From 8b2f38307b4db60d647dafb36bb824bc46137f7b Mon Sep 17 00:00:00 2001 From: bicarus <202771338+bicarus-dev@users.noreply.github.com> Date: Tue, 18 Aug 2026 00:22:45 -0700 Subject: [PATCH] graphics: rewrite screenshot and api capture image processing (#870) ## Link to GitHub Issue or related Pull Request, if one exists #0 ## Description of change Significantly speeds up API screen capture and D3D9 screenshots saving. Two reasons for doing this: 1. We now have a 4K game (GITADORA) and existing capture code was taking multiple seconds. 2. Renewed user interest on streaming as we have a couple more companion apps in active development. **API screen capture (streaming), 1280x720:** 14.3ms -> 6.3ms per frame. Back buffer copies go to pooled `D3DPOOL_SYSTEMMEM` surfaces via `GetRenderTargetData` instead of allocating a lockable render target every frame, and TooJpeg is replaced with libjpeg-turbo (encode 9.8ms -> 3.0ms). MSAA remains unsupported **Screenshots for GITADORA arena model, across 4 screens with one of them 4K**: 4068ms -> 124ms. `D3DXSaveSurfaceToFileA` is replaced with fpng (encode 4043ms -> 76ms) and the screens encode in parallel. Dropping D3DX also removes the `d3dx9_43.dll` ... `d3dx9_24.dll` probing loop, so screenshots no longer fail outright on machines with no D3DX9 runtime installed. Screenshot surfaces are read on the present thread, so no D3D call reaches another thread for screenshots. This fixes a hang in DDR X2 introduced earlier in the branch: its device has no internal locking, and reading the surface on a pool thread while the present thread sat inside `GetRenderTargetData` left the game's own render thread deadlocked. ## Testing - **GITADORA** (arena model, D3D9Ex, 4K main plus three subscreens, windowed) with `-screenshotsub`: three sets of four screenshots, images verified correct. Completion order differs between sets, so the screens really are encoding in parallel. - **LovePlus** (KLP, plain D3D9, 768x1360): covers the inline path used by games whose image processing must not leave the present thread. - **API screen capture** through a companion app: live video correct throughout. - **Print Screen** bound as the screenshot key: the clipboard copy succeeded on every shot. - Quitting the game after capturing leaves no `IDirect3DDevice9` reference count warning, so the pooled readback surfaces are released along with the device. --- src/spice2x/CMakeLists.txt | 38 +- src/spice2x/api/modules/capture.cpp | 7 +- src/spice2x/external/docker/Dockerfile | 1 + src/spice2x/external/fpng/fpng.cpp | 3222 +++++++++++++ src/spice2x/external/fpng/fpng.h | 122 + .../external/libjpeg-turbo/CMakeLists.txt | 143 + src/spice2x/external/libjpeg-turbo/LICENSE.md | 135 + src/spice2x/external/libjpeg-turbo/README.ijg | 258 + src/spice2x/external/libjpeg-turbo/jaricom.c | 157 + src/spice2x/external/libjpeg-turbo/jcapimin.c | 318 ++ src/spice2x/external/libjpeg-turbo/jcapistd.c | 187 + src/spice2x/external/libjpeg-turbo/jcarith.c | 932 ++++ src/spice2x/external/libjpeg-turbo/jccoefct.c | 454 ++ src/spice2x/external/libjpeg-turbo/jccolext.c | 152 + src/spice2x/external/libjpeg-turbo/jccolor.c | 732 +++ src/spice2x/external/libjpeg-turbo/jcdctmgr.c | 748 +++ src/spice2x/external/libjpeg-turbo/jcdiffct.c | 411 ++ src/spice2x/external/libjpeg-turbo/jchuff.c | 1175 +++++ src/spice2x/external/libjpeg-turbo/jchuff.h | 44 + src/spice2x/external/libjpeg-turbo/jcicc.c | 105 + src/spice2x/external/libjpeg-turbo/jcinit.c | 149 + src/spice2x/external/libjpeg-turbo/jclhuff.c | 587 +++ src/spice2x/external/libjpeg-turbo/jclossls.c | 319 ++ src/spice2x/external/libjpeg-turbo/jcmainct.c | 173 + src/spice2x/external/libjpeg-turbo/jcmarker.c | 670 +++ src/spice2x/external/libjpeg-turbo/jcmaster.c | 793 ++++ src/spice2x/external/libjpeg-turbo/jcmaster.h | 43 + src/spice2x/external/libjpeg-turbo/jcomapi.c | 110 + .../external/libjpeg-turbo/jconfig.h.in | 60 + .../external/libjpeg-turbo/jconfigint.h.in | 76 + src/spice2x/external/libjpeg-turbo/jcparam.c | 585 +++ src/spice2x/external/libjpeg-turbo/jcphuff.c | 1102 +++++ src/spice2x/external/libjpeg-turbo/jcprepct.c | 365 ++ src/spice2x/external/libjpeg-turbo/jcsample.c | 543 +++ src/spice2x/external/libjpeg-turbo/jctrans.c | 415 ++ src/spice2x/external/libjpeg-turbo/jdapimin.c | 417 ++ src/spice2x/external/libjpeg-turbo/jdapistd.c | 752 +++ src/spice2x/external/libjpeg-turbo/jdarith.c | 782 ++++ src/spice2x/external/libjpeg-turbo/jdatadst.c | 277 ++ src/spice2x/external/libjpeg-turbo/jdatasrc.c | 289 ++ src/spice2x/external/libjpeg-turbo/jdcoefct.c | 885 ++++ src/spice2x/external/libjpeg-turbo/jdcoefct.h | 88 + src/spice2x/external/libjpeg-turbo/jdcol565.c | 392 ++ src/spice2x/external/libjpeg-turbo/jdcolext.c | 145 + src/spice2x/external/libjpeg-turbo/jdcolor.c | 925 ++++ src/spice2x/external/libjpeg-turbo/jdct.h | 221 + src/spice2x/external/libjpeg-turbo/jddctmgr.c | 365 ++ src/spice2x/external/libjpeg-turbo/jddiffct.c | 403 ++ src/spice2x/external/libjpeg-turbo/jdhuff.c | 836 ++++ src/spice2x/external/libjpeg-turbo/jdhuff.h | 250 + src/spice2x/external/libjpeg-turbo/jdicc.c | 167 + src/spice2x/external/libjpeg-turbo/jdinput.c | 419 ++ src/spice2x/external/libjpeg-turbo/jdlhuff.c | 302 ++ src/spice2x/external/libjpeg-turbo/jdlossls.c | 289 ++ src/spice2x/external/libjpeg-turbo/jdmainct.c | 469 ++ src/spice2x/external/libjpeg-turbo/jdmainct.h | 78 + src/spice2x/external/libjpeg-turbo/jdmarker.c | 1384 ++++++ src/spice2x/external/libjpeg-turbo/jdmaster.c | 891 ++++ src/spice2x/external/libjpeg-turbo/jdmaster.h | 28 + src/spice2x/external/libjpeg-turbo/jdmerge.c | 594 +++ src/spice2x/external/libjpeg-turbo/jdmerge.h | 48 + src/spice2x/external/libjpeg-turbo/jdmrg565.c | 355 ++ src/spice2x/external/libjpeg-turbo/jdmrgext.c | 184 + src/spice2x/external/libjpeg-turbo/jdphuff.c | 681 +++ src/spice2x/external/libjpeg-turbo/jdpostct.c | 315 ++ src/spice2x/external/libjpeg-turbo/jdsample.c | 540 +++ src/spice2x/external/libjpeg-turbo/jdsample.h | 53 + src/spice2x/external/libjpeg-turbo/jdtrans.c | 162 + src/spice2x/external/libjpeg-turbo/jerror.c | 243 + src/spice2x/external/libjpeg-turbo/jerror.h | 336 ++ src/spice2x/external/libjpeg-turbo/jfdctflt.c | 169 + src/spice2x/external/libjpeg-turbo/jfdctfst.c | 227 + src/spice2x/external/libjpeg-turbo/jfdctint.c | 288 ++ src/spice2x/external/libjpeg-turbo/jidctflt.c | 240 + src/spice2x/external/libjpeg-turbo/jidctfst.c | 371 ++ src/spice2x/external/libjpeg-turbo/jidctint.c | 2627 +++++++++++ src/spice2x/external/libjpeg-turbo/jidctred.c | 409 ++ src/spice2x/external/libjpeg-turbo/jinclude.h | 147 + src/spice2x/external/libjpeg-turbo/jlossls.h | 101 + src/spice2x/external/libjpeg-turbo/jmemmgr.c | 1289 +++++ src/spice2x/external/libjpeg-turbo/jmemnobs.c | 110 + src/spice2x/external/libjpeg-turbo/jmemsys.h | 147 + src/spice2x/external/libjpeg-turbo/jmorecfg.h | 385 ++ .../external/libjpeg-turbo/jpeg_nbits.c | 4134 +++++++++++++++++ .../external/libjpeg-turbo/jpeg_nbits.h | 43 + .../external/libjpeg-turbo/jpegapicomp.h | 32 + src/spice2x/external/libjpeg-turbo/jpegint.h | 600 +++ src/spice2x/external/libjpeg-turbo/jpeglib.h | 1218 +++++ src/spice2x/external/libjpeg-turbo/jpegtran.c | 729 +++ src/spice2x/external/libjpeg-turbo/jquant1.c | 864 ++++ src/spice2x/external/libjpeg-turbo/jquant2.c | 1293 ++++++ .../external/libjpeg-turbo/jsamplecomp.h | 336 ++ src/spice2x/external/libjpeg-turbo/jsimd.h | 127 + src/spice2x/external/libjpeg-turbo/jsimddct.h | 70 + src/spice2x/external/libjpeg-turbo/jstdhuff.c | 144 + src/spice2x/external/libjpeg-turbo/jutils.c | 148 + .../external/libjpeg-turbo/jversion.h.in | 56 + .../libjpeg-turbo/simd/i386/jccolext-avx2.asm | 574 +++ .../libjpeg-turbo/simd/i386/jccolext-mmx.asm | 472 ++ .../libjpeg-turbo/simd/i386/jccolext-sse2.asm | 499 ++ .../libjpeg-turbo/simd/i386/jccolor-avx2.asm | 117 + .../libjpeg-turbo/simd/i386/jccolor-mmx.asm | 117 + .../libjpeg-turbo/simd/i386/jccolor-sse2.asm | 116 + .../libjpeg-turbo/simd/i386/jcgray-avx2.asm | 109 + .../libjpeg-turbo/simd/i386/jcgray-mmx.asm | 109 + .../libjpeg-turbo/simd/i386/jcgray-sse2.asm | 108 + .../libjpeg-turbo/simd/i386/jcgryext-avx2.asm | 453 ++ .../libjpeg-turbo/simd/i386/jcgryext-mmx.asm | 351 ++ .../libjpeg-turbo/simd/i386/jcgryext-sse2.asm | 378 ++ .../libjpeg-turbo/simd/i386/jchuff-sse2.asm | 758 +++ .../libjpeg-turbo/simd/i386/jcphuff-sse2.asm | 658 +++ .../libjpeg-turbo/simd/i386/jcsample-avx2.asm | 384 ++ .../libjpeg-turbo/simd/i386/jcsample-mmx.asm | 320 ++ .../libjpeg-turbo/simd/i386/jcsample-sse2.asm | 347 ++ .../libjpeg-turbo/simd/i386/jdcolext-avx2.asm | 511 ++ .../libjpeg-turbo/simd/i386/jdcolext-mmx.asm | 400 ++ .../libjpeg-turbo/simd/i386/jdcolext-sse2.asm | 454 ++ .../libjpeg-turbo/simd/i386/jdcolor-avx2.asm | 114 + .../libjpeg-turbo/simd/i386/jdcolor-mmx.asm | 113 + .../libjpeg-turbo/simd/i386/jdcolor-sse2.asm | 113 + .../libjpeg-turbo/simd/i386/jdmerge-avx2.asm | 132 + .../libjpeg-turbo/simd/i386/jdmerge-mmx.asm | 119 + .../libjpeg-turbo/simd/i386/jdmerge-sse2.asm | 131 + .../libjpeg-turbo/simd/i386/jdmrgext-avx2.asm | 571 +++ .../libjpeg-turbo/simd/i386/jdmrgext-mmx.asm | 456 ++ .../libjpeg-turbo/simd/i386/jdmrgext-sse2.asm | 513 ++ .../libjpeg-turbo/simd/i386/jdsample-avx2.asm | 756 +++ .../libjpeg-turbo/simd/i386/jdsample-mmx.asm | 727 +++ .../libjpeg-turbo/simd/i386/jdsample-sse2.asm | 720 +++ .../libjpeg-turbo/simd/i386/jfdctflt-3dn.asm | 314 ++ .../libjpeg-turbo/simd/i386/jfdctflt-sse.asm | 365 ++ .../libjpeg-turbo/simd/i386/jfdctfst-mmx.asm | 391 ++ .../libjpeg-turbo/simd/i386/jfdctfst-sse2.asm | 399 ++ .../libjpeg-turbo/simd/i386/jfdctint-avx2.asm | 327 ++ .../libjpeg-turbo/simd/i386/jfdctint-mmx.asm | 616 +++ .../libjpeg-turbo/simd/i386/jfdctint-sse2.asm | 629 +++ .../libjpeg-turbo/simd/i386/jidctflt-3dn.asm | 447 ++ .../libjpeg-turbo/simd/i386/jidctflt-sse.asm | 567 +++ .../libjpeg-turbo/simd/i386/jidctflt-sse2.asm | 493 ++ .../libjpeg-turbo/simd/i386/jidctfst-mmx.asm | 495 ++ .../libjpeg-turbo/simd/i386/jidctfst-sse2.asm | 497 ++ .../libjpeg-turbo/simd/i386/jidctint-avx2.asm | 449 ++ .../libjpeg-turbo/simd/i386/jidctint-mmx.asm | 847 ++++ .../libjpeg-turbo/simd/i386/jidctint-sse2.asm | 854 ++++ .../libjpeg-turbo/simd/i386/jidctred-mmx.asm | 700 +++ .../libjpeg-turbo/simd/i386/jidctred-sse2.asm | 588 +++ .../libjpeg-turbo/simd/i386/jquant-3dn.asm | 226 + .../libjpeg-turbo/simd/i386/jquant-mmx.asm | 272 ++ .../libjpeg-turbo/simd/i386/jquant-sse.asm | 204 + .../libjpeg-turbo/simd/i386/jquantf-sse2.asm | 164 + .../libjpeg-turbo/simd/i386/jquanti-avx2.asm | 184 + .../libjpeg-turbo/simd/i386/jquanti-sse2.asm | 197 + .../external/libjpeg-turbo/simd/i386/jsimd.c | 1312 ++++++ .../libjpeg-turbo/simd/i386/jsimdcpu.asm | 131 + .../external/libjpeg-turbo/simd/jsimd.h | 1258 +++++ .../libjpeg-turbo/simd/nasm/jcolsamp.inc | 135 + .../external/libjpeg-turbo/simd/nasm/jdct.inc | 31 + .../libjpeg-turbo/simd/nasm/jsimdcfg.inc | 93 + .../libjpeg-turbo/simd/nasm/jsimdcfg.inc.h | 133 + .../libjpeg-turbo/simd/nasm/jsimdext.inc | 542 +++ .../simd/x86_64/jccolext-avx2.asm | 557 +++ .../simd/x86_64/jccolext-sse2.asm | 482 ++ .../simd/x86_64/jccolor-avx2.asm | 117 + .../simd/x86_64/jccolor-sse2.asm | 116 + .../libjpeg-turbo/simd/x86_64/jcgray-avx2.asm | 109 + .../libjpeg-turbo/simd/x86_64/jcgray-sse2.asm | 108 + .../simd/x86_64/jcgryext-avx2.asm | 436 ++ .../simd/x86_64/jcgryext-sse2.asm | 361 ++ .../libjpeg-turbo/simd/x86_64/jchuff-sse2.asm | 581 +++ .../simd/x86_64/jcphuff-sse2.asm | 629 +++ .../simd/x86_64/jcsample-avx2.asm | 363 ++ .../simd/x86_64/jcsample-sse2.asm | 326 ++ .../simd/x86_64/jdcolext-avx2.asm | 494 ++ .../simd/x86_64/jdcolext-sse2.asm | 437 ++ .../simd/x86_64/jdcolor-avx2.asm | 114 + .../simd/x86_64/jdcolor-sse2.asm | 113 + .../simd/x86_64/jdmerge-avx2.asm | 132 + .../simd/x86_64/jdmerge-sse2.asm | 131 + .../simd/x86_64/jdmrgext-avx2.asm | 594 +++ .../simd/x86_64/jdmrgext-sse2.asm | 536 +++ .../simd/x86_64/jdsample-avx2.asm | 694 +++ .../simd/x86_64/jdsample-sse2.asm | 663 +++ .../simd/x86_64/jfdctflt-sse.asm | 353 ++ .../simd/x86_64/jfdctfst-sse2.asm | 387 ++ .../simd/x86_64/jfdctint-avx2.asm | 316 ++ .../simd/x86_64/jfdctint-sse2.asm | 617 +++ .../simd/x86_64/jidctflt-sse2.asm | 478 ++ .../simd/x86_64/jidctfst-sse2.asm | 487 ++ .../simd/x86_64/jidctint-avx2.asm | 414 ++ .../simd/x86_64/jidctint-sse2.asm | 843 ++++ .../simd/x86_64/jidctred-sse2.asm | 570 +++ .../simd/x86_64/jquantf-sse2.asm | 151 + .../simd/x86_64/jquanti-avx2.asm | 159 + .../simd/x86_64/jquanti-sse2.asm | 184 + .../libjpeg-turbo/simd/x86_64/jsimd.c | 1110 +++++ .../libjpeg-turbo/simd/x86_64/jsimdcpu.asm | 86 + src/spice2x/external/toojpeg/LICENSE | 10 - src/spice2x/external/toojpeg/toojpeg.cpp | 665 --- src/spice2x/external/toojpeg/toojpeg.h | 62 - .../graphics/backends/d3d9/d3d9_device.cpp | 3 + .../graphics/backends/d3d9/d3d9_readback.cpp | 241 + .../graphics/backends/d3d9/d3d9_readback.h | 44 + .../backends/d3d9/d3d9_screenshot.cpp | 781 ++-- src/spice2x/hooks/graphics/graphics.cpp | 19 +- src/spice2x/hooks/graphics/graphics.h | 10 +- src/spice2x/hooks/graphics/jpeg_encoder.cpp | 140 + src/spice2x/hooks/graphics/jpeg_encoder.h | 15 + src/spice2x/licenses.txt | 85 +- src/spice2x/misc/clipboard.cpp | 9 +- 209 files changed, 87027 insertions(+), 1065 deletions(-) create mode 100644 src/spice2x/external/fpng/fpng.cpp create mode 100644 src/spice2x/external/fpng/fpng.h create mode 100644 src/spice2x/external/libjpeg-turbo/CMakeLists.txt create mode 100644 src/spice2x/external/libjpeg-turbo/LICENSE.md create mode 100644 src/spice2x/external/libjpeg-turbo/README.ijg create mode 100644 src/spice2x/external/libjpeg-turbo/jaricom.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcapimin.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcapistd.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcarith.c create mode 100644 src/spice2x/external/libjpeg-turbo/jccoefct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jccolext.c create mode 100644 src/spice2x/external/libjpeg-turbo/jccolor.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcdctmgr.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcdiffct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jchuff.c create mode 100644 src/spice2x/external/libjpeg-turbo/jchuff.h create mode 100644 src/spice2x/external/libjpeg-turbo/jcicc.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcinit.c create mode 100644 src/spice2x/external/libjpeg-turbo/jclhuff.c create mode 100644 src/spice2x/external/libjpeg-turbo/jclossls.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcmainct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcmarker.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcmaster.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcmaster.h create mode 100644 src/spice2x/external/libjpeg-turbo/jcomapi.c create mode 100644 src/spice2x/external/libjpeg-turbo/jconfig.h.in create mode 100644 src/spice2x/external/libjpeg-turbo/jconfigint.h.in create mode 100644 src/spice2x/external/libjpeg-turbo/jcparam.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcphuff.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcprepct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jcsample.c create mode 100644 src/spice2x/external/libjpeg-turbo/jctrans.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdapimin.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdapistd.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdarith.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdatadst.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdatasrc.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdcoefct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdcoefct.h create mode 100644 src/spice2x/external/libjpeg-turbo/jdcol565.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdcolext.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdcolor.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdct.h create mode 100644 src/spice2x/external/libjpeg-turbo/jddctmgr.c create mode 100644 src/spice2x/external/libjpeg-turbo/jddiffct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdhuff.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdhuff.h create mode 100644 src/spice2x/external/libjpeg-turbo/jdicc.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdinput.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdlhuff.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdlossls.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdmainct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdmainct.h create mode 100644 src/spice2x/external/libjpeg-turbo/jdmarker.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdmaster.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdmaster.h create mode 100644 src/spice2x/external/libjpeg-turbo/jdmerge.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdmerge.h create mode 100644 src/spice2x/external/libjpeg-turbo/jdmrg565.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdmrgext.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdphuff.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdpostct.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdsample.c create mode 100644 src/spice2x/external/libjpeg-turbo/jdsample.h create mode 100644 src/spice2x/external/libjpeg-turbo/jdtrans.c create mode 100644 src/spice2x/external/libjpeg-turbo/jerror.c create mode 100644 src/spice2x/external/libjpeg-turbo/jerror.h create mode 100644 src/spice2x/external/libjpeg-turbo/jfdctflt.c create mode 100644 src/spice2x/external/libjpeg-turbo/jfdctfst.c create mode 100644 src/spice2x/external/libjpeg-turbo/jfdctint.c create mode 100644 src/spice2x/external/libjpeg-turbo/jidctflt.c create mode 100644 src/spice2x/external/libjpeg-turbo/jidctfst.c create mode 100644 src/spice2x/external/libjpeg-turbo/jidctint.c create mode 100644 src/spice2x/external/libjpeg-turbo/jidctred.c create mode 100644 src/spice2x/external/libjpeg-turbo/jinclude.h create mode 100644 src/spice2x/external/libjpeg-turbo/jlossls.h create mode 100644 src/spice2x/external/libjpeg-turbo/jmemmgr.c create mode 100644 src/spice2x/external/libjpeg-turbo/jmemnobs.c create mode 100644 src/spice2x/external/libjpeg-turbo/jmemsys.h create mode 100644 src/spice2x/external/libjpeg-turbo/jmorecfg.h create mode 100644 src/spice2x/external/libjpeg-turbo/jpeg_nbits.c create mode 100644 src/spice2x/external/libjpeg-turbo/jpeg_nbits.h create mode 100644 src/spice2x/external/libjpeg-turbo/jpegapicomp.h create mode 100644 src/spice2x/external/libjpeg-turbo/jpegint.h create mode 100644 src/spice2x/external/libjpeg-turbo/jpeglib.h create mode 100644 src/spice2x/external/libjpeg-turbo/jpegtran.c create mode 100644 src/spice2x/external/libjpeg-turbo/jquant1.c create mode 100644 src/spice2x/external/libjpeg-turbo/jquant2.c create mode 100644 src/spice2x/external/libjpeg-turbo/jsamplecomp.h create mode 100644 src/spice2x/external/libjpeg-turbo/jsimd.h create mode 100644 src/spice2x/external/libjpeg-turbo/jsimddct.h create mode 100644 src/spice2x/external/libjpeg-turbo/jstdhuff.c create mode 100644 src/spice2x/external/libjpeg-turbo/jutils.c create mode 100644 src/spice2x/external/libjpeg-turbo/jversion.h.in create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jchuff-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcphuff-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-3dn.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-sse.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-3dn.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jquant-3dn.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jquant-mmx.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jquant-sse.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jquantf-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jsimd.c create mode 100644 src/spice2x/external/libjpeg-turbo/simd/i386/jsimdcpu.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/jsimd.h create mode 100644 src/spice2x/external/libjpeg-turbo/simd/nasm/jcolsamp.inc create mode 100644 src/spice2x/external/libjpeg-turbo/simd/nasm/jdct.inc create mode 100644 src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc create mode 100644 src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc.h create mode 100644 src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdext.inc create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jchuff-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jcphuff-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctflt-sse.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctfst-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctflt-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctfst-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctred-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jquantf-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-avx2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-sse2.asm create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimd.c create mode 100644 src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimdcpu.asm delete mode 100644 src/spice2x/external/toojpeg/LICENSE delete mode 100644 src/spice2x/external/toojpeg/toojpeg.cpp delete mode 100644 src/spice2x/external/toojpeg/toojpeg.h create mode 100644 src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.cpp create mode 100644 src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.h create mode 100644 src/spice2x/hooks/graphics/jpeg_encoder.cpp create mode 100644 src/spice2x/hooks/graphics/jpeg_encoder.h diff --git a/src/spice2x/CMakeLists.txt b/src/spice2x/CMakeLists.txt index addc95e..00e74e8 100644 --- a/src/spice2x/CMakeLists.txt +++ b/src/spice2x/CMakeLists.txt @@ -256,6 +256,28 @@ add_subdirectory(external/imgui EXCLUDE_FROM_ALL) add_subdirectory(external/minhook EXCLUDE_FROM_ALL) add_subdirectory(external/cpu_features EXCLUDE_FROM_ALL) +# libjpeg-turbo's SIMD needs nasm; enable_language belongs at top level. Without +# it the library still builds, just without SIMD. +include(CheckLanguage) +check_language(ASM_NASM) +if(CMAKE_ASM_NASM_COMPILER) + enable_language(ASM_NASM) +else() + message(WARNING + "nasm not found: libjpeg-turbo will be built without SIMD, " + "which makes api screen capture several times slower") +endif() +add_subdirectory(external/libjpeg-turbo EXCLUDE_FROM_ALL) + +# fpng's SIMD needs the whole unit built for SSE4.1, which its runtime CPU check +# cannot undo, so keep it scalar rather than raising the CPU baseline +set_source_files_properties(external/fpng/fpng.cpp PROPERTIES + COMPILE_DEFINITIONS "FPNG_NO_SSE=1") +if(NOT MSVC) + set_source_files_properties(external/fpng/fpng.cpp PROPERTIES + COMPILE_OPTIONS "-fno-strict-aliasing") +endif() + # set link time optimizations (disabled for Debug builds for speed, disabled # for RelWithDebInfo builds due to "lto1: error: two or more sections for" # errors) @@ -380,7 +402,7 @@ set(SOURCE_FILES ${SOURCE_FILES} external/tinyxml2/tinyxml2.cpp external/http-parser/http_parser.c external/usbhidusage/usb-hid-usage.c - external/toojpeg/toojpeg.cpp + external/fpng/fpng.cpp external/scard/scard.cpp # games @@ -532,10 +554,12 @@ set(SOURCE_FILES ${SOURCE_FILES} hooks/devicehook.cpp hooks/graphics/graphics.cpp hooks/graphics/graphics_windowed.cpp + hooks/graphics/jpeg_encoder.cpp hooks/graphics/nvapi_impl.cpp hooks/graphics/nvapi_hook.cpp hooks/graphics/nvenc_hook.cpp hooks/graphics/backends/d3d9/d3d9_backend.cpp + hooks/graphics/backends/d3d9/d3d9_readback.cpp hooks/graphics/backends/d3d9/d3d9_screenshot.cpp hooks/graphics/backends/d3d9/d3d9_device.cpp hooks/graphics/backends/d3d9/d3d9_gfdm.cpp @@ -740,7 +764,7 @@ endfunction() add_library(spicetools_spice_objs OBJECT ${SOURCE_FILES}) target_link_libraries(spicetools_spice_objs PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp - PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features) + PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static) target_link_libraries(spicetools_spice_objs PUBLIC winscard) if(NOT MSVC) @@ -780,7 +804,7 @@ set(RESOURCE_FILES build/manifest.manifest build/manifest.rc build/icon.rc cfg/W add_executable(spicetools_spice_linux ${SOURCE_FILES} ${RESOURCE_FILES}) target_link_libraries(spicetools_spice_linux PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp - PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features) + PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static) set_target_properties(spicetools_spice_linux PROPERTIES PREFIX "") set_target_properties(spicetools_spice_linux PROPERTIES OUTPUT_NAME "spice_linux") target_compile_definitions(spicetools_spice_linux PRIVATE NO_SCARD=1 PRIVATE SPICE_LINUX=1) @@ -798,7 +822,7 @@ add_executable(spicetools_spice64 ${SOURCE_FILES} ${RESOURCE_FILES}) # do NOT link against: mf, mfplat, mfreadwrite; otherwise unity games will break target_link_libraries(spicetools_spice64 PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp mfuuid strmiids dxva2 - PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features) + PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static) target_link_libraries(spicetools_spice64 PUBLIC winscard) set_target_properties(spicetools_spice64 PROPERTIES PREFIX "") set_target_properties(spicetools_spice64 PROPERTIES OUTPUT_NAME "spice64") @@ -821,7 +845,7 @@ add_executable(spicetools_spice64_linux ${SOURCE_FILES} ${RESOURCE_FILES}) # do NOT link against: mf, mfplat, mfreadwrite; otherwise unity games will break target_link_libraries(spicetools_spice64_linux PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp mfuuid strmiids dxva2 - PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features) + PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static) set_target_properties(spicetools_spice64_linux PROPERTIES PREFIX "") set_target_properties(spicetools_spice64_linux PROPERTIES OUTPUT_NAME "spice64_linux") target_compile_definitions(spicetools_spice64_linux PRIVATE SPICE64=1) @@ -842,7 +866,7 @@ set(RESOURCE_FILES cfg/manifest.manifest cfg/manifest.rc cfg/icon.rc cfg/Win32D. add_executable(spicetools_cfg WIN32 ${SOURCE_FILES} ${RESOURCE_FILES}) target_link_libraries(spicetools_cfg PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp strmiids - PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features) + PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static) target_link_libraries(spicetools_cfg PUBLIC winscard) set_target_properties(spicetools_cfg PROPERTIES PREFIX "") set_target_properties(spicetools_cfg PROPERTIES OUTPUT_NAME "spicecfg") @@ -860,7 +884,7 @@ set(RESOURCE_FILES cfg/manifest.manifest cfg/manifest.rc cfg/icon.rc cfg/Win32D. add_executable(spicetools_cfg_linux WIN32 ${SOURCE_FILES} ${RESOURCE_FILES}) target_link_libraries(spicetools_cfg_linux PUBLIC d3d9 ws2_32 version comctl32 shlwapi iphlpapi hid secur32 setupapi psapi winmm winhttp strmiids - PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features) + PRIVATE fmt::fmt-header-only discord-rpc imgui hash-library minhook imm32 dwmapi CpuFeatures::cpu_features jpeg-static) set_target_properties(spicetools_cfg_linux PROPERTIES PREFIX "") set_target_properties(spicetools_cfg_linux PROPERTIES OUTPUT_NAME "spicecfg_linux") target_compile_definitions(spicetools_cfg_linux PRIVATE SPICETOOLS_SPICECFG_STANDALONE=1) diff --git a/src/spice2x/api/modules/capture.cpp b/src/spice2x/api/modules/capture.cpp index 7d393f1..ed96937 100644 --- a/src/spice2x/api/modules/capture.cpp +++ b/src/spice2x/api/modules/capture.cpp @@ -93,8 +93,6 @@ namespace api::modules { * reduce: uint for dividing image size */ void Capture::get_jpg(Request &req, Response &res) { - CAPTURE_BUFFER.clear(); - CAPTURE_BUFFER.reserve(1024 * 128); // settings int screen = 0; @@ -121,9 +119,8 @@ namespace api::modules { int width = 0; int height = 0; graphics_capture_trigger(screen); - bool success = graphics_capture_receive_jpeg(screen, [] (uint8_t byte) { - CAPTURE_BUFFER.push_back(byte); - }, true, quality, true, divide, ×tamp, &width, &height); + bool success = graphics_capture_receive_jpeg( + screen, CAPTURE_BUFFER, quality, divide, ×tamp, &width, &height); if (success) { add_jpeg_response(screen, timestamp, width, height, CAPTURE_BUFFER, res); diff --git a/src/spice2x/external/docker/Dockerfile b/src/spice2x/external/docker/Dockerfile index 29a6476..4413c20 100644 --- a/src/spice2x/external/docker/Dockerfile +++ b/src/spice2x/external/docker/Dockerfile @@ -7,6 +7,7 @@ RUN pacman --noconfirm -Syu git \ ninja \ cmake \ unzip \ + nasm \ mingw-w64-crt \ mingw-w64-winpthreads \ mingw-w64-gcc \ diff --git a/src/spice2x/external/fpng/fpng.cpp b/src/spice2x/external/fpng/fpng.cpp new file mode 100644 index 0000000..68be28d --- /dev/null +++ b/src/spice2x/external/fpng/fpng.cpp @@ -0,0 +1,3222 @@ +// fpng.cpp 1.0.6 - Fast 24/32bpp .PNG image writer/reader. See unlicense at the end of this file. +// PNG's generated by this code have been tested to load successfully with stb_image.h, lodepng.cpp, wuffs, libpng, and pngcheck. +// +// Uses code from the simple PNG writer function by Alex Evans, 2011. Released into the public domain: https://gist.github.com/908299 +// Some low-level Deflate/Huffman functions derived from the original 2011 Google Code version of miniz (public domain by R. Geldreich, Jr.): https://code.google.com/archive/p/miniz/ +// Low-level Huffman code size function: public domain, originally written by: Alistair Moffat, alistair@cs.mu.oz.au, Jyrki Katajainen, jyrki@diku.dk, November 1996. +// +// Optional config macros: +// FPNG_NO_SSE - Set to 1 to completely disable SSE usage, even on x86/x64. By default, on x86/x64 it's enabled. +// FPNG_DISABLE_DECODE_CRC32_CHECKS - Set to 1 to disable PNG chunk CRC-32 tests, for improved fuzzing. Defaults to 0. +// FPNG_USE_UNALIGNED_LOADS - Set to 1 to indicate it's OK to read/write unaligned 32-bit/64-bit values. Defaults to 0, unless x86/x64. +// +// With gcc/clang on x86, compile with -msse4.1 -mpclmul -fno-strict-aliasing +// Only tested with -fno-strict-aliasing (which the Linux kernel uses, and MSVC's default). +// +#include "fpng.h" +#include +#include + +#ifdef _MSC_VER + #pragma warning (disable:4127) // conditional expression is constant +#endif + +// Set FPNG_NO_SSE to 1 to completely disable SSE usage. +#ifndef FPNG_NO_SSE + #define FPNG_NO_SSE (0) +#endif + +// Detect if we're compiling on x86/x64 +#if defined(_M_IX86) || defined(_M_X64) || defined(__i386__) || defined(__i386) || defined(__i486__) || defined(__i486) || defined(i386) || defined(__ia64__) || defined(__x86_64__) + #define FPNG_X86_OR_X64_CPU (1) +#else + #define FPNG_X86_OR_X64_CPU (0) +#endif + +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + #ifdef _MSC_VER + #include + #endif + #include // SSE + #include // SSE2 + #include // SSE4.1 + #include // pclmul +#endif + +#ifndef FPNG_NO_STDIO + #include +#endif + +// Allow the disabling of the chunk data CRC32 checks, for fuzz testing of the decoder +#ifndef FPNG_DISABLE_DECODE_CRC32_CHECKS + #define FPNG_DISABLE_DECODE_CRC32_CHECKS (0) +#endif + +// Using unaligned loads and stores causes errors when using UBSan. Jam it off. +#if defined(__has_feature) + #if __has_feature(undefined_behavior_sanitizer) + #undef FPNG_USE_UNALIGNED_LOADS + #define FPNG_USE_UNALIGNED_LOADS (0) + #endif +#endif + +// Set to 0 if your platform doesn't support unaligned 32-bit/64-bit reads/writes. +#ifndef FPNG_USE_UNALIGNED_LOADS + #if FPNG_X86_OR_X64_CPU + // On x86/x64 we default to enabled, for a noticeable perf gain. + #define FPNG_USE_UNALIGNED_LOADS (1) + #else + #define FPNG_USE_UNALIGNED_LOADS (0) + #endif +#endif + +#if defined(_MSC_VER) || defined(__MINGW32__) || FPNG_X86_OR_X64_CPU + #ifndef __LITTLE_ENDIAN + #define __LITTLE_ENDIAN 1234 + #endif + #ifndef __BIG_ENDIAN + #define __BIG_ENDIAN 4321 + #endif + + // Assume little endian on Windows/x86/x64. + #define __BYTE_ORDER __LITTLE_ENDIAN +#elif defined(__APPLE__) + #define __BYTE_ORDER __BYTE_ORDER__ + #define __LITTLE_ENDIAN __LITTLE_ENDIAN__ + #define __BIG_ENDIAN __BIG_ENDIAN__ +#else + // for __BYTE_ORDER (__LITTLE_ENDIAN or __BIG_ENDIAN) + #include + + #ifndef __LITTLE_ENDIAN + #define __LITTLE_ENDIAN 1234 + #endif + #ifndef __BIG_ENDIAN + #define __BIG_ENDIAN 4321 + #endif +#endif + +#if !defined(__BYTE_ORDER) + #error __BYTE_ORDER undefined. Compile with -D__BYTE_ORDER=1234 for little endian or -D__BYTE_ORDER=4321 for big endian. +#endif + +namespace fpng +{ + static const int FPNG_FALSE = 0; + static const uint8_t FPNG_FDEC_VERSION = 0; + static const uint32_t FPNG_MAX_SUPPORTED_DIM = 1 << 24; + + template static inline S maximum(S a, S b) { return (a > b) ? a : b; } + template static inline S minimum(S a, S b) { return (a < b) ? a : b; } + + static inline uint32_t simple_swap32(uint32_t x) { return (x >> 24) | ((x >> 8) & 0x0000FF00) | ((x << 8) & 0x00FF0000) | (x << 24); } + static inline uint64_t simple_swap64(uint64_t x) { return (((uint64_t)simple_swap32((uint32_t)x)) << 32U) | simple_swap32((uint32_t)(x >> 32U)); } + + static inline uint32_t swap32(uint32_t x) + { +#if defined(__GNUC__) || defined(__clang__) + return __builtin_bswap32(x); +#else + return simple_swap32(x); +#endif + } + + static inline uint64_t swap64(uint64_t x) + { +#if defined(__GNUC__) || defined(__clang__) + return __builtin_bswap64(x); +#else + return simple_swap64(x); +#endif + } + +#if FPNG_USE_UNALIGNED_LOADS + #if __BYTE_ORDER == __BIG_ENDIAN + #define READ_LE32(p) swap32(*reinterpret_cast(p)) + #define WRITE_LE32(p, v) *reinterpret_cast(p) = swap32((uint32_t)(v)) + #define WRITE_LE64(p, v) *reinterpret_cast(p) = swap64((uint64_t)(v)) + + #define READ_BE32(p) *reinterpret_cast(p) + #else + #define READ_LE32(p) (*reinterpret_cast(p)) + #define WRITE_LE32(p, v) *reinterpret_cast(p) = (uint32_t)(v) + #define WRITE_LE64(p, v) *reinterpret_cast(p) = (uint64_t)(v) + + #define READ_BE32(p) swap32(*reinterpret_cast(p)) + #endif +#else + // A good compiler should be able to optimize these routines - hopefully. They are crucial for performance. + static inline uint32_t READ_LE32(const void* p) + { + const uint8_t* pBytes = (const uint8_t*)p; + return ((uint32_t)pBytes[0]) | (((uint32_t)pBytes[1]) << 8U) | (((uint32_t)pBytes[2]) << 16U) | (((uint32_t)pBytes[3]) << 24U); + } + + static inline uint32_t READ_BE32(const void* p) + { + const uint8_t* pBytes = (const uint8_t*)p; + return ((uint32_t)pBytes[3]) | (((uint32_t)pBytes[2]) << 8U) | (((uint32_t)pBytes[1]) << 16U) | (((uint32_t)pBytes[0]) << 24U); + } + + static inline void WRITE_LE32(const void* p, uint32_t v) + { + uint8_t* pBytes = (uint8_t*)p; + pBytes[0] = (uint8_t)(v); + pBytes[1] = (uint8_t)(v >> 8); + pBytes[2] = (uint8_t)(v >> 16); + pBytes[3] = (uint8_t)(v >> 24); + } + + static inline void WRITE_LE64(const void* p, uint64_t v) + { + uint8_t* pBytes = (uint8_t*)p; + pBytes[0] = (uint8_t)(v); + pBytes[1] = (uint8_t)(v >> 8); + pBytes[2] = (uint8_t)(v >> 16); + pBytes[3] = (uint8_t)(v >> 24); + pBytes[4] = (uint8_t)(v >> 32); + pBytes[5] = (uint8_t)(v >> 40); + pBytes[6] = (uint8_t)(v >> 48); + pBytes[7] = (uint8_t)(v >> 56); + } +#endif + + // Customized the very common case of reading a 24bpp pixel from memory + static inline uint32_t READ_RGB_PIXEL(const void* p) + { +#if FPNG_USE_UNALIGNED_LOADS + return READ_LE32(p) & 0xFFFFFF; +#else + const uint8_t* pBytes = (const uint8_t*)p; + return ((uint32_t)pBytes[0]) | (((uint32_t)pBytes[1]) << 8U) | (((uint32_t)pBytes[2]) << 16U); +#endif + } + + // See "Slicing by 4" CRC-32 algorithm here: + // https://create.stephan-brumme.com/crc32/ + + // Precomputed 4KB of CRC-32 tables + static const uint32_t g_crc32_4[4][256] = { + {00, 016701630226, 035603460454, 023102250672, 0733342031, 016032572217, 035130722465, 023631112643, 01666704062, 017167134244, 034065364436, 022764554610, 01155446053, 017654276275, 034756026407, 022057616621, 03555610144, 015254020362, 036356270510, 020457440736, 03266552175, 015567362353, 036465132521, 020364702707, 02333114126, 014432724300, 037530574572, 021231344754, 02400256117, 014301466331, 037203636543, 021502006765, + 07333420310, 011432210136, 032530040744, 024231670562, 07400762321, 011301152107, 032203302775, 024502532553, 06555324372, 010254514154, 033356744726, 025457174500, 06266066343, 010567656165, 033465406717, 025364236531, 04666230254, 012167400072, 031065650600, 027764060426, 04155172265, 012654742043, 031756512631, 027057322417, 05000534236, 013701304010, 030603154662, 026102764444, 05733676207, 013032046021, 030130216653, 026631426475, + 016667040620, 0166670406, 023064420274, 035765210052, 016154302611, 0655532437, 023757762245, 035056152063, 017001744642, 01700174464, 022602324216, 034103514030, 017732406673, 01033236455, 022131066227, 034630656001, 015332650764, 03433060542, 020531230330, 036230400116, 015401512755, 03300322573, 020202172301, 036503742127, 014554154706, 02255764520, 021357534352, 037456304174, 014267216737, 02566426511, 021464676363, 037365046145, + 011554460530, 07255250716, 024357000164, 032456630342, 011267722501, 07566112727, 024464342155, 032365572373, 010332364552, 06433554774, 025531704106, 033230134320, 010401026563, 06300616745, 025202446137, 033503276311, 012001270474, 04700440652, 027602610020, 031103020206, 012732132445, 04033702663, 027131552011, 031630362237, 013667574416, 05166344630, 026064114042, 030765724264, 013154636427, 05655006601, 026757256073, 030056466255, + 035556101440, 023257731666, 0355561014, 016454351232, 035265243471, 023564473657, 0466623025, 016367013203, 034330605422, 022431035604, 01533265076, 017232455250, 034403547413, 022302377635, 01200127047, 017501717261, 036003711504, 020702121722, 03600371150, 015101541376, 036730453535, 020031263713, 03133033161, 015632603347, 037665015566, 021164625740, 02066475132, 014767245314, 037156357557, 021657567771, 02755737103, 014054107325, + 032665521750, 024164311576, 07066141304, 011767771122, 032156663761, 024657053547, 07755203335, 011054433113, 033003225732, 025702415514, 06600645366, 010101075140, 033730167703, 025031757525, 06133507357, 010632337171, 031330331614, 027431501432, 04533751240, 012232161066, 031403073625, 027302643403, 04200413271, 012501223057, 030556435676, 026257205450, 05355055222, 013454665004, 030265777647, 026564147461, 05466317213, 013367527035, + 023331141260, 035430771046, 016532521634, 0233311412, 023402203251, 035303433077, 016201663605, 0500053423, 022557645202, 034256075024, 017354225656, 01455415470, 022264507233, 034565337015, 017467167667, 01366757441, 020664751324, 036165161102, 015067331770, 03766501556, 020157413315, 036656223133, 015754073741, 03055643567, 021002055346, 037703665160, 014601435712, 02100205534, 021731317377, 037030527151, 014132777723, 02633147505, + 024002561170, 032703351356, 011601101524, 07100731702, 024731623141, 032030013367, 011132243515, 07633473733, 025664265112, 033165455334, 010067605546, 06766035760, 025157127123, 033656717305, 010754547577, 06055377751, 027557371034, 031256541212, 012354711460, 04455121646, 027264033005, 031565603223, 012467453451, 04366263677, 026331475056, 030430245270, 013532015402, 05233625624, 026402737067, 030303107241, 013201357433, 05500567615, + }, { 00,03106630501,06215461202,05313251703,014433142404,017535772105,012626523606,011720313307,031066305010,032160535511,037273764212,034375154713,025455247414,026553477115,023640626616,020746016317,011260411121,012366221420,017075070323,014173640622,05653553525,06755363024,03446132727,0540702226,020206714131,023300124430,026013375333,025115545632,034635656535,037733066034,032420237737,031526407236, + 022541022242,021447612743,024754443040,027652273541,036172160646,035074750347,030367501444,033261331145,013527327252,010421517753,015732746050,016634176551,07114265656,04012455357,01301604454,02207034155,033721433363,030627203662,035534052161,036432662460,027312571767,024214341266,021107110565,022001720064,02747736373,01641106672,04552357171,07454567470,016374674777,015272044276,010161215575,013067425074, + 036036247405,035130477104,030223626607,033325016306,022405305001,021503535500,024610764203,027716154702,07050142415,04156772114,01245523617,02343313316,013463000011,010565630510,015676461213,016770251712,027256656524,024350066025,021043237726,022145407227,033665714120,030763124421,035470375322,036576545623,016230553534,015336363035,010025132736,013123702237,02603411130,01705221431,04416070332,07510640633, + 014577265647,017471455346,012762604445,011664034144,0144327243,03042517742,06351746041,05257176540,025511160657,026417750356,023704501455,020602331154,031122022253,032024612752,037337443051,034231273550,05717674766,06611044267,03502215564,0404425065,011324736362,012222106663,017131357160,014037567461,034771571776,037677341277,032564110574,031462720075,020342433372,023244203673,026157052170,025051662471, + 07340714113,04246124412,01155375311,02053545610,013773656517,010675066016,015566237715,016460407214,036326411103,035220221402,030133070301,033035640600,022715553507,021613363006,024500132705,027406702204,016120305032,015026535533,010335764230,013233154731,02513247436,01415477137,04706626634,07600016335,027146000022,024040630523,021353461220,022255251721,033575142426,030473772127,035760523624,036666313325, + 025601736351,026707106650,023414357153,020512567452,031232674755,032334044254,037027215557,034121425056,014667433341,017761203640,012472052143,011574662442,0254571745,03352341244,06041110547,05147720046,034461327270,037567517771,032674746072,031772176573,020052265674,023154455375,026247604476,025341034177,05407022260,06501612761,03612443062,0714273563,011034160664,012132750365,017221501466,014327331167, + 031376553516,032270363017,037163132714,034065702215,025745411112,026643221413,023550070310,020456640611,0310656506,03216066007,06105237704,05003407205,014723714102,017625124403,012536375300,011430545601,020116142437,023010772136,026303523635,025205313334,034525000033,037423630532,032730461231,031636251730,011170247427,012076477126,017365626625,014263016324,05543305023,06445535522,03756764221,0650154720, + 013637571754,010731341255,015422110556,016524720057,07204433350,04302203651,01011052152,02117662453,022651674744,021757044245,024444215546,027542425047,036262736340,035364106641,030077357142,033171567443,02457160675,01551750374,04642501477,07744331176,016064022271,015162612770,010271443073,013377273572,033431265665,030537455364,035624604467,036722034166,027002327261,024104517760,021217746063,022311176562, + }, { 00,0160465067,0341152156,0221537131,0702324334,0662741353,0443276262,0523613205,01604650670,01764235617,01545702726,01425367741,01106574544,01066111523,01247426412,01327043475,03411521560,03571144507,03750473436,03630016451,03313605654,03273260633,03052757702,03132332765,02215371310,02375714377,02154223246,02034646221,02517055024,02477430043,02656107172,02736562115, + 07023243340,07143626327,07362311216,07202774271,07721167074,07641502013,07460035122,07500450145,06627413530,06747076557,06566541466,06406124401,06125737604,06045352663,06264665752,06304200735,04432762620,04552307647,04773630776,04613255711,04330446514,04250023573,04071514442,04111171425,05236132050,05356557037,05177060106,05017405161,05534216364,05454673303,05675344232,05715721255, + 016046506700,016126163767,016307454656,016267031631,016744622434,016624247453,016405770562,016565315505,017642356170,017722733117,017503204026,017463661041,017140072244,017020417223,017201120312,017361545375,015457027260,015537442207,015716175336,015676510351,015355303154,015235766133,015014251002,015174634065,014253677410,014333212477,014112725546,014072340521,014551553724,014431136743,014610401672,014770064615, + 011065745440,011105320427,011324617516,011244272571,011767461774,011607004713,011426533622,011546156645,010661115230,010701570257,010520047366,010440422301,010163231104,010003654163,010222363052,010342706035,012474264120,012514601147,012735336076,012655753011,012376140214,012216525273,012037012342,012157477325,013270434750,013310051737,013131566606,013051103661,013572710464,013412375403,013633642532,013753227555, + 034115215600,034075670667,034254347756,034334722731,034617131534,034777554553,034556063462,034436406405,035711445070,035671020017,035450517126,035530172141,035013761344,035173304323,035352633212,035232256275,037504734360,037464351307,037645666236,037725203251,037206410054,037366075033,037147542102,037027127165,036300164510,036260501577,036041036446,036121453421,036402240624,036562625643,036743312772,036623777715, + 033136056540,033056433527,033277104416,033317561471,033634372674,033754717613,033575220722,033415645745,032732606330,032652263357,032473754266,032513331201,032030522004,032150147063,032371470152,032211015135,030527577020,030447112047,030666425176,030706040111,030225653314,030345236373,030164701242,030004364225,031323327650,031243742637,031062275706,031102610761,031421003564,031541466503,031760151432,031600534455, + 022153713100,022033376167,022212641056,022372224031,022651437234,022731052253,022510565362,022470100305,023757143770,023637526717,023416011626,023576474641,023055267444,023135602423,023314335512,023274750575,021542232460,021422657407,021603360536,021763705551,021240116754,021320573733,021101044602,021061421665,020346462210,020226007277,020007530346,020167155321,020444746124,020524323143,020705614072,020665271015, + 025170550240,025010135227,025231402316,025351067371,025672674174,025712211113,025533726022,025453343045,024774300430,024614765457,024435252566,024555637501,024076024704,024116441763,024337176652,024257513635,026561071720,026401414747,026620123676,026740546611,026263355414,026303730473,026122207542,026042662525,027365621150,027205244137,027024773006,027144316061,027467505264,027507160203,027726457332,027646032355, + }, { 00,027057063545,025202344213,02255327756,021730513527,06767570062,04532657734,023565634271,030555024357,017502047612,015757360144,032700303401,011265537670,036232554335,034067673463,013030610126,012006253637,035051230372,037204117424,010253174161,033736740310,014761723655,016534404103,031563467446,022553277560,05504214025,07751133773,020706150236,03263764047,024234707502,026061420254,01036443711, + 024014527476,03043544133,01216663665,026241600320,05724034151,022773057414,020526370342,07571313607,014541503721,033516560264,031743647532,016714624077,035271010206,012226073743,010073354015,037024337550,036012774241,011045717704,013210430052,034247453517,017722267766,030775204223,032520123575,015577140030,06547750116,021510733453,023745414305,04712477640,027277243431,0220220174,02075107622,025022164367, + 023305054075,04352037530,06107310266,021150373723,02435547552,025462524017,027637603741,0660660204,013650070322,034607013667,036452334131,011405357474,032160563605,015137500340,017362627416,030335644153,031303207642,016354264307,014101143451,033156120114,010433714365,037464777620,035631450176,012666433433,01656223515,026601240050,024454167706,03403104243,020166730032,07131753577,05364474221,022333417764, + 07311573403,020346510146,022113637610,05144654355,026421060124,01476003461,03623324337,024674347672,037644557754,010613534211,012446613547,035411670002,016174044273,031123027736,033376300060,014321363525,015317720234,032340743771,030115464027,017142407562,034427233713,013470250256,011625177500,036672114045,025642704163,02615767426,0440440370,027417423635,04172217444,023125274101,021370153657,06327130312, + 035526333073,012571350536,010724077260,037773014725,014216620554,033241643011,031014564747,016043507202,05073317324,022024374661,020271053137,07226030472,024743604603,03714667346,01541540410,026516523155,027520160644,0577103301,02722224457,025775247112,06210473363,021247410626,023012737170,04045754435,017075144513,030022127056,032277200700,015220263245,036745457034,011712434571,013547713227,034510770762, + 011532614405,036565677140,034730550616,013767533353,030202307122,017255364467,015000043331,032057020674,021067630752,06030653217,04265574541,023232517004,0757323275,027700340730,025555067066,02502004523,03534447232,024563424777,026736703021,01761760564,022204154715,05253137250,07006210506,020051273043,033061463165,014036400420,016263727376,031234744633,012751170442,035706113107,037553234651,010504257314, + 016623367006,031674304543,033421023215,014476040750,037113674521,010144617064,012311530732,035346553277,026376343351,01321320614,03174007142,024123064407,07446650676,020411633333,022644514465,05613577120,04625134631,023672157374,021427270422,06470213167,025115427316,02142444653,0317763105,027340700440,034370110566,013327173023,011172254775,036125237230,015440403041,032417460504,030642747252,017615724717, + 032637640470,015660623135,017435504663,030462567326,013107353157,034150330412,036305017344,011352074601,02362664727,025335607262,027160520534,0137543071,023452377200,04405314745,06650033013,021607050556,020631413247,07666470702,05433757054,022464734511,01101100760,026156163225,024303244573,03354227036,010364437110,037333454455,035166773303,012131710646,031454124437,016403147172,014656260624,033601203361, + } }; + + static uint32_t crc32_slice_by_4(const void* pData, size_t data_len, uint32_t cur_crc32 = 0) + { + uint32_t crc = ~cur_crc32; + const uint32_t* pData32 = static_cast(pData); + + for (; data_len >= sizeof(uint32_t); ++pData32, data_len -= 4) + { + uint32_t v = READ_LE32(pData32) ^ crc; + crc = g_crc32_4[0][v >> 24] ^ g_crc32_4[1][(v >> 16) & 0xFF] ^ g_crc32_4[2][(v >> 8) & 0xFF] ^ g_crc32_4[3][v & 0xFF]; + } + + for (const uint8_t* pData8 = reinterpret_cast(pData32); data_len; --data_len) + crc = (crc >> 8) ^ g_crc32_4[0][(crc & 0xFF) ^ *pData8++]; + + return ~crc; + } + +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + // See Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction": + // https://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf + // Requires PCLMUL and SSE 4.1. This function skips Step 1 (fold by 4) for simplicity/less code. + static uint32_t crc32_pclmul(const uint8_t* p, size_t size, uint32_t crc) + { + assert(size >= 16); + + // See page 22 (bit reflected constants for gzip) +#ifdef _MSC_VER + static const uint64_t __declspec(align(16)) +#else + static const uint64_t __attribute__((aligned(16))) +#endif + s_u[2] = { 0x1DB710641, 0x1F7011641 }, s_k5k0[2] = { 0x163CD6124, 0 }, s_k3k4[2] = { 0x1751997D0, 0xCCAA009E }; + + // Load first 16 bytes, apply initial CRC32 + __m128i b = _mm_xor_si128(_mm_cvtsi32_si128(~crc), _mm_loadu_si128(reinterpret_cast(p))); + + // We're skipping directly to Step 2 page 12 - iteratively folding by 1 (by 4 is overkill for our needs) + const __m128i k3k4 = _mm_load_si128(reinterpret_cast(s_k3k4)); + + for (size -= 16, p += 16; size >= 16; size -= 16, p += 16) + b = _mm_xor_si128(_mm_xor_si128(_mm_clmulepi64_si128(b, k3k4, 17), _mm_loadu_si128(reinterpret_cast(p))), _mm_clmulepi64_si128(b, k3k4, 0)); + + // Final stages: fold to 64-bits, 32-bit Barrett reduction + const __m128i z = _mm_set_epi32(0, ~0, 0, ~0), u = _mm_load_si128(reinterpret_cast(s_u)); + b = _mm_xor_si128(_mm_srli_si128(b, 8), _mm_clmulepi64_si128(b, k3k4, 16)); + b = _mm_xor_si128(_mm_clmulepi64_si128(_mm_and_si128(b, z), _mm_loadl_epi64(reinterpret_cast(s_k5k0)), 0), _mm_srli_si128(b, 4)); + return ~_mm_extract_epi32(_mm_xor_si128(b, _mm_clmulepi64_si128(_mm_and_si128(_mm_clmulepi64_si128(_mm_and_si128(b, z), u, 16), z), u, 0)), 1); + } + + static uint32_t crc32_sse41_simd(const unsigned char* buf, size_t len, uint32_t prev_crc32) + { + if (len < 16) + return crc32_slice_by_4(buf, len, prev_crc32); + + uint32_t simd_len = len & ~15; + uint32_t c = crc32_pclmul(buf, simd_len, prev_crc32); + return crc32_slice_by_4(buf + simd_len, len - simd_len, c); + } +#endif + +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + +#ifndef _MSC_VER + static void do_cpuid(uint32_t eax, uint32_t ecx, uint32_t* regs) + { + uint32_t ebx = 0, edx = 0; + +#if defined(__PIC__) && defined(__i386__) + __asm__("movl %%ebx, %%edi;" + "cpuid;" + "xchgl %%ebx, %%edi;" + : "=D"(ebx), "+a"(eax), "+c"(ecx), "=d"(edx)); +#else + __asm__("cpuid;" : "+b"(ebx), "+a"(eax), "+c"(ecx), "=d"(edx)); +#endif + + regs[0] = eax; regs[1] = ebx; regs[2] = ecx; regs[3] = edx; + } +#endif + + struct cpu_info + { + cpu_info() { memset(this, 0, sizeof(*this)); } + + bool m_initialized, m_has_fpu, m_has_mmx, m_has_sse, m_has_sse2, m_has_sse3, m_has_ssse3, m_has_sse41, m_has_sse42, m_has_avx, m_has_avx2, m_has_pclmulqdq; + + void init() + { + if (m_initialized) + return; + + int regs[4]; + +#ifdef _MSC_VER + __cpuid(regs, 0); +#else + do_cpuid(0, 0, (uint32_t*)regs); +#endif + + const uint32_t max_eax = regs[0]; + if (max_eax >= 1U) + { +#ifdef _MSC_VER + __cpuid(regs, 1); +#else + do_cpuid(1, 0, (uint32_t*)regs); +#endif + extract_x86_flags(regs[2], regs[3]); + } + + if (max_eax >= 7U) + { +#ifdef _MSC_VER + __cpuidex(regs, 7, 0); +#else + do_cpuid(7, 0, (uint32_t*)regs); +#endif + extract_x86_extended_flags(regs[1]); + } + + m_initialized = true; + } + + bool can_use_sse41() const { return m_has_sse && m_has_sse2 && m_has_sse3 && m_has_ssse3 && m_has_sse41; } + bool can_use_pclmul() const { return m_has_pclmulqdq && can_use_sse41(); } + + private: + void extract_x86_flags(uint32_t ecx, uint32_t edx) + { + m_has_fpu = (edx & (1 << 0)) != 0; m_has_mmx = (edx & (1 << 23)) != 0; m_has_sse = (edx & (1 << 25)) != 0; m_has_sse2 = (edx & (1 << 26)) != 0; + m_has_sse3 = (ecx & (1 << 0)) != 0; m_has_ssse3 = (ecx & (1 << 9)) != 0; m_has_sse41 = (ecx & (1 << 19)) != 0; m_has_sse42 = (ecx & (1 << 20)) != 0; + m_has_pclmulqdq = (ecx & (1 << 1)) != 0; m_has_avx = (ecx & (1 << 28)) != 0; + } + + void extract_x86_extended_flags(uint32_t ebx) { m_has_avx2 = (ebx & (1 << 5)) != 0; } + }; + + cpu_info g_cpu_info; + + void fpng_init() + { + g_cpu_info.init(); + } +#else + void fpng_init() + { + } +#endif + + bool fpng_cpu_supports_sse41() + { +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + assert(g_cpu_info.m_initialized); + return g_cpu_info.can_use_sse41(); +#else + return false; +#endif + } + + uint32_t fpng_crc32(const void* pData, size_t size, uint32_t prev_crc32) + { +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + if (g_cpu_info.can_use_pclmul()) + return crc32_sse41_simd(static_cast(pData), size, prev_crc32); +#endif + + return crc32_slice_by_4(pData, size, prev_crc32); + } + +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + // See "Fast Computation of Adler32 Checksums": + // https://www.intel.com/content/www/us/en/developer/articles/technical/fast-computation-of-adler32-checksums.html + // SSE 4.1, 16 bytes per iteration + static uint32_t adler32_sse_16(const uint8_t* p, size_t len, uint32_t initial) + { + uint32_t s1 = initial & 0xFFFF, s2 = initial >> 16; + const uint32_t K = 65521; + + while (len >= 16) + { + __m128i a = _mm_setr_epi32(s1, 0, 0, 0), b = _mm_setzero_si128(), c = _mm_setzero_si128(), d = _mm_setzero_si128(), + e = _mm_setzero_si128(), f = _mm_setzero_si128(), g = _mm_setzero_si128(), h = _mm_setzero_si128(); + + const size_t n = minimum(len >> 4, 5552); + + for (size_t i = 0; i < n; i++) + { + const __m128i v = _mm_loadu_si128((const __m128i*)(p + i * 16)); + a = _mm_add_epi32(a, _mm_cvtepu8_epi32(_mm_shuffle_epi32(v, _MM_SHUFFLE(0, 0, 0, 0)))); b = _mm_add_epi32(b, a); + c = _mm_add_epi32(c, _mm_cvtepu8_epi32(_mm_shuffle_epi32(v, _MM_SHUFFLE(1, 1, 1, 1)))); d = _mm_add_epi32(d, c); + e = _mm_add_epi32(e, _mm_cvtepu8_epi32(_mm_shuffle_epi32(v, _MM_SHUFFLE(2, 2, 2, 2)))); f = _mm_add_epi32(f, e); + g = _mm_add_epi32(g, _mm_cvtepu8_epi32(_mm_shuffle_epi32(v, _MM_SHUFFLE(3, 3, 3, 3)))); h = _mm_add_epi32(h, g); + } + + uint32_t sa[16], sb[16]; + _mm_storeu_si128((__m128i*)sa, a); _mm_storeu_si128((__m128i*)(sa + 4), c); + _mm_storeu_si128((__m128i*)sb, b); _mm_storeu_si128((__m128i*)(sb + 4), d); + _mm_storeu_si128((__m128i*)(sa + 8), e); _mm_storeu_si128((__m128i*)(sa + 12), g); + _mm_storeu_si128((__m128i*)(sb + 8), f); _mm_storeu_si128((__m128i*)(sb + 12), h); + + // This could be vectorized, but it's only executed every 5552*16 iterations. + uint64_t vs1 = 0; + for (uint32_t i = 0; i < 16; i++) + vs1 += sa[i]; + + uint64_t vs2_a = 0; + for (uint32_t i = 0; i < 16; i++) + vs2_a += sa[i] * (uint64_t)i; + uint64_t vs2_b = 0; + for (uint32_t i = 0; i < 16; i++) + vs2_b += sb[i]; + vs2_b *= 16U; + uint64_t vs2 = vs2_b - vs2_a + s2; + + s1 = (uint32_t)(vs1 % K); + s2 = (uint32_t)(vs2 % K); + + p += n * 16; + len -= n * 16; + } + + for (; len; len--) + { + s1 += *p++; + s2 += s1; + } + + return (s1 % K) | ((s2 % K) << 16); + } +#endif + + static uint32_t fpng_adler32_scalar(const uint8_t* ptr, size_t buf_len, uint32_t adler) + { + uint32_t i, s1 = (uint32_t)(adler & 0xffff), s2 = (uint32_t)(adler >> 16); uint32_t block_len = (uint32_t)(buf_len % 5552); + if (!ptr) return FPNG_ADLER32_INIT; + while (buf_len) { + for (i = 0; i + 7 < block_len; i += 8, ptr += 8) { + s1 += ptr[0], s2 += s1; s1 += ptr[1], s2 += s1; s1 += ptr[2], s2 += s1; s1 += ptr[3], s2 += s1; + s1 += ptr[4], s2 += s1; s1 += ptr[5], s2 += s1; s1 += ptr[6], s2 += s1; s1 += ptr[7], s2 += s1; + } + for (; i < block_len; ++i) s1 += *ptr++, s2 += s1; + s1 %= 65521U, s2 %= 65521U; buf_len -= block_len; block_len = 5552; + } + return (s2 << 16) + s1; + } + + uint32_t fpng_adler32(const void* pData, size_t size, uint32_t adler) + { +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + if (g_cpu_info.can_use_sse41()) + return adler32_sse_16((const uint8_t*)pData, size, adler); +#endif + return fpng_adler32_scalar((const uint8_t*)pData, size, adler); + } + + // Ensure we've been configured for endianness correctly. + static inline bool endian_check() + { + uint32_t endian_check = 0; + WRITE_LE32(&endian_check, 0x1234ABCD); + const uint32_t first_byte = reinterpret_cast(&endian_check)[0]; + return first_byte == 0xCD; + } + + static const uint16_t g_defl_len_sym[256] = { + 257,258,259,260,261,262,263,264,265,265,266,266,267,267,268,268,269,269,269,269,270,270,270,270,271,271,271,271,272,272,272,272, + 273,273,273,273,273,273,273,273,274,274,274,274,274,274,274,274,275,275,275,275,275,275,275,275,276,276,276,276,276,276,276,276, + 277,277,277,277,277,277,277,277,277,277,277,277,277,277,277,277,278,278,278,278,278,278,278,278,278,278,278,278,278,278,278,278, + 279,279,279,279,279,279,279,279,279,279,279,279,279,279,279,279,280,280,280,280,280,280,280,280,280,280,280,280,280,280,280,280, + 281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281,281, + 282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282,282, + 283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283,283, + 284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,284,285 }; + + static const uint8_t g_defl_len_extra[256] = { + 0,0,0,0,0,0,0,0,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3,3, + 4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4, + 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5, + 5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,5,0 }; + + static const uint8_t g_defl_small_dist_sym[512] = { + 0,1,2,3,4,4,5,5,6,6,6,6,7,7,7,7,8,8,8,8,8,8,8,8,9,9,9,9,9,9,9,9,10,10,10,10,10,10,10,10,10,10,10,10,10,10,10,10,11,11,11,11,11,11, + 11,11,11,11,11,11,11,11,11,11,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,12,13, + 13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,13,14,14,14,14,14,14,14,14,14,14,14,14, + 14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14,14, + 14,14,14,14,14,14,14,14,14,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15, + 15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,16,16,16,16,16,16,16,16,16,16,16,16,16, + 16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16, + 16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16, + 16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,17,17,17,17,17,17,17,17,17,17,17,17,17,17, + 17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17, + 17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17, + 17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17,17 }; + + static const uint32_t g_bitmasks[17] = { 0x0000, 0x0001, 0x0003, 0x0007, 0x000F, 0x001F, 0x003F, 0x007F, 0x00FF, 0x01FF, 0x03FF, 0x07FF, 0x0FFF, 0x1FFF, 0x3FFF, 0x7FFF, 0xFFFF }; + + // Huffman tables generated by fpng_test -t @filelist.txt. Total alpha files : 1440, Total opaque files : 5627. + // Feel free to retrain the encoder on your opaque/alpha PNG files by setting FPNG_TRAIN_HUFFMAN_TABLES and running fpng_test with the -t option. + static const uint8_t g_dyn_huff_3[] = { + 120, 1, 237, 195, 3, 176, 110, 89, 122, 128, 225, 247, 251, 214, 218, 248, 113, 124, 173, 190, 109, 12, 50, 201, 196, 182, 109, 219, 182, 109, 219, 182, + 109, 219, 201, 36, 147, 153, 105, 235, 246, 53, 142, 207, 143, 141, 181, 214, 151, 93, 117, 170, 78, 117, 117, 58, 206, 77, 210, 217, 169, 122 }; + const uint32_t DYN_HUFF_3_BITBUF = 30, DYN_HUFF_3_BITBUF_SIZE = 7; + static const struct { uint8_t m_code_size; uint16_t m_code; } g_dyn_huff_3_codes[288] = { + {2,0},{4,2},{4,10},{5,14},{5,30},{6,25},{6,57},{6,5},{6,37},{7,3},{7,67},{7,35},{7,99},{8,11},{8,139},{8,75},{8,203},{8,43},{8,171},{8,107},{9,135},{9,391},{9,71},{9,327},{9,199},{9,455},{9,39},{9,295},{9,167},{9,423},{9,103},{10,183}, + {9,359},{10,695},{10,439},{10,951},{10,119},{10,631},{10,375},{10,887},{10,247},{10,759},{10,503},{11,975},{11,1999},{11,47},{11,1071},{12,1199},{11,559},{12,3247},{12,687},{11,1583},{12,2735},{12,1711},{12,3759},{12,431},{12,2479},{12,1455},{12,3503},{12,943},{12,2991},{12,1967},{12,4015},{12,111}, + {12,2159},{12,1135},{12,3183},{12,623},{12,2671},{12,1647},{12,3695},{12,367},{12,2415},{12,1391},{12,3439},{12,879},{12,2927},{12,1903},{12,3951},{12,239},{12,2287},{12,1263},{12,3311},{12,751},{12,2799},{12,1775},{12,3823},{12,495},{12,2543},{12,1519},{12,3567},{12,1007},{12,3055},{12,2031},{12,4079},{12,31}, + {12,2079},{12,1055},{12,3103},{12,543},{12,2591},{12,1567},{12,3615},{12,287},{12,2335},{12,1311},{12,3359},{12,799},{12,2847},{12,1823},{12,3871},{12,159},{12,2207},{12,1183},{12,3231},{12,671},{12,2719},{12,1695},{12,3743},{12,415},{12,2463},{12,1439},{12,3487},{12,927},{12,2975},{12,1951},{12,3999},{12,95}, + {12,2143},{12,1119},{12,3167},{12,607},{12,2655},{12,1631},{12,3679},{12,351},{12,2399},{12,1375},{12,3423},{12,863},{12,2911},{12,1887},{12,3935},{12,223},{12,2271},{12,1247},{12,3295},{12,735},{12,2783},{12,1759},{12,3807},{12,479},{12,2527},{12,1503},{12,3551},{12,991},{12,3039},{12,2015},{12,4063},{12,63}, + {12,2111},{12,1087},{12,3135},{12,575},{12,2623},{12,1599},{12,3647},{12,319},{12,2367},{12,1343},{12,3391},{12,831},{12,2879},{12,1855},{12,3903},{12,191},{12,2239},{12,1215},{12,3263},{12,703},{12,2751},{12,1727},{12,3775},{12,447},{12,2495},{12,1471},{12,3519},{12,959},{12,3007},{12,1983},{12,4031},{12,127}, + {12,2175},{12,1151},{12,3199},{12,639},{12,2687},{12,1663},{12,3711},{12,383},{12,2431},{12,1407},{12,3455},{12,895},{12,2943},{11,303},{12,1919},{12,3967},{11,1327},{12,255},{11,815},{11,1839},{11,175},{10,1015},{10,15},{10,527},{10,271},{10,783},{10,143},{10,655},{10,399},{10,911},{10,79},{10,591}, + {9,231},{10,335},{9,487},{9,23},{9,279},{9,151},{9,407},{9,87},{9,343},{9,215},{9,471},{9,55},{8,235},{8,27},{8,155},{8,91},{8,219},{8,59},{8,187},{8,123},{7,19},{7,83},{7,51},{7,115},{6,21},{6,53},{6,13},{6,45},{5,1},{5,17},{5,9},{4,6}, + {12,2303},{6,29},{0,0},{0,0},{8,251},{0,0},{0,0},{8,7},{0,0},{10,847},{0,0},{10,207},{12,1279},{10,719},{12,3327},{12,767},{12,2815},{12,1791},{12,3839},{12,511},{12,2559},{12,1535},{9,311},{12,3583},{12,1023},{12,3071},{10,463},{12,2047},{6,61},{12,4095},{0,0},{0,0} + }; + + static const uint8_t g_dyn_huff_4[] = { + 120, 1, 229, 196, 99, 180, 37, 103, 218, 128, 225, 251, 121, 171, 106, 243, 216, 231, 180, 109, 196, 182, 51, 51, 73, 6, 201, 216, 182, 109, 219, 182, + 17, 140, 98, 219, 102, 219, 60, 125, 172, 205, 170, 122, 159, 111, 213, 143, 179, 214, 94, 189, 58, 153, 104, 166, 103, 190, 247, 199, 117 }; + const uint32_t DYN_HUFF_4_BITBUF = 1, DYN_HUFF_4_BITBUF_SIZE = 2; + static const struct { uint8_t m_code_size; uint16_t m_code; } g_dyn_huff_4_codes[288] = { + {2,0},{4,2},{5,6},{6,30},{6,62},{6,1},{7,41},{7,105},{7,25},{7,89},{7,57},{7,121},{8,117},{8,245},{8,13},{8,141},{8,77},{8,205},{8,45},{8,173},{8,109},{8,237},{8,29},{8,157},{8,93},{8,221},{8,61},{9,83},{9,339},{9,211},{9,467},{9,51}, + {9,307},{9,179},{9,435},{9,115},{9,371},{9,243},{9,499},{9,11},{9,267},{9,139},{9,395},{9,75},{9,331},{9,203},{9,459},{9,43},{9,299},{10,7},{10,519},{10,263},{10,775},{10,135},{10,647},{10,391},{10,903},{10,71},{10,583},{10,327},{10,839},{10,199},{10,711},{10,455}, + {10,967},{10,39},{10,551},{10,295},{10,807},{10,167},{10,679},{10,423},{10,935},{10,103},{10,615},{11,463},{11,1487},{11,975},{10,359},{10,871},{10,231},{11,1999},{11,47},{11,1071},{11,559},{10,743},{10,487},{11,1583},{11,303},{11,1327},{11,815},{11,1839},{11,175},{11,1199},{11,687},{11,1711}, + {11,431},{11,1455},{11,943},{11,1967},{11,111},{11,1135},{11,623},{11,1647},{11,367},{11,1391},{11,879},{11,1903},{11,239},{11,1263},{11,751},{11,1775},{11,495},{11,1519},{11,1007},{11,2031},{11,31},{11,1055},{11,543},{11,1567},{11,287},{11,1311},{11,799},{11,1823},{11,159},{11,1183},{11,671},{11,1695}, + {11,415},{11,1439},{11,927},{11,1951},{11,95},{11,1119},{11,607},{11,1631},{11,351},{11,1375},{11,863},{11,1887},{11,223},{11,1247},{11,735},{11,1759},{11,479},{11,1503},{11,991},{11,2015},{11,63},{11,1087},{11,575},{11,1599},{11,319},{11,1343},{11,831},{11,1855},{11,191},{11,1215},{11,703},{11,1727}, + {11,447},{11,1471},{11,959},{11,1983},{11,127},{11,1151},{11,639},{11,1663},{11,383},{10,999},{10,23},{10,535},{10,279},{11,1407},{11,895},{11,1919},{11,255},{11,1279},{10,791},{10,151},{10,663},{10,407},{10,919},{10,87},{10,599},{10,343},{10,855},{10,215},{10,727},{10,471},{10,983},{10,55}, + {10,567},{10,311},{10,823},{10,183},{10,695},{10,439},{10,951},{10,119},{10,631},{10,375},{10,887},{10,247},{10,759},{10,503},{10,1015},{10,15},{10,527},{10,271},{10,783},{10,143},{10,655},{10,399},{9,171},{9,427},{9,107},{9,363},{9,235},{9,491},{9,27},{9,283},{9,155},{9,411}, + {9,91},{9,347},{9,219},{9,475},{9,59},{9,315},{9,187},{9,443},{8,189},{9,123},{8,125},{8,253},{8,3},{8,131},{8,67},{8,195},{8,35},{8,163},{8,99},{8,227},{8,19},{7,5},{7,69},{7,37},{7,101},{7,21},{7,85},{6,33},{6,17},{6,49},{5,22},{4,10}, + {12,2047},{0,0},{6,9},{0,0},{0,0},{0,0},{8,147},{0,0},{0,0},{7,53},{0,0},{9,379},{0,0},{9,251},{10,911},{10,79},{11,767},{10,591},{10,335},{10,847},{10,207},{10,719},{11,1791},{11,511},{9,507},{11,1535},{11,1023},{12,4095},{5,14},{0,0},{0,0},{0,0} + }; + +#define PUT_BITS(bb, ll) do { uint32_t b = bb, l = ll; assert((l) >= 0 && (l) <= 16); assert((b) < (1ULL << (l))); bit_buf |= (((uint64_t)(b)) << bit_buf_size); bit_buf_size += (l); assert(bit_buf_size <= 64); } while(0) +#define PUT_BITS_CZ(bb, ll) do { uint32_t b = bb, l = ll; assert((l) >= 1 && (l) <= 16); assert((b) < (1ULL << (l))); bit_buf |= (((uint64_t)(b)) << bit_buf_size); bit_buf_size += (l); assert(bit_buf_size <= 64); } while(0) + +#define PUT_BITS_FLUSH do { \ + if ((dst_ofs + 8) > dst_buf_size) \ + return 0; \ + WRITE_LE64(pDst + dst_ofs, bit_buf); \ + uint32_t bits_to_shift = bit_buf_size & ~7; \ + dst_ofs += (bits_to_shift >> 3); \ + assert(bits_to_shift < 64); \ + bit_buf = bit_buf >> bits_to_shift; \ + bit_buf_size -= bits_to_shift; \ +} while(0) + +#define PUT_BITS_FORCE_FLUSH do { \ + while (bit_buf_size > 0) \ + { \ + if ((dst_ofs + 1) > dst_buf_size) \ + return 0; \ + *(uint8_t*)(pDst + dst_ofs) = (uint8_t)bit_buf; \ + dst_ofs++; \ + bit_buf >>= 8; \ + bit_buf_size -= 8; \ + } \ +} while(0) + + enum + { + DEFL_MAX_HUFF_TABLES = 3, + DEFL_MAX_HUFF_SYMBOLS = 288, + DEFL_MAX_HUFF_SYMBOLS_0 = 288, + DEFL_MAX_HUFF_SYMBOLS_1 = 32, + DEFL_MAX_HUFF_SYMBOLS_2 = 19, + DEFL_LZ_DICT_SIZE = 32768, + DEFL_LZ_DICT_SIZE_MASK = DEFL_LZ_DICT_SIZE - 1, + DEFL_MIN_MATCH_LEN = 3, + DEFL_MAX_MATCH_LEN = 258 + }; + +#if FPNG_TRAIN_HUFFMAN_TABLES + uint64_t g_huff_counts[HUFF_COUNTS_SIZE]; +#endif + + struct defl_huff + { + uint16_t m_huff_count[DEFL_MAX_HUFF_TABLES][DEFL_MAX_HUFF_SYMBOLS]; + uint16_t m_huff_codes[DEFL_MAX_HUFF_TABLES][DEFL_MAX_HUFF_SYMBOLS]; + uint8_t m_huff_code_sizes[DEFL_MAX_HUFF_TABLES][DEFL_MAX_HUFF_SYMBOLS]; + }; + + struct defl_sym_freq + { + uint16_t m_key; + uint16_t m_sym_index; + }; + +#define DEFL_CLEAR_OBJ(obj) memset(&(obj), 0, sizeof(obj)) + + static defl_sym_freq* defl_radix_sort_syms(uint32_t num_syms, defl_sym_freq* pSyms0, defl_sym_freq* pSyms1) + { + uint32_t total_passes = 2, pass_shift, pass, i, hist[256 * 2]; defl_sym_freq* pCur_syms = pSyms0, * pNew_syms = pSyms1; DEFL_CLEAR_OBJ(hist); + for (i = 0; i < num_syms; i++) { uint32_t freq = pSyms0[i].m_key; hist[freq & 0xFF]++; hist[256 + ((freq >> 8) & 0xFF)]++; } + while ((total_passes > 1) && (num_syms == hist[(total_passes - 1) * 256])) total_passes--; + for (pass_shift = 0, pass = 0; pass < total_passes; pass++, pass_shift += 8) + { + const uint32_t* pHist = &hist[pass << 8]; + uint32_t offsets[256], cur_ofs = 0; + for (i = 0; i < 256; i++) { offsets[i] = cur_ofs; cur_ofs += pHist[i]; } + for (i = 0; i < num_syms; i++) pNew_syms[offsets[(pCur_syms[i].m_key >> pass_shift) & 0xFF]++] = pCur_syms[i]; + { defl_sym_freq* t = pCur_syms; pCur_syms = pNew_syms; pNew_syms = t; } + } + return pCur_syms; + } + + // defl_calculate_minimum_redundancy() originally written by: Alistair Moffat, alistair@cs.mu.oz.au, Jyrki Katajainen, jyrki@diku.dk, November 1996. + static void defl_calculate_minimum_redundancy(defl_sym_freq* A, int n) + { + int root, leaf, next, avbl, used, dpth; + if (n == 0) return; else if (n == 1) { A[0].m_key = 1; return; } + A[0].m_key += A[1].m_key; root = 0; leaf = 2; + for (next = 1; next < n - 1; next++) + { + if (leaf >= n || A[root].m_key < A[leaf].m_key) { A[next].m_key = A[root].m_key; A[root++].m_key = (uint16_t)next; } + else A[next].m_key = A[leaf++].m_key; + if (leaf >= n || (root < next && A[root].m_key < A[leaf].m_key)) { A[next].m_key = (uint16_t)(A[next].m_key + A[root].m_key); A[root++].m_key = (uint16_t)next; } + else A[next].m_key = (uint16_t)(A[next].m_key + A[leaf++].m_key); + } + A[n - 2].m_key = 0; for (next = n - 3; next >= 0; next--) A[next].m_key = A[A[next].m_key].m_key + 1; + avbl = 1; used = dpth = 0; root = n - 2; next = n - 1; + while (avbl > 0) + { + while (root >= 0 && (int)A[root].m_key == dpth) { used++; root--; } + while (avbl > used) { A[next--].m_key = (uint16_t)(dpth); avbl--; } + avbl = 2 * used; dpth++; used = 0; + } + } + + // Limits canonical Huffman code table's max code size. + enum { DEFL_MAX_SUPPORTED_HUFF_CODESIZE = 32 }; + static void defl_huffman_enforce_max_code_size(int* pNum_codes, int code_list_len, int max_code_size) + { + int i; uint32_t total = 0; if (code_list_len <= 1) return; + for (i = max_code_size + 1; i <= DEFL_MAX_SUPPORTED_HUFF_CODESIZE; i++) pNum_codes[max_code_size] += pNum_codes[i]; + for (i = max_code_size; i > 0; i--) total += (((uint32_t)pNum_codes[i]) << (max_code_size - i)); + while (total != (1UL << max_code_size)) + { + pNum_codes[max_code_size]--; + for (i = max_code_size - 1; i > 0; i--) if (pNum_codes[i]) { pNum_codes[i]--; pNum_codes[i + 1] += 2; break; } + total--; + } + } + + static void defl_optimize_huffman_table(defl_huff* d, int table_num, int table_len, int code_size_limit, int static_table) + { + int i, j, l, num_codes[1 + DEFL_MAX_SUPPORTED_HUFF_CODESIZE]; uint32_t next_code[DEFL_MAX_SUPPORTED_HUFF_CODESIZE + 1]; DEFL_CLEAR_OBJ(num_codes); + if (static_table) + { + for (i = 0; i < table_len; i++) num_codes[d->m_huff_code_sizes[table_num][i]]++; + } + else + { + defl_sym_freq syms0[DEFL_MAX_HUFF_SYMBOLS], syms1[DEFL_MAX_HUFF_SYMBOLS], * pSyms; + int num_used_syms = 0; + const uint16_t* pSym_count = &d->m_huff_count[table_num][0]; + for (i = 0; i < table_len; i++) if (pSym_count[i]) { syms0[num_used_syms].m_key = (uint16_t)pSym_count[i]; syms0[num_used_syms++].m_sym_index = (uint16_t)i; } + + pSyms = defl_radix_sort_syms(num_used_syms, syms0, syms1); defl_calculate_minimum_redundancy(pSyms, num_used_syms); + + for (i = 0; i < num_used_syms; i++) num_codes[pSyms[i].m_key]++; + + defl_huffman_enforce_max_code_size(num_codes, num_used_syms, code_size_limit); + + DEFL_CLEAR_OBJ(d->m_huff_code_sizes[table_num]); DEFL_CLEAR_OBJ(d->m_huff_codes[table_num]); + for (i = 1, j = num_used_syms; i <= code_size_limit; i++) + for (l = num_codes[i]; l > 0; l--) d->m_huff_code_sizes[table_num][pSyms[--j].m_sym_index] = (uint8_t)(i); + } + + next_code[1] = 0; for (j = 0, i = 2; i <= code_size_limit; i++) next_code[i] = j = ((j + num_codes[i - 1]) << 1); + + for (i = 0; i < table_len; i++) + { + uint32_t rev_code = 0, code, code_size; if ((code_size = d->m_huff_code_sizes[table_num][i]) == 0) continue; + code = next_code[code_size]++; for (l = code_size; l > 0; l--, code >>= 1) rev_code = (rev_code << 1) | (code & 1); + d->m_huff_codes[table_num][i] = (uint16_t)rev_code; + } + } + +#define DEFL_RLE_PREV_CODE_SIZE() { if (rle_repeat_count) { \ + if (rle_repeat_count < 3) { \ + d->m_huff_count[2][prev_code_size] = (uint16_t)(d->m_huff_count[2][prev_code_size] + rle_repeat_count); \ + while (rle_repeat_count--) packed_code_sizes[num_packed_code_sizes++] = prev_code_size; \ + } else { \ + d->m_huff_count[2][16] = (uint16_t)(d->m_huff_count[2][16] + 1); packed_code_sizes[num_packed_code_sizes++] = 16; packed_code_sizes[num_packed_code_sizes++] = (uint8_t)(rle_repeat_count - 3); \ +} rle_repeat_count = 0; } } + +#define DEFL_RLE_ZERO_CODE_SIZE() { if (rle_z_count) { \ + if (rle_z_count < 3) { \ + d->m_huff_count[2][0] = (uint16_t)(d->m_huff_count[2][0] + rle_z_count); while (rle_z_count--) packed_code_sizes[num_packed_code_sizes++] = 0; \ + } else if (rle_z_count <= 10) { \ + d->m_huff_count[2][17] = (uint16_t)(d->m_huff_count[2][17] + 1); packed_code_sizes[num_packed_code_sizes++] = 17; packed_code_sizes[num_packed_code_sizes++] = (uint8_t)(rle_z_count - 3); \ + } else { \ + d->m_huff_count[2][18] = (uint16_t)(d->m_huff_count[2][18] + 1); packed_code_sizes[num_packed_code_sizes++] = 18; packed_code_sizes[num_packed_code_sizes++] = (uint8_t)(rle_z_count - 11); \ +} rle_z_count = 0; } } + + static uint8_t g_defl_packed_code_size_syms_swizzle[] = { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 }; + +#define DEFL_DYN_PUT_BITS(bb, ll) \ +do { \ + uint32_t b = (bb), l = (ll); \ + assert((l) >= 1 && (l) <= 16); assert((b) < (1ULL << (l))); \ + bit_buf |= (((uint64_t)(b)) << bit_buf_size); bit_buf_size += (l); assert(bit_buf_size <= 64); \ + while (bit_buf_size >= 8) \ + { \ + if ((dst_ofs + 1) > dst_buf_size) \ + return false; \ + *(uint8_t*)(pDst + dst_ofs) = (uint8_t)bit_buf; \ + dst_ofs++; \ + bit_buf >>= 8; \ + bit_buf_size -= 8; \ + } \ +} while(0) + + static bool defl_start_dynamic_block(defl_huff* d, uint8_t* pDst, uint32_t& dst_ofs, uint32_t dst_buf_size, uint64_t& bit_buf, int& bit_buf_size) + { + int num_lit_codes, num_dist_codes, num_bit_lengths; uint32_t i, total_code_sizes_to_pack, num_packed_code_sizes, rle_z_count, rle_repeat_count, packed_code_sizes_index; + uint8_t code_sizes_to_pack[DEFL_MAX_HUFF_SYMBOLS_0 + DEFL_MAX_HUFF_SYMBOLS_1], packed_code_sizes[DEFL_MAX_HUFF_SYMBOLS_0 + DEFL_MAX_HUFF_SYMBOLS_1], prev_code_size = 0xFF; + +#if FPNG_TRAIN_HUFFMAN_TABLES + assert(HUFF_COUNTS_SIZE == DEFL_MAX_HUFF_SYMBOLS_0); + for (uint32_t i = 0; i < DEFL_MAX_HUFF_SYMBOLS_0; i++) + g_huff_counts[i] += d->m_huff_count[0][i]; +#endif + + d->m_huff_count[0][256] = 1; + + defl_optimize_huffman_table(d, 0, DEFL_MAX_HUFF_SYMBOLS_0, 12, FPNG_FALSE); + defl_optimize_huffman_table(d, 1, DEFL_MAX_HUFF_SYMBOLS_1, 12, FPNG_FALSE); + + for (num_lit_codes = 286; num_lit_codes > 257; num_lit_codes--) if (d->m_huff_code_sizes[0][num_lit_codes - 1]) break; + for (num_dist_codes = 30; num_dist_codes > 1; num_dist_codes--) if (d->m_huff_code_sizes[1][num_dist_codes - 1]) break; + + memcpy(code_sizes_to_pack, &d->m_huff_code_sizes[0][0], num_lit_codes); + memcpy(code_sizes_to_pack + num_lit_codes, &d->m_huff_code_sizes[1][0], num_dist_codes); + total_code_sizes_to_pack = num_lit_codes + num_dist_codes; num_packed_code_sizes = 0; rle_z_count = 0; rle_repeat_count = 0; + + memset(&d->m_huff_count[2][0], 0, sizeof(d->m_huff_count[2][0]) * DEFL_MAX_HUFF_SYMBOLS_2); + for (i = 0; i < total_code_sizes_to_pack; i++) + { + uint8_t code_size = code_sizes_to_pack[i]; + if (!code_size) + { + DEFL_RLE_PREV_CODE_SIZE(); + if (++rle_z_count == 138) { DEFL_RLE_ZERO_CODE_SIZE(); } + } + else + { + DEFL_RLE_ZERO_CODE_SIZE(); + if (code_size != prev_code_size) + { + DEFL_RLE_PREV_CODE_SIZE(); + d->m_huff_count[2][code_size] = (uint16_t)(d->m_huff_count[2][code_size] + 1); packed_code_sizes[num_packed_code_sizes++] = code_size; + } + else if (++rle_repeat_count == 6) + { + DEFL_RLE_PREV_CODE_SIZE(); + } + } + prev_code_size = code_size; + } + if (rle_repeat_count) { DEFL_RLE_PREV_CODE_SIZE(); } + else { DEFL_RLE_ZERO_CODE_SIZE(); } + + defl_optimize_huffman_table(d, 2, DEFL_MAX_HUFF_SYMBOLS_2, 7, FPNG_FALSE); + + // max of 2+5+5+4+18*3+(288+32)*7=2310 bits + DEFL_DYN_PUT_BITS(2, 2); + + DEFL_DYN_PUT_BITS(num_lit_codes - 257, 5); + DEFL_DYN_PUT_BITS(num_dist_codes - 1, 5); + + for (num_bit_lengths = 18; num_bit_lengths >= 0; num_bit_lengths--) if (d->m_huff_code_sizes[2][g_defl_packed_code_size_syms_swizzle[num_bit_lengths]]) break; + num_bit_lengths = maximum(4, (num_bit_lengths + 1)); DEFL_DYN_PUT_BITS(num_bit_lengths - 4, 4); + for (i = 0; (int)i < num_bit_lengths; i++) DEFL_DYN_PUT_BITS(d->m_huff_code_sizes[2][g_defl_packed_code_size_syms_swizzle[i]], 3); + + for (packed_code_sizes_index = 0; packed_code_sizes_index < num_packed_code_sizes; ) + { + uint32_t code = packed_code_sizes[packed_code_sizes_index++]; assert(code < DEFL_MAX_HUFF_SYMBOLS_2); + DEFL_DYN_PUT_BITS(d->m_huff_codes[2][code], d->m_huff_code_sizes[2][code]); + if (code >= 16) DEFL_DYN_PUT_BITS(packed_code_sizes[packed_code_sizes_index++], "\02\03\07"[code - 16]); + } + + return true; + } + + static uint32_t write_raw_block(const uint8_t* pSrc, uint32_t src_len, uint8_t* pDst, uint32_t dst_buf_size) + { + if (dst_buf_size < 2) + return 0; + + pDst[0] = 0x78; + pDst[1] = 0x01; + + uint32_t dst_ofs = 2; + + uint32_t src_ofs = 0; + while (src_ofs < src_len) + { + const uint32_t src_remaining = src_len - src_ofs; + const uint32_t block_size = minimum(UINT16_MAX, src_remaining); + const bool final_block = (block_size == src_remaining); + + if ((dst_ofs + 5 + block_size) > dst_buf_size) + return 0; + + pDst[dst_ofs + 0] = final_block ? 1 : 0; + + pDst[dst_ofs + 1] = block_size & 0xFF; + pDst[dst_ofs + 2] = (block_size >> 8) & 0xFF; + + pDst[dst_ofs + 3] = (~block_size) & 0xFF; + pDst[dst_ofs + 4] = ((~block_size) >> 8) & 0xFF; + + memcpy(pDst + dst_ofs + 5, pSrc + src_ofs, block_size); + + src_ofs += block_size; + dst_ofs += 5 + block_size; + } + + uint32_t src_adler32 = fpng_adler32(pSrc, src_len, FPNG_ADLER32_INIT); + + for (uint32_t i = 0; i < 4; i++) + { + if (dst_ofs + 1 > dst_buf_size) + return 0; + + pDst[dst_ofs] = (uint8_t)(src_adler32 >> 24); + dst_ofs++; + + src_adler32 <<= 8; + } + + return dst_ofs; + } + + static void adjust_freq32(uint32_t num_freq, uint32_t* pFreq, uint16_t* pFreq16) + { + uint32_t total_freq = 0; + for (uint32_t i = 0; i < num_freq; i++) + total_freq += pFreq[i]; + + if (!total_freq) + { + memset(pFreq16, 0, num_freq * sizeof(uint16_t)); + return; + } + + uint32_t total_freq16 = 0; + for (uint32_t i = 0; i < num_freq; i++) + { + uint64_t f = pFreq[i]; + if (!f) + { + pFreq16[i] = 0; + continue; + } + + pFreq16[i] = (uint16_t)maximum(1, (uint32_t)((f * UINT16_MAX) / total_freq)); + + total_freq16 += pFreq16[i]; + } + + while (total_freq16 > UINT16_MAX) + { + total_freq16 = 0; + for (uint32_t i = 0; i < num_freq; i++) + { + if (pFreq[i]) + { + pFreq[i] = maximum(1, pFreq[i] >> 1); + total_freq16 += pFreq[i]; + } + } + } + } + +#if FPNG_TRAIN_HUFFMAN_TABLES + bool create_dynamic_block_prefix(uint64_t* pFreq, uint32_t num_chans, std::vector& prefix, uint64_t& bit_buf, int &bit_buf_size, uint32_t* pCodes, uint8_t* pCodesizes) + { + assert((num_chans == 3) || (num_chans == 4)); + assert(HUFF_COUNTS_SIZE == DEFL_MAX_HUFF_SYMBOLS_0); // must be equal + + defl_huff dh; + memset(&dh, 0, sizeof(dh)); + + uint32_t lit_freq[DEFL_MAX_HUFF_SYMBOLS_0]; + + uint32_t shift_len = 0; + for (; ; ) + { + uint32_t i; + for (i = 0; i < DEFL_MAX_HUFF_SYMBOLS_0; i++) + { + uint64_t f = pFreq[i]; + if (f) + f = maximum(1U, f >> shift_len); + + if (f > UINT32_MAX) + break; + + lit_freq[i] = (uint32_t)pFreq[i]; + } + + if (i == DEFL_MAX_HUFF_SYMBOLS_0) + break; + + shift_len++; + } + + // Ensure all valid Deflate literal/EOB/length syms are non-zero, so anything can be coded. + for (uint32_t i = 0; i <= 256; i++) + { + if (!lit_freq[i]) + lit_freq[i] = 1; + } + + for (uint32_t len = num_chans; len <= DEFL_MAX_MATCH_LEN; len += num_chans) + { + uint32_t sym = g_defl_len_sym[len - 3]; + if (!lit_freq[sym]) + lit_freq[sym] = 1; + } + + adjust_freq32(DEFL_MAX_HUFF_SYMBOLS_0, lit_freq, &dh.m_huff_count[0][0]); + + const uint32_t dist_sym = g_defl_small_dist_sym[num_chans - 1]; + dh.m_huff_count[1][dist_sym] = 1; + dh.m_huff_count[1][dist_sym + 1] = 1; // to workaround a bug in wuffs decoder + + prefix.resize(4096); + uint8_t* pDst = prefix.data(); + uint32_t dst_buf_size = (uint32_t)prefix.size(); + + uint32_t dst_ofs = 0; + + // zlib header + PUT_BITS(0x78, 8); + PUT_BITS(0x01, 8); + + // write BFINAL bit + PUT_BITS(1, 1); + + if (!defl_start_dynamic_block(&dh, pDst, dst_ofs, dst_buf_size, bit_buf, bit_buf_size)) + return false; + + prefix.resize(dst_ofs); + + for (uint32_t i = 0; i < DEFL_MAX_HUFF_SYMBOLS_0; i++) + { + pCodes[i] = dh.m_huff_codes[0][i]; + pCodesizes[i] = dh.m_huff_code_sizes[0][i]; + } + + return true; + } +#endif + + static uint32_t pixel_deflate_dyn_3_rle( + const uint8_t* pImg, uint32_t w, uint32_t h, + uint8_t* pDst, uint32_t dst_buf_size) + { + const uint32_t bpl = 1 + w * 3; + + uint64_t bit_buf = 0; + int bit_buf_size = 0; + + uint32_t dst_ofs = 0; + + // zlib header + PUT_BITS(0x78, 8); + PUT_BITS(0x01, 8); + + // write BFINAL bit + PUT_BITS(1, 1); + + std::vector codes((w + 1) * h); + uint32_t* pDst_codes = codes.data(); + + uint32_t lit_freq[DEFL_MAX_HUFF_SYMBOLS_0]; + memset(lit_freq, 0, sizeof(lit_freq)); + + const uint8_t* pSrc = pImg; + uint32_t src_ofs = 0; + + uint32_t src_adler32 = fpng_adler32(pImg, bpl * h, FPNG_ADLER32_INIT); + + const uint32_t dist_sym = g_defl_small_dist_sym[3 - 1]; + + for (uint32_t y = 0; y < h; y++) + { + const uint32_t end_src_ofs = src_ofs + bpl; + + const uint32_t filter_lit = pSrc[src_ofs++]; + *pDst_codes++ = 1 | (filter_lit << 8); + lit_freq[filter_lit]++; + + uint32_t prev_lits; + + { + uint32_t lits = READ_RGB_PIXEL(pSrc + src_ofs); + + *pDst_codes++ = lits << 8; + + lit_freq[lits & 0xFF]++; + lit_freq[(lits >> 8) & 0xFF]++; + lit_freq[lits >> 16]++; + + src_ofs += 3; + + prev_lits = lits; + } + + while (src_ofs < end_src_ofs) + { + uint32_t lits = READ_RGB_PIXEL(pSrc + src_ofs); + + if (lits == prev_lits) + { + uint32_t match_len = 3; + uint32_t max_match_len = minimum(255, (int)(end_src_ofs - src_ofs)); + + while (match_len < max_match_len) + { + if (READ_RGB_PIXEL(pSrc + src_ofs + match_len) != lits) + break; + match_len += 3; + } + + *pDst_codes++ = match_len - 1; + + uint32_t adj_match_len = match_len - 3; + + lit_freq[g_defl_len_sym[adj_match_len]]++; + + src_ofs += match_len; + } + else + { + *pDst_codes++ = lits << 8; + + lit_freq[lits & 0xFF]++; + lit_freq[(lits >> 8) & 0xFF]++; + lit_freq[lits >> 16]++; + + prev_lits = lits; + + src_ofs += 3; + } + + } // while (src_ofs < end_src_ofs) + + } // y + + assert(src_ofs == h * bpl); + const uint32_t total_codes = (uint32_t)(pDst_codes - codes.data()); + assert(total_codes <= codes.size()); + + defl_huff dh; + + lit_freq[256] = 1; + + adjust_freq32(DEFL_MAX_HUFF_SYMBOLS_0, lit_freq, &dh.m_huff_count[0][0]); + + memset(&dh.m_huff_count[1][0], 0, sizeof(dh.m_huff_count[1][0]) * DEFL_MAX_HUFF_SYMBOLS_1); + dh.m_huff_count[1][dist_sym] = 1; + dh.m_huff_count[1][dist_sym + 1] = 1; // to workaround a bug in wuffs decoder + + if (!defl_start_dynamic_block(&dh, pDst, dst_ofs, dst_buf_size, bit_buf, bit_buf_size)) + return 0; + + assert(bit_buf_size <= 7); + assert(dh.m_huff_codes[1][dist_sym] == 0 && dh.m_huff_code_sizes[1][dist_sym] == 1); + + for (uint32_t i = 0; i < total_codes; i++) + { + uint32_t c = codes[i]; + + uint32_t c_type = c & 0xFF; + if (c_type == 0) + { + uint32_t lits = c >> 8; + + PUT_BITS_CZ(dh.m_huff_codes[0][lits & 0xFF], dh.m_huff_code_sizes[0][lits & 0xFF]); + lits >>= 8; + + PUT_BITS_CZ(dh.m_huff_codes[0][lits & 0xFF], dh.m_huff_code_sizes[0][lits & 0xFF]); + lits >>= 8; + + PUT_BITS_CZ(dh.m_huff_codes[0][lits], dh.m_huff_code_sizes[0][lits]); + } + else if (c_type == 1) + { + uint32_t lit = c >> 8; + PUT_BITS_CZ(dh.m_huff_codes[0][lit], dh.m_huff_code_sizes[0][lit]); + } + else + { + uint32_t match_len = c_type + 1; + + uint32_t adj_match_len = match_len - 3; + + PUT_BITS_CZ(dh.m_huff_codes[0][g_defl_len_sym[adj_match_len]], dh.m_huff_code_sizes[0][g_defl_len_sym[adj_match_len]]); + PUT_BITS(adj_match_len & g_bitmasks[g_defl_len_extra[adj_match_len]], g_defl_len_extra[adj_match_len] + 1); // up to 6 bits, +1 for the match distance Huff code which is always 0 + + // no need to write the distance code, it's always 0 + //PUT_BITS_CZ(dh.m_huff_codes[1][dist_sym], dh.m_huff_code_sizes[1][dist_sym]); + } + + // up to 55 bits + PUT_BITS_FLUSH; + } + + PUT_BITS_CZ(dh.m_huff_codes[0][256], dh.m_huff_code_sizes[0][256]); + + PUT_BITS_FORCE_FLUSH; + + // Write zlib adler32 + for (uint32_t i = 0; i < 4; i++) + { + if ((dst_ofs + 1) > dst_buf_size) + return 0; + *(uint8_t*)(pDst + dst_ofs) = (uint8_t)(src_adler32 >> 24); + dst_ofs++; + + src_adler32 <<= 8; + } + + return dst_ofs; + } + + static uint32_t pixel_deflate_dyn_3_rle_one_pass( + const uint8_t* pImg, uint32_t w, uint32_t h, + uint8_t* pDst, uint32_t dst_buf_size) + { + const uint32_t bpl = 1 + w * 3; + + if (dst_buf_size < sizeof(g_dyn_huff_3)) + return false; + memcpy(pDst, g_dyn_huff_3, sizeof(g_dyn_huff_3)); + uint32_t dst_ofs = sizeof(g_dyn_huff_3); + + uint64_t bit_buf = DYN_HUFF_3_BITBUF; + int bit_buf_size = DYN_HUFF_3_BITBUF_SIZE; + + const uint8_t* pSrc = pImg; + uint32_t src_ofs = 0; + + uint32_t src_adler32 = fpng_adler32(pImg, bpl * h, FPNG_ADLER32_INIT); + + for (uint32_t y = 0; y < h; y++) + { + const uint32_t end_src_ofs = src_ofs + bpl; + + const uint32_t filter_lit = pSrc[src_ofs++]; + PUT_BITS_CZ(g_dyn_huff_3_codes[filter_lit].m_code, g_dyn_huff_3_codes[filter_lit].m_code_size); + + uint32_t prev_lits; + + { + uint32_t lits = READ_RGB_PIXEL(pSrc + src_ofs); + + PUT_BITS_CZ(g_dyn_huff_3_codes[lits & 0xFF].m_code, g_dyn_huff_3_codes[lits & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_3_codes[(lits >> 8) & 0xFF].m_code, g_dyn_huff_3_codes[(lits >> 8) & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_3_codes[(lits >> 16)].m_code, g_dyn_huff_3_codes[(lits >> 16)].m_code_size); + + src_ofs += 3; + + prev_lits = lits; + } + + PUT_BITS_FLUSH; + + while (src_ofs < end_src_ofs) + { + uint32_t lits = READ_RGB_PIXEL(pSrc + src_ofs); + + if (lits == prev_lits) + { + uint32_t match_len = 3; + uint32_t max_match_len = minimum(255, (int)(end_src_ofs - src_ofs)); + + while (match_len < max_match_len) + { + if (READ_RGB_PIXEL(pSrc + src_ofs + match_len) != lits) + break; + match_len += 3; + } + + uint32_t adj_match_len = match_len - 3; + + PUT_BITS_CZ(g_dyn_huff_3_codes[g_defl_len_sym[adj_match_len]].m_code, g_dyn_huff_3_codes[g_defl_len_sym[adj_match_len]].m_code_size); + PUT_BITS(adj_match_len & g_bitmasks[g_defl_len_extra[adj_match_len]], g_defl_len_extra[adj_match_len] + 1); // up to 6 bits, +1 for the match distance Huff code which is always 0 + + src_ofs += match_len; + } + else + { + PUT_BITS_CZ(g_dyn_huff_3_codes[lits & 0xFF].m_code, g_dyn_huff_3_codes[lits & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_3_codes[(lits >> 8) & 0xFF].m_code, g_dyn_huff_3_codes[(lits >> 8) & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_3_codes[(lits >> 16)].m_code, g_dyn_huff_3_codes[(lits >> 16)].m_code_size); + + prev_lits = lits; + + src_ofs += 3; + } + + PUT_BITS_FLUSH; + + } // while (src_ofs < end_src_ofs) + + } // y + + assert(src_ofs == h * bpl); + + assert(bit_buf_size <= 7); + + PUT_BITS_CZ(g_dyn_huff_3_codes[256].m_code, g_dyn_huff_3_codes[256].m_code_size); + + PUT_BITS_FORCE_FLUSH; + + // Write zlib adler32 + for (uint32_t i = 0; i < 4; i++) + { + if ((dst_ofs + 1) > dst_buf_size) + return 0; + *(uint8_t*)(pDst + dst_ofs) = (uint8_t)(src_adler32 >> 24); + dst_ofs++; + + src_adler32 <<= 8; + } + + return dst_ofs; + } + + static uint32_t pixel_deflate_dyn_4_rle( + const uint8_t* pImg, uint32_t w, uint32_t h, + uint8_t* pDst, uint32_t dst_buf_size) + { + const uint32_t bpl = 1 + w * 4; + + uint64_t bit_buf = 0; + int bit_buf_size = 0; + + uint32_t dst_ofs = 0; + + // zlib header + PUT_BITS(0x78, 8); + PUT_BITS(0x01, 8); + + // write BFINAL bit + PUT_BITS(1, 1); + + std::vector codes; + codes.resize((w + 1) * h); + uint64_t* pDst_codes = codes.data(); + + uint32_t lit_freq[DEFL_MAX_HUFF_SYMBOLS_0]; + memset(lit_freq, 0, sizeof(lit_freq)); + + const uint8_t* pSrc = pImg; + uint32_t src_ofs = 0; + + uint32_t src_adler32 = fpng_adler32(pImg, bpl * h, FPNG_ADLER32_INIT); + + const uint32_t dist_sym = g_defl_small_dist_sym[4 - 1]; + + for (uint32_t y = 0; y < h; y++) + { + const uint32_t end_src_ofs = src_ofs + bpl; + + const uint32_t filter_lit = pSrc[src_ofs++]; + *pDst_codes++ = 1 | (filter_lit << 8); + lit_freq[filter_lit]++; + + uint32_t prev_lits; + { + uint32_t lits = READ_LE32(pSrc + src_ofs); + + *pDst_codes++ = (uint64_t)lits << 8; + + lit_freq[lits & 0xFF]++; + lit_freq[(lits >> 8) & 0xFF]++; + lit_freq[(lits >> 16) & 0xFF]++; + lit_freq[lits >> 24]++; + + src_ofs += 4; + + prev_lits = lits; + } + + while (src_ofs < end_src_ofs) + { + uint32_t lits = READ_LE32(pSrc + src_ofs); + + if (lits == prev_lits) + { + uint32_t match_len = 4; + uint32_t max_match_len = minimum(252, (int)(end_src_ofs - src_ofs)); + + while (match_len < max_match_len) + { + if (READ_LE32(pSrc + src_ofs + match_len) != lits) + break; + match_len += 4; + } + + *pDst_codes++ = match_len - 1; + + uint32_t adj_match_len = match_len - 3; + + lit_freq[g_defl_len_sym[adj_match_len]]++; + + src_ofs += match_len; + } + else + { + *pDst_codes++ = (uint64_t)lits << 8; + + lit_freq[lits & 0xFF]++; + lit_freq[(lits >> 8) & 0xFF]++; + lit_freq[(lits >> 16) & 0xFF]++; + lit_freq[lits >> 24]++; + + prev_lits = lits; + + src_ofs += 4; + } + + } // while (src_ofs < end_src_ofs) + + } // y + + assert(src_ofs == h * bpl); + const uint32_t total_codes = (uint32_t)(pDst_codes - codes.data()); + assert(total_codes <= codes.size()); + + defl_huff dh; + + lit_freq[256] = 1; + + adjust_freq32(DEFL_MAX_HUFF_SYMBOLS_0, lit_freq, &dh.m_huff_count[0][0]); + + memset(&dh.m_huff_count[1][0], 0, sizeof(dh.m_huff_count[1][0]) * DEFL_MAX_HUFF_SYMBOLS_1); + dh.m_huff_count[1][dist_sym] = 1; + dh.m_huff_count[1][dist_sym + 1] = 1; // to workaround a bug in wuffs decoder + + if (!defl_start_dynamic_block(&dh, pDst, dst_ofs, dst_buf_size, bit_buf, bit_buf_size)) + return 0; + + assert(bit_buf_size <= 7); + assert(dh.m_huff_codes[1][dist_sym] == 0 && dh.m_huff_code_sizes[1][dist_sym] == 1); + + for (uint32_t i = 0; i < total_codes; i++) + { + uint64_t c = codes[i]; + + uint32_t c_type = (uint32_t)(c & 0xFF); + if (c_type == 0) + { + uint32_t lits = (uint32_t)(c >> 8); + + PUT_BITS_CZ(dh.m_huff_codes[0][lits & 0xFF], dh.m_huff_code_sizes[0][lits & 0xFF]); + lits >>= 8; + + PUT_BITS_CZ(dh.m_huff_codes[0][lits & 0xFF], dh.m_huff_code_sizes[0][lits & 0xFF]); + lits >>= 8; + + PUT_BITS_CZ(dh.m_huff_codes[0][lits & 0xFF], dh.m_huff_code_sizes[0][lits & 0xFF]); + lits >>= 8; + + if (bit_buf_size >= 49) + { + PUT_BITS_FLUSH; + } + + PUT_BITS_CZ(dh.m_huff_codes[0][lits], dh.m_huff_code_sizes[0][lits]); + } + else if (c_type == 1) + { + uint32_t lit = (uint32_t)(c >> 8); + PUT_BITS_CZ(dh.m_huff_codes[0][lit], dh.m_huff_code_sizes[0][lit]); + } + else + { + uint32_t match_len = c_type + 1; + + uint32_t adj_match_len = match_len - 3; + + PUT_BITS_CZ(dh.m_huff_codes[0][g_defl_len_sym[adj_match_len]], dh.m_huff_code_sizes[0][g_defl_len_sym[adj_match_len]]); + PUT_BITS(adj_match_len & g_bitmasks[g_defl_len_extra[adj_match_len]], g_defl_len_extra[adj_match_len] + 1); // up to 6 bits, +1 for the match distance Huff code which is always 0 + + // no need to write the distance code, it's always 0 + } + + // up to 55 bits + PUT_BITS_FLUSH; + } + + PUT_BITS_CZ(dh.m_huff_codes[0][256], dh.m_huff_code_sizes[0][256]); + + PUT_BITS_FORCE_FLUSH; + + // Write zlib adler32 + for (uint32_t i = 0; i < 4; i++) + { + if ((dst_ofs + 1) > dst_buf_size) + return 0; + *(uint8_t*)(pDst + dst_ofs) = (uint8_t)(src_adler32 >> 24); + dst_ofs++; + + src_adler32 <<= 8; + } + + return dst_ofs; + } + + static uint32_t pixel_deflate_dyn_4_rle_one_pass( + const uint8_t* pImg, uint32_t w, uint32_t h, + uint8_t* pDst, uint32_t dst_buf_size) + { + const uint32_t bpl = 1 + w * 4; + + if (dst_buf_size < sizeof(g_dyn_huff_4)) + return false; + memcpy(pDst, g_dyn_huff_4, sizeof(g_dyn_huff_4)); + uint32_t dst_ofs = sizeof(g_dyn_huff_4); + + uint64_t bit_buf = DYN_HUFF_4_BITBUF; + int bit_buf_size = DYN_HUFF_4_BITBUF_SIZE; + + const uint8_t* pSrc = pImg; + uint32_t src_ofs = 0; + + uint32_t src_adler32 = fpng_adler32(pImg, bpl * h, FPNG_ADLER32_INIT); + + for (uint32_t y = 0; y < h; y++) + { + const uint32_t end_src_ofs = src_ofs + bpl; + + const uint32_t filter_lit = pSrc[src_ofs++]; + PUT_BITS_CZ(g_dyn_huff_4_codes[filter_lit].m_code, g_dyn_huff_4_codes[filter_lit].m_code_size); + + PUT_BITS_FLUSH; + + uint32_t prev_lits; + { + uint32_t lits = READ_LE32(pSrc + src_ofs); + + PUT_BITS_CZ(g_dyn_huff_4_codes[lits & 0xFF].m_code, g_dyn_huff_4_codes[lits & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_4_codes[(lits >> 8) & 0xFF].m_code, g_dyn_huff_4_codes[(lits >> 8) & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_4_codes[(lits >> 16) & 0xFF].m_code, g_dyn_huff_4_codes[(lits >> 16) & 0xFF].m_code_size); + + if (bit_buf_size >= 49) + { + PUT_BITS_FLUSH; + } + + PUT_BITS_CZ(g_dyn_huff_4_codes[(lits >> 24)].m_code, g_dyn_huff_4_codes[(lits >> 24)].m_code_size); + + src_ofs += 4; + + prev_lits = lits; + } + + PUT_BITS_FLUSH; + + while (src_ofs < end_src_ofs) + { + uint32_t lits = READ_LE32(pSrc + src_ofs); + + if (lits == prev_lits) + { + uint32_t match_len = 4; + uint32_t max_match_len = minimum(252, (int)(end_src_ofs - src_ofs)); + + while (match_len < max_match_len) + { + if (READ_LE32(pSrc + src_ofs + match_len) != lits) + break; + match_len += 4; + } + + uint32_t adj_match_len = match_len - 3; + + const uint32_t match_code_bits = g_dyn_huff_4_codes[g_defl_len_sym[adj_match_len]].m_code_size; + const uint32_t len_extra_bits = g_defl_len_extra[adj_match_len]; + + if (match_len == 4) + { + // This check is optional - see if just encoding 4 literals would be cheaper than using a short match. + uint32_t lit_bits = g_dyn_huff_4_codes[lits & 0xFF].m_code_size + g_dyn_huff_4_codes[(lits >> 8) & 0xFF].m_code_size + + g_dyn_huff_4_codes[(lits >> 16) & 0xFF].m_code_size + g_dyn_huff_4_codes[(lits >> 24)].m_code_size; + + if ((match_code_bits + len_extra_bits + 1) > lit_bits) + goto do_literals; + } + + PUT_BITS_CZ(g_dyn_huff_4_codes[g_defl_len_sym[adj_match_len]].m_code, match_code_bits); + PUT_BITS(adj_match_len & g_bitmasks[g_defl_len_extra[adj_match_len]], len_extra_bits + 1); // up to 6 bits, +1 for the match distance Huff code which is always 0 + + src_ofs += match_len; + } + else + { +do_literals: + PUT_BITS_CZ(g_dyn_huff_4_codes[lits & 0xFF].m_code, g_dyn_huff_4_codes[lits & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_4_codes[(lits >> 8) & 0xFF].m_code, g_dyn_huff_4_codes[(lits >> 8) & 0xFF].m_code_size); + PUT_BITS_CZ(g_dyn_huff_4_codes[(lits >> 16) & 0xFF].m_code, g_dyn_huff_4_codes[(lits >> 16) & 0xFF].m_code_size); + + if (bit_buf_size >= 49) + { + PUT_BITS_FLUSH; + } + + PUT_BITS_CZ(g_dyn_huff_4_codes[(lits >> 24)].m_code, g_dyn_huff_4_codes[(lits >> 24)].m_code_size); + + src_ofs += 4; + + prev_lits = lits; + } + + PUT_BITS_FLUSH; + + } // while (src_ofs < end_src_ofs) + + } // y + + assert(src_ofs == h * bpl); + + assert(bit_buf_size <= 7); + + PUT_BITS_CZ(g_dyn_huff_4_codes[256].m_code, g_dyn_huff_4_codes[256].m_code_size); + + PUT_BITS_FORCE_FLUSH; + + // Write zlib adler32 + for (uint32_t i = 0; i < 4; i++) + { + if ((dst_ofs + 1) > dst_buf_size) + return 0; + *(uint8_t*)(pDst + dst_ofs) = (uint8_t)(src_adler32 >> 24); + dst_ofs++; + + src_adler32 <<= 8; + } + + return dst_ofs; + } + + static void vector_append(std::vector& buf, const void* pData, size_t len) + { + if (len) + { + size_t l = buf.size(); + buf.resize(l + len); + memcpy(buf.data() + l, pData, len); + } + } + + static void apply_filter(uint32_t filter, int w, int h, uint32_t num_chans, uint32_t bpl, const uint8_t* pSrc, const uint8_t* pPrev_src, uint8_t* pDst) + { + (void)h; + + switch (filter) + { + case 0: + { + *pDst++ = 0; + + memcpy(pDst, pSrc, bpl); + break; + } + case 2: + { + assert(pPrev_src); + + // Previous scanline + *pDst++ = 2; + +#if FPNG_X86_OR_X64_CPU && !FPNG_NO_SSE + if (g_cpu_info.can_use_sse41()) + { + uint32_t bytes_to_process = w * num_chans, ofs = 0; + for (; bytes_to_process >= 16; bytes_to_process -= 16, ofs += 16) + _mm_storeu_si128((__m128i*)(pDst + ofs), _mm_sub_epi8(_mm_loadu_si128((const __m128i*)(pSrc + ofs)), _mm_loadu_si128((const __m128i*)(pPrev_src + ofs)))); + + for (; bytes_to_process; bytes_to_process--, ofs++) + pDst[ofs] = (uint8_t)(pSrc[ofs] - pPrev_src[ofs]); + } + else +#endif + { + if (num_chans == 3) + { + for (uint32_t x = 0; x < (uint32_t)w; x++) + { + pDst[0] = (uint8_t)(pSrc[0] - pPrev_src[0]); + pDst[1] = (uint8_t)(pSrc[1] - pPrev_src[1]); + pDst[2] = (uint8_t)(pSrc[2] - pPrev_src[2]); + + pSrc += 3; + pPrev_src += 3; + pDst += 3; + } + } + else + { + for (uint32_t x = 0; x < (uint32_t)w; x++) + { + pDst[0] = (uint8_t)(pSrc[0] - pPrev_src[0]); + pDst[1] = (uint8_t)(pSrc[1] - pPrev_src[1]); + pDst[2] = (uint8_t)(pSrc[2] - pPrev_src[2]); + pDst[3] = (uint8_t)(pSrc[3] - pPrev_src[3]); + + pSrc += 4; + pPrev_src += 4; + pDst += 4; + } + } + } + + break; + } + default: + assert(0); + break; + } + } + + bool fpng_encode_image_to_memory(const void* pImage, uint32_t w, uint32_t h, uint32_t num_chans, std::vector& out_buf, uint32_t flags) + { + if (!endian_check()) + { + assert(0); + return false; + } + + if ((w < 1) || (h < 1) || (w * (uint64_t)h > UINT32_MAX) || (w > FPNG_MAX_SUPPORTED_DIM) || (h > FPNG_MAX_SUPPORTED_DIM)) + { + assert(0); + return false; + } + + if ((num_chans != 3) && (num_chans != 4)) + { + assert(0); + return false; + } + + int i, bpl = w * num_chans; + uint32_t y; + + std::vector temp_buf; + temp_buf.resize((bpl + 1) * h + 7); + uint32_t temp_buf_ofs = 0; + + for (y = 0; y < h; ++y) + { + const uint8_t* pSrc = (uint8_t*)pImage + y * bpl; + const uint8_t* pPrev_src = y ? ((uint8_t*)pImage + (y - 1) * bpl) : nullptr; + + uint8_t* pDst = &temp_buf[temp_buf_ofs]; + + apply_filter(y ? 2 : 0, w, h, num_chans, bpl, pSrc, pPrev_src, pDst); + + temp_buf_ofs += 1 + bpl; + } + + const uint32_t PNG_HEADER_SIZE = 58; + + uint32_t out_ofs = PNG_HEADER_SIZE; + + out_buf.resize((out_ofs + (bpl + 1) * h + 7) & ~7); + + uint32_t defl_size = 0; + if ((flags & FPNG_FORCE_UNCOMPRESSED) == 0) + { + if (num_chans == 3) + { + if (flags & FPNG_ENCODE_SLOWER) + defl_size = pixel_deflate_dyn_3_rle(temp_buf.data(), w, h, &out_buf[out_ofs], (uint32_t)out_buf.size() - out_ofs); + else + defl_size = pixel_deflate_dyn_3_rle_one_pass(temp_buf.data(), w, h, &out_buf[out_ofs], (uint32_t)out_buf.size() - out_ofs); + } + else + { + if (flags & FPNG_ENCODE_SLOWER) + defl_size = pixel_deflate_dyn_4_rle(temp_buf.data(), w, h, &out_buf[out_ofs], (uint32_t)out_buf.size() - out_ofs); + else + defl_size = pixel_deflate_dyn_4_rle_one_pass(temp_buf.data(), w, h, &out_buf[out_ofs], (uint32_t)out_buf.size() - out_ofs); + } + } + + uint32_t zlib_size = defl_size; + + if (!defl_size) + { + // Dynamic block failed to compress - fall back to uncompressed blocks, filter 0. + + temp_buf_ofs = 0; + + for (y = 0; y < h; ++y) + { + const uint8_t* pSrc = (uint8_t*)pImage + y * bpl; + + uint8_t* pDst = &temp_buf[temp_buf_ofs]; + + apply_filter(0, w, h, num_chans, bpl, pSrc, nullptr, pDst); + + temp_buf_ofs += 1 + bpl; + } + + assert(temp_buf_ofs <= temp_buf.size()); + + out_buf.resize(out_ofs + 6 + temp_buf_ofs + ((temp_buf_ofs + 65534) / 65535) * 5); + + uint32_t raw_size = write_raw_block(temp_buf.data(), (uint32_t)temp_buf_ofs, out_buf.data() + out_ofs, (uint32_t)out_buf.size() - out_ofs); + if (!raw_size) + { + // Somehow we miscomputed the size of the output buffer. + assert(0); + return false; + } + + zlib_size = raw_size; + } + + assert((out_ofs + zlib_size) <= out_buf.size()); + + out_buf.resize(out_ofs + zlib_size); + + const uint32_t idat_len = (uint32_t)out_buf.size() - PNG_HEADER_SIZE; + + // Write real PNG header, fdEC chunk, and the beginning of the IDAT chunk + { + static const uint8_t s_color_type[] = { 0x00, 0x00, 0x04, 0x02, 0x06 }; + + uint8_t pnghdr[58] = { + 0x89,0x50,0x4e,0x47,0x0d,0x0a,0x1a,0x0a, // PNG sig + 0x00,0x00,0x00,0x0d, 'I','H','D','R', // IHDR chunk len, type + 0,0,(uint8_t)(w >> 8),(uint8_t)w, // width + 0,0,(uint8_t)(h >> 8),(uint8_t)h, // height + 8, //bit_depth + s_color_type[num_chans], // color_type + 0, // compression + 0, // filter + 0, // interlace + 0, 0, 0, 0, // IHDR crc32 + 0, 0, 0, 5, 'f', 'd', 'E', 'C', 82, 36, 147, 227, FPNG_FDEC_VERSION, 0xE5, 0xAB, 0x62, 0x99, // our custom private, ancillary, do not copy, fdEC chunk + (uint8_t)(idat_len >> 24),(uint8_t)(idat_len >> 16),(uint8_t)(idat_len >> 8),(uint8_t)idat_len, 'I','D','A','T' // IDATA chunk len, type + }; + + // Compute IHDR CRC32 + uint32_t c = (uint32_t)fpng_crc32(pnghdr + 12, 17, FPNG_CRC32_INIT); + for (i = 0; i < 4; ++i, c <<= 8) + ((uint8_t*)(pnghdr + 29))[i] = (uint8_t)(c >> 24); + + memcpy(out_buf.data(), pnghdr, PNG_HEADER_SIZE); + } + + // Write IDAT chunk's CRC32 and a 0 length IEND chunk + vector_append(out_buf, "\0\0\0\0\0\0\0\0\x49\x45\x4e\x44\xae\x42\x60\x82", 16); // IDAT CRC32, followed by the IEND chunk + + // Compute IDAT crc32 + uint32_t c = (uint32_t)fpng_crc32(out_buf.data() + PNG_HEADER_SIZE - 4, idat_len + 4, FPNG_CRC32_INIT); + + for (i = 0; i < 4; ++i, c <<= 8) + (out_buf.data() + out_buf.size() - 16)[i] = (uint8_t)(c >> 24); + + return true; + } + +#ifndef FPNG_NO_STDIO + bool fpng_encode_image_to_file(const char* pFilename, const void* pImage, uint32_t w, uint32_t h, uint32_t num_chans, uint32_t flags) + { + std::vector out_buf; + if (!fpng_encode_image_to_memory(pImage, w, h, num_chans, out_buf, flags)) + return false; + + FILE* pFile = nullptr; +#ifdef _MSC_VER + fopen_s(&pFile, pFilename, "wb"); +#else + pFile = fopen(pFilename, "wb"); +#endif + if (!pFile) + return false; + + if (fwrite(out_buf.data(), 1, out_buf.size(), pFile) != out_buf.size()) + { + fclose(pFile); + return false; + } + + return (fclose(pFile) != EOF); + } +#endif + + // Decompression + + const uint32_t FPNG_DECODER_TABLE_BITS = 12; + const uint32_t FPNG_DECODER_TABLE_SIZE = 1 << FPNG_DECODER_TABLE_BITS; + + static bool build_decoder_table(uint32_t num_syms, uint8_t* pCode_sizes, uint32_t* pTable) + { + uint32_t num_codes[16]; + + memset(num_codes, 0, sizeof(num_codes)); + for (uint32_t i = 0; i < num_syms; i++) + { + assert(pCode_sizes[i] <= FPNG_DECODER_TABLE_SIZE); + num_codes[pCode_sizes[i]]++; + } + + uint32_t next_code[17]; + next_code[0] = next_code[1] = 0; + uint32_t total = 0; + for (uint32_t i = 1; i <= 15; i++) + next_code[i + 1] = (uint32_t)(total = ((total + ((uint32_t)num_codes[i])) << 1)); + + if (total != 0x10000) + { + uint32_t j = 0; + + for (uint32_t i = 15; i != 0; i--) + if ((j += num_codes[i]) > 1) + return false; + + if (j != 1) + return false; + } + + uint32_t rev_codes[DEFL_MAX_HUFF_SYMBOLS]; + + for (uint32_t i = 0; i < num_syms; i++) + rev_codes[i] = next_code[pCode_sizes[i]]++; + + memset(pTable, 0, sizeof(uint32_t) * FPNG_DECODER_TABLE_SIZE); + + for (uint32_t i = 0; i < num_syms; i++) + { + const uint32_t code_size = pCode_sizes[i]; + if (!code_size) + continue; + + uint32_t old_code = rev_codes[i], new_code = 0; + for (uint32_t j = code_size; j != 0; j--) + { + new_code = (new_code << 1) | (old_code & 1); + old_code >>= 1; + } + + uint32_t j = 1 << code_size; + + while (new_code < FPNG_DECODER_TABLE_SIZE) + { + pTable[new_code] = i | (code_size << 9); + new_code += j; + } + } + + return true; + } + + static const uint16_t g_run_len3_to_4[259] = + { + 0, + 0, 0, 4, 0, 0, 8, 0, 0, 12, 0, 0, 16, 0, 0, 20, 0, 0, 24, 0, 0, 28, 0, 0, + 32, 0, 0, 36, 0, 0, 40, 0, 0, 44, 0, 0, 48, 0, 0, 52, 0, 0, 56, 0, 0, + 60, 0, 0, 64, 0, 0, 68, 0, 0, 72, 0, 0, 76, 0, 0, 80, 0, 0, 84, 0, 0, + 88, 0, 0, 92, 0, 0, 96, 0, 0, 100, 0, 0, 104, 0, 0, 108, 0, 0, 112, 0, 0, + 116, 0, 0, 120, 0, 0, 124, 0, 0, 128, 0, 0, 132, 0, 0, 136, 0, 0, 140, 0, 0, + 144, 0, 0, 148, 0, 0, 152, 0, 0, 156, 0, 0, 160, 0, 0, 164, 0, 0, 168, 0, 0, + 172, 0, 0, 176, 0, 0, 180, 0, 0, 184, 0, 0, 188, 0, 0, 192, 0, 0, 196, 0, 0, + 200, 0, 0, 204, 0, 0, 208, 0, 0, 212, 0, 0, 216, 0, 0, 220, 0, 0, 224, 0, 0, + 228, 0, 0, 232, 0, 0, 236, 0, 0, 240, 0, 0, 244, 0, 0, 248, 0, 0, 252, 0, 0, + 256, 0, 0, 260, 0, 0, 264, 0, 0, 268, 0, 0, 272, 0, 0, 276, 0, 0, 280, 0, 0, + 284, 0, 0, 288, 0, 0, 292, 0, 0, 296, 0, 0, 300, 0, 0, 304, 0, 0, 308, 0, 0, + 312, 0, 0, 316, 0, 0, 320, 0, 0, 324, 0, 0, 328, 0, 0, 332, 0, 0, 336, 0, 0, + 340, 0, 0, + 344, + }; + + static const int s_length_extra[] = { 0,0,0,0, 0,0,0,0, 1,1,1,1, 2,2,2,2, 3,3,3,3, 4,4,4,4, 5,5,5,5, 0, 0,0 }; + static const int s_length_range[] = { 3,4,5,6, 7,8,9,10, 11,13,15,17, 19,23,27,31, 35,43,51,59, 67,83,99,115, 131,163,195,227, 258, 0,0 }; + +#define ENSURE_32BITS() do { \ + if (bit_buf_size < 32) { \ + if ((src_ofs + 4) > src_len) return false; \ + bit_buf |= ((uint64_t)READ_LE32(pSrc + src_ofs)) << bit_buf_size; \ + src_ofs += 4; bit_buf_size += 32; } \ + } while(0) + +#define GET_BITS(b, ll) do { \ + uint32_t l = ll; assert(l && (l <= 32)); \ + b = (uint32_t)(bit_buf & g_bitmasks[l]); \ + bit_buf >>= l; \ + bit_buf_size -= l; \ + ENSURE_32BITS(); \ + } while(0) + +#define SKIP_BITS(ll) do { \ + uint32_t l = ll; assert(l <= 32); \ + bit_buf >>= l; \ + bit_buf_size -= l; \ + ENSURE_32BITS(); \ + } while(0) + +#define GET_BITS_NE(b, ll) do { \ + uint32_t l = ll; assert(l && (l <= 32) && (bit_buf_size >= l)); \ + b = (uint32_t)(bit_buf & g_bitmasks[l]); \ + bit_buf >>= l; \ + bit_buf_size -= l; \ + } while(0) + +#define SKIP_BITS_NE(ll) do { \ + uint32_t l = ll; assert(l <= 32 && (bit_buf_size >= l)); \ + bit_buf >>= l; \ + bit_buf_size -= l; \ + } while(0) + + static bool prepare_dynamic_block( + const uint8_t* pSrc, uint32_t src_len, uint32_t& src_ofs, + uint32_t& bit_buf_size, uint64_t& bit_buf, + uint32_t* pLit_table, uint32_t num_chans) + { + static const uint8_t s_bit_length_order[] = { 16, 17, 18, 0, 8, 7, 9, 6, 10, 5, 11, 4, 12, 3, 13, 2, 14, 1, 15 }; + + uint32_t num_lit_codes, num_dist_codes, num_clen_codes; + + GET_BITS(num_lit_codes, 5); + num_lit_codes += 257; + + GET_BITS(num_dist_codes, 5); + num_dist_codes += 1; + + uint32_t total_codes = num_lit_codes + num_dist_codes; + if (total_codes > (DEFL_MAX_HUFF_SYMBOLS_0 + DEFL_MAX_HUFF_SYMBOLS_1)) + return false; + + uint8_t code_sizes[DEFL_MAX_HUFF_SYMBOLS_0 + DEFL_MAX_HUFF_SYMBOLS_1]; + memset(code_sizes, 0, sizeof(code_sizes)); + + GET_BITS(num_clen_codes, 4); + num_clen_codes += 4; + + uint8_t clen_codesizes[DEFL_MAX_HUFF_SYMBOLS_2]; + memset(clen_codesizes, 0, sizeof(clen_codesizes)); + + for (uint32_t i = 0; i < num_clen_codes; i++) + { + uint32_t len = 0; + GET_BITS(len, 3); + clen_codesizes[s_bit_length_order[i]] = (uint8_t)len; + } + + uint32_t clen_table[FPNG_DECODER_TABLE_SIZE]; + if (!build_decoder_table(DEFL_MAX_HUFF_SYMBOLS_2, clen_codesizes, clen_table)) + return false; + + uint32_t min_code_size = 15; + + for (uint32_t cur_code = 0; cur_code < total_codes; ) + { + uint32_t sym = clen_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t sym_len = sym >> 9; + if (!sym_len) + return false; + SKIP_BITS(sym_len); + sym &= 511; + + if (sym <= 15) + { + // Can't be a fpng Huffman table + if (sym > FPNG_DECODER_TABLE_BITS) + return false; + + if (sym) + min_code_size = minimum(min_code_size, sym); + + code_sizes[cur_code++] = (uint8_t)sym; + continue; + } + + uint32_t rep_len = 0, rep_code_size = 0; + + switch (sym) + { + case 16: + { + GET_BITS(rep_len, 2); + rep_len += 3; + if (!cur_code) + return false; + rep_code_size = code_sizes[cur_code - 1]; + break; + } + case 17: + { + GET_BITS(rep_len, 3); + rep_len += 3; + rep_code_size = 0; + break; + } + case 18: + { + GET_BITS(rep_len, 7); + rep_len += 11; + rep_code_size = 0; + break; + } + } + + if ((cur_code + rep_len) > total_codes) + return false; + + for (; rep_len; rep_len--) + code_sizes[cur_code++] = (uint8_t)rep_code_size; + } + + uint8_t lit_codesizes[DEFL_MAX_HUFF_SYMBOLS_0]; + + memcpy(lit_codesizes, code_sizes, num_lit_codes); + memset(lit_codesizes + num_lit_codes, 0, DEFL_MAX_HUFF_SYMBOLS_0 - num_lit_codes); + + uint32_t total_valid_distcodes = 0; + for (uint32_t i = 0; i < num_dist_codes; i++) + total_valid_distcodes += (code_sizes[num_lit_codes + i] == 1); + + // 1 or 2 because the first version of FPNG only issued 1 valid distance code, but that upset wuffs. So we let 1 or 2 through. + if ((total_valid_distcodes < 1) || (total_valid_distcodes > 2)) + return false; + + if (code_sizes[num_lit_codes + (num_chans - 1)] != 1) + return false; + + if (total_valid_distcodes == 2) + { + // If there are two valid distance codes, make sure the first is 1 bit. + if (code_sizes[num_lit_codes + num_chans] != 1) + return false; + } + + if (!build_decoder_table(num_lit_codes, lit_codesizes, pLit_table)) + return false; + + // Add next symbol to decoder table, when it fits + for (uint32_t i = 0; i < FPNG_DECODER_TABLE_SIZE; i++) + { + uint32_t sym = pLit_table[i] & 511; + if (sym >= 256) + continue; + + uint32_t sym_bits = (pLit_table[i] >> 9) & 15; + if (!sym_bits) + continue; + assert(sym_bits <= FPNG_DECODER_TABLE_BITS); + + uint32_t bits_left = FPNG_DECODER_TABLE_BITS - sym_bits; + if (bits_left < min_code_size) + continue; + + uint32_t next_bits = i >> sym_bits; + uint32_t next_sym = pLit_table[next_bits] & 511; + uint32_t next_sym_bits = (pLit_table[next_bits] >> 9) & 15; + if ((!next_sym_bits) || (bits_left < next_sym_bits)) + continue; + + pLit_table[i] |= (next_sym << 16) | (next_sym_bits << (16 + 9)); + } + + return true; + } + + static bool fpng_pixel_zlib_raw_decompress( + const uint8_t* pSrc, uint32_t src_len, uint32_t zlib_len, + uint8_t* pDst, uint32_t w, uint32_t h, + uint32_t src_chans, uint32_t dst_chans) + { + assert((src_chans == 3) || (src_chans == 4)); + assert((dst_chans == 3) || (dst_chans == 4)); + + const uint32_t src_bpl = w * src_chans; + const uint32_t dst_bpl = w * dst_chans; + const uint32_t dst_len = dst_bpl * h; + + uint32_t src_ofs = 2; + uint32_t dst_ofs = 0; + uint32_t raster_ofs = 0; + uint32_t comp_ofs = 0; + + for (; ; ) + { + if ((src_ofs + 1) > src_len) + return false; + + const bool bfinal = (pSrc[src_ofs] & 1) != 0; + const uint32_t btype = (pSrc[src_ofs] >> 1) & 3; + if (btype != 0) + return false; + + src_ofs++; + + if ((src_ofs + 4) > src_len) + return false; + uint32_t len = pSrc[src_ofs + 0] | (pSrc[src_ofs + 1] << 8); + uint32_t nlen = pSrc[src_ofs + 2] | (pSrc[src_ofs + 3] << 8); + src_ofs += 4; + + if (len != (~nlen & 0xFFFF)) + return false; + + if ((src_ofs + len) > src_len) + return false; + + // Raw blocks are a relatively uncommon case so this isn't well optimized. + // Supports 3->4 and 4->3 byte/pixel conversion. + for (uint32_t i = 0; i < len; i++) + { + uint32_t c = pSrc[src_ofs + i]; + + if (!raster_ofs) + { + // Check filter type + if (c != 0) + return false; + + assert(!comp_ofs); + } + else + { + if (comp_ofs < dst_chans) + { + if (dst_ofs == dst_len) + return false; + + pDst[dst_ofs++] = (uint8_t)c; + } + + if (++comp_ofs == src_chans) + { + if (dst_chans > src_chans) + { + if (dst_ofs == dst_len) + return false; + + pDst[dst_ofs++] = (uint8_t)0xFF; + } + + comp_ofs = 0; + } + } + + if (++raster_ofs == (src_bpl + 1)) + { + assert(!comp_ofs); + raster_ofs = 0; + } + } + + src_ofs += len; + + if (bfinal) + break; + } + + if (comp_ofs != 0) + return false; + + // Check for zlib adler32 + if ((src_ofs + 4) != zlib_len) + return false; + + return (dst_ofs == dst_len); + } + + template + static bool fpng_pixel_zlib_decompress_3( + const uint8_t* pSrc, uint32_t src_len, uint32_t zlib_len, + uint8_t* pDst, uint32_t w, uint32_t h) + { + assert(src_len >= (zlib_len + 4)); + + const uint32_t dst_bpl = w * dst_comps; + //const uint32_t dst_len = dst_bpl * h; + + if (zlib_len < 7) + return false; + + // check zlib header + if ((pSrc[0] != 0x78) || (pSrc[1] != 0x01)) + return false; + + uint32_t src_ofs = 2; + + if ((pSrc[src_ofs] & 6) == 0) + return fpng_pixel_zlib_raw_decompress(pSrc, src_len, zlib_len, pDst, w, h, 3, dst_comps); + + if ((src_ofs + 4) > src_len) + return false; + uint64_t bit_buf = READ_LE32(pSrc + src_ofs); + src_ofs += 4; + + uint32_t bit_buf_size = 32; + + uint32_t bfinal, btype; + GET_BITS(bfinal, 1); + GET_BITS(btype, 2); + + // Must be the final block or it's not valid, and type=2 (dynamic) + if ((bfinal != 1) || (btype != 2)) + return false; + + uint32_t lit_table[FPNG_DECODER_TABLE_SIZE]; + if (!prepare_dynamic_block(pSrc, src_len, src_ofs, bit_buf_size, bit_buf, lit_table, 3)) + return false; + + const uint8_t* pPrev_scanline = nullptr; + uint8_t* pCur_scanline = pDst; + + for (uint32_t y = 0; y < h; y++) + { + // At start of PNG scanline, so read the filter literal + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + uint32_t filter = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t filter_len = (filter >> 9) & 15; + if (!filter_len) + return false; + SKIP_BITS(filter_len); + filter &= 511; + + uint32_t expected_filter = (y ? 2 : 0); + if (filter != expected_filter) + return false; + + uint32_t x_ofs = 0; + uint8_t prev_delta_r = 0, prev_delta_g = 0, prev_delta_b = 0; + do + { + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + uint32_t lit0_tab = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + + uint32_t lit0 = lit0_tab; + uint32_t lit0_len = (lit0_tab >> 9) & 15; + if (!lit0_len) + return false; + SKIP_BITS(lit0_len); + + if (lit0 & 256) + { + lit0 &= 511; + + // Can't be EOB - we still have more pixels to decompress. + if (lit0 == 256) + return false; + + // Must be an RLE match against the previous pixel. + uint32_t run_len = s_length_range[lit0 - 257]; + if (lit0 >= 265) + { + uint32_t e; + GET_BITS_NE(e, s_length_extra[lit0 - 257]); + + run_len += e; + } + + // Skip match distance - it's always the same (3) + SKIP_BITS_NE(1); + + // Matches must always be a multiple of 3/4 bytes + assert((run_len % 3) == 0); + + if (dst_comps == 4) + { + const uint32_t x_ofs_end = x_ofs + g_run_len3_to_4[run_len]; + + // Check for valid run lengths + if (x_ofs == x_ofs_end) + return false; + + // Matches cannot cross scanlines. + if (x_ofs_end > dst_bpl) + return false; + + if (pPrev_scanline) + { + if ((prev_delta_r | prev_delta_g | prev_delta_b) == 0) + { + memcpy(pCur_scanline + x_ofs, pPrev_scanline + x_ofs, x_ofs_end - x_ofs); + x_ofs = x_ofs_end; + } + else + { + do + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + prev_delta_r); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + prev_delta_g); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + prev_delta_b); + pCur_scanline[x_ofs + 3] = 0xFF; + x_ofs += 4; + } while (x_ofs < x_ofs_end); + } + } + else + { + do + { + pCur_scanline[x_ofs] = prev_delta_r; + pCur_scanline[x_ofs + 1] = prev_delta_g; + pCur_scanline[x_ofs + 2] = prev_delta_b; + pCur_scanline[x_ofs + 3] = 0xFF; + x_ofs += 4; + } while (x_ofs < x_ofs_end); + } + } + else + { + // Check for valid run lengths + if (!g_run_len3_to_4[run_len]) + return false; + + const uint32_t x_ofs_end = x_ofs + run_len; + + // Matches cannot cross scanlines. + if (x_ofs_end > dst_bpl) + return false; + + if (pPrev_scanline) + { + if ((prev_delta_r | prev_delta_g | prev_delta_b) == 0) + { + memcpy(pCur_scanline + x_ofs, pPrev_scanline + x_ofs, run_len); + x_ofs = x_ofs_end; + } + else + { + do + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + prev_delta_r); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + prev_delta_g); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + prev_delta_b); + x_ofs += 3; + } while (x_ofs < x_ofs_end); + } + } + else + { + do + { + pCur_scanline[x_ofs] = prev_delta_r; + pCur_scanline[x_ofs + 1] = prev_delta_g; + pCur_scanline[x_ofs + 2] = prev_delta_b; + x_ofs += 3; + } while (x_ofs < x_ofs_end); + } + } + } + else + { + uint32_t lit1, lit2; + + uint32_t lit1_spec_len = (lit0_tab >> (16 + 9)); + uint32_t lit2_len; + if (lit1_spec_len) + { + lit1 = (lit0_tab >> 16) & 511; + SKIP_BITS_NE(lit1_spec_len); + + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit2 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + lit2_len = (lit2 >> 9) & 15; + if (!lit2_len) + return false; + } + else + { + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit1 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t lit1_len = (lit1 >> 9) & 15; + if (!lit1_len) + return false; + SKIP_BITS_NE(lit1_len); + + lit2_len = (lit1 >> (16 + 9)); + if (lit2_len) + lit2 = lit1 >> 16; + else + { + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit2 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + lit2_len = (lit2 >> 9) & 15; + if (!lit2_len) + return false; + } + } + + SKIP_BITS(lit2_len); + + // Check for matches + if ((lit1 | lit2) & 256) + return false; + + if (dst_comps == 4) + { + if (pPrev_scanline) + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + lit0); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + lit1); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + lit2); + pCur_scanline[x_ofs + 3] = 0xFF; + } + else + { + pCur_scanline[x_ofs] = (uint8_t)lit0; + pCur_scanline[x_ofs + 1] = (uint8_t)lit1; + pCur_scanline[x_ofs + 2] = (uint8_t)lit2; + pCur_scanline[x_ofs + 3] = 0xFF; + } + x_ofs += 4; + } + else + { + if (pPrev_scanline) + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + lit0); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + lit1); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + lit2); + } + else + { + pCur_scanline[x_ofs] = (uint8_t)lit0; + pCur_scanline[x_ofs + 1] = (uint8_t)lit1; + pCur_scanline[x_ofs + 2] = (uint8_t)lit2; + } + x_ofs += 3; + } + + prev_delta_r = (uint8_t)lit0; + prev_delta_g = (uint8_t)lit1; + prev_delta_b = (uint8_t)lit2; + + // See if we can decode one more pixel. + uint32_t spec_next_len0_len = lit2 >> (16 + 9); + if ((spec_next_len0_len) && (x_ofs < dst_bpl)) + { + lit0 = (lit2 >> 16) & 511; + if (lit0 < 256) + { + SKIP_BITS_NE(spec_next_len0_len); + + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit1 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t lit1_len = (lit1 >> 9) & 15; + if (!lit1_len) + return false; + SKIP_BITS(lit1_len); + + lit2_len = (lit1 >> (16 + 9)); + if (lit2_len) + lit2 = lit1 >> 16; + else + { + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit2 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + lit2_len = (lit2 >> 9) & 15; + if (!lit2_len) + return false; + } + + SKIP_BITS_NE(lit2_len); + + // Check for matches + if ((lit1 | lit2) & 256) + return false; + + if (dst_comps == 4) + { + if (pPrev_scanline) + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + lit0); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + lit1); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + lit2); + pCur_scanline[x_ofs + 3] = 0xFF; + } + else + { + pCur_scanline[x_ofs] = (uint8_t)lit0; + pCur_scanline[x_ofs + 1] = (uint8_t)lit1; + pCur_scanline[x_ofs + 2] = (uint8_t)lit2; + pCur_scanline[x_ofs + 3] = 0xFF; + } + x_ofs += 4; + } + else + { + if (pPrev_scanline) + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + lit0); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + lit1); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + lit2); + } + else + { + pCur_scanline[x_ofs] = (uint8_t)lit0; + pCur_scanline[x_ofs + 1] = (uint8_t)lit1; + pCur_scanline[x_ofs + 2] = (uint8_t)lit2; + } + x_ofs += 3; + } + + prev_delta_r = (uint8_t)lit0; + prev_delta_g = (uint8_t)lit1; + prev_delta_b = (uint8_t)lit2; + + } // if (lit0 < 256) + + } // if ((spec_next_len0_len) && (x_ofs < bpl)) + } + + } while (x_ofs < dst_bpl); + + pPrev_scanline = pCur_scanline; + pCur_scanline += dst_bpl; + + } // y + + // The last symbol should be EOB + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + uint32_t lit0 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t lit0_len = (lit0 >> 9) & 15; + if (!lit0_len) + return false; + lit0 &= 511; + if (lit0 != 256) + return false; + + bit_buf_size -= lit0_len; + bit_buf >>= lit0_len; + + uint32_t align_bits = bit_buf_size & 7; + bit_buf_size -= align_bits; + bit_buf >>= align_bits; + + if (src_ofs < (bit_buf_size >> 3)) + return false; + src_ofs -= (bit_buf_size >> 3); + + // We should be at the very end, because the bit buf reads ahead 32-bits (which contains the zlib adler32). + if ((src_ofs + 4) != zlib_len) + return false; + + return true; + } + + template + static bool fpng_pixel_zlib_decompress_4( + const uint8_t* pSrc, uint32_t src_len, uint32_t zlib_len, + uint8_t* pDst, uint32_t w, uint32_t h) + { + assert(src_len >= (zlib_len + 4)); + + const uint32_t dst_bpl = w * dst_comps; + //const uint32_t dst_len = dst_bpl * h; + + if (zlib_len < 7) + return false; + + // check zlib header + if ((pSrc[0] != 0x78) || (pSrc[1] != 0x01)) + return false; + + uint32_t src_ofs = 2; + + if ((pSrc[src_ofs] & 6) == 0) + return fpng_pixel_zlib_raw_decompress(pSrc, src_len, zlib_len, pDst, w, h, 4, dst_comps); + + if ((src_ofs + 4) > src_len) + return false; + uint64_t bit_buf = READ_LE32(pSrc + src_ofs); + src_ofs += 4; + + uint32_t bit_buf_size = 32; + + uint32_t bfinal, btype; + GET_BITS(bfinal, 1); + GET_BITS(btype, 2); + + // Must be the final block or it's not valid, and type=2 (dynamic) + if ((bfinal != 1) || (btype != 2)) + return false; + + uint32_t lit_table[FPNG_DECODER_TABLE_SIZE]; + if (!prepare_dynamic_block(pSrc, src_len, src_ofs, bit_buf_size, bit_buf, lit_table, 4)) + return false; + + const uint8_t* pPrev_scanline = nullptr; + uint8_t* pCur_scanline = pDst; + + for (uint32_t y = 0; y < h; y++) + { + // At start of PNG scanline, so read the filter literal + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + uint32_t filter = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t filter_len = (filter >> 9) & 15; + if (!filter_len) + return false; + SKIP_BITS(filter_len); + filter &= 511; + + uint32_t expected_filter = (y ? 2 : 0); + if (filter != expected_filter) + return false; + + uint32_t x_ofs = 0; + uint8_t prev_delta_r = 0, prev_delta_g = 0, prev_delta_b = 0, prev_delta_a = 0; + do + { + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + uint32_t lit0_tab = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + + uint32_t lit0 = lit0_tab; + uint32_t lit0_len = (lit0_tab >> 9) & 15; + if (!lit0_len) + return false; + SKIP_BITS(lit0_len); + + if (lit0 & 256) + { + lit0 &= 511; + + // Can't be EOB - we still have more pixels to decompress. + if (lit0 == 256) + return false; + + // Must be an RLE match against the previous pixel. + uint32_t run_len = s_length_range[lit0 - 257]; + if (lit0 >= 265) + { + uint32_t e; + GET_BITS_NE(e, s_length_extra[lit0 - 257]); + + run_len += e; + } + + // Skip match distance - it's always the same (4) + SKIP_BITS_NE(1); + + // Matches must always be a multiple of 3/4 bytes + if (run_len & 3) + return false; + + if (dst_comps == 3) + { + const uint32_t run_len3 = (run_len >> 2) * 3; + const uint32_t x_ofs_end = x_ofs + run_len3; + + // Matches cannot cross scanlines. + if (x_ofs_end > dst_bpl) + return false; + + if (pPrev_scanline) + { + if ((prev_delta_r | prev_delta_g | prev_delta_b | prev_delta_a) == 0) + { + memcpy(pCur_scanline + x_ofs, pPrev_scanline + x_ofs, run_len3); + x_ofs = x_ofs_end; + } + else + { + do + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + prev_delta_r); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + prev_delta_g); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + prev_delta_b); + x_ofs += 3; + } while (x_ofs < x_ofs_end); + } + } + else + { + do + { + pCur_scanline[x_ofs] = prev_delta_r; + pCur_scanline[x_ofs + 1] = prev_delta_g; + pCur_scanline[x_ofs + 2] = prev_delta_b; + x_ofs += 3; + } while (x_ofs < x_ofs_end); + } + } + else + { + const uint32_t x_ofs_end = x_ofs + run_len; + + // Matches cannot cross scanlines. + if (x_ofs_end > dst_bpl) + return false; + + if (pPrev_scanline) + { + if ((prev_delta_r | prev_delta_g | prev_delta_b | prev_delta_a) == 0) + { + memcpy(pCur_scanline + x_ofs, pPrev_scanline + x_ofs, run_len); + x_ofs = x_ofs_end; + } + else + { + do + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + prev_delta_r); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + prev_delta_g); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + prev_delta_b); + pCur_scanline[x_ofs + 3] = (uint8_t)(pPrev_scanline[x_ofs + 3] + prev_delta_a); + x_ofs += 4; + } while (x_ofs < x_ofs_end); + } + } + else + { + do + { + pCur_scanline[x_ofs] = prev_delta_r; + pCur_scanline[x_ofs + 1] = prev_delta_g; + pCur_scanline[x_ofs + 2] = prev_delta_b; + pCur_scanline[x_ofs + 3] = prev_delta_a; + x_ofs += 4; + } while (x_ofs < x_ofs_end); + } + } + } + else + { + uint32_t lit1, lit2; + + uint32_t lit1_spec_len = (lit0_tab >> (16 + 9)); + uint32_t lit2_len; + if (lit1_spec_len) + { + lit1 = (lit0_tab >> 16) & 511; + SKIP_BITS_NE(lit1_spec_len); + + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit2 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + lit2_len = (lit2 >> 9) & 15; + if (!lit2_len) + return false; + } + else + { + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit1 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t lit1_len = (lit1 >> 9) & 15; + if (!lit1_len) + return false; + SKIP_BITS_NE(lit1_len); + + lit2_len = (lit1 >> (16 + 9)); + if (lit2_len) + lit2 = lit1 >> 16; + else + { + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit2 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + lit2_len = (lit2 >> 9) & 15; + if (!lit2_len) + return false; + } + } + + uint32_t lit3; + uint32_t lit3_len = lit2 >> (16 + 9); + + if (lit3_len) + { + lit3 = (lit2 >> 16); + SKIP_BITS(lit2_len + lit3_len); + } + else + { + SKIP_BITS(lit2_len); + + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + lit3 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + lit3_len = (lit3 >> 9) & 15; + if (!lit3_len) + return false; + + SKIP_BITS_NE(lit3_len); + } + + // Check for matches + if ((lit1 | lit2 | lit3) & 256) + return false; + + if (dst_comps == 3) + { + if (pPrev_scanline) + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + lit0); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + lit1); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + lit2); + } + else + { + pCur_scanline[x_ofs] = (uint8_t)lit0; + pCur_scanline[x_ofs + 1] = (uint8_t)lit1; + pCur_scanline[x_ofs + 2] = (uint8_t)lit2; + } + + x_ofs += 3; + } + else + { + if (pPrev_scanline) + { + pCur_scanline[x_ofs] = (uint8_t)(pPrev_scanline[x_ofs] + lit0); + pCur_scanline[x_ofs + 1] = (uint8_t)(pPrev_scanline[x_ofs + 1] + lit1); + pCur_scanline[x_ofs + 2] = (uint8_t)(pPrev_scanline[x_ofs + 2] + lit2); + pCur_scanline[x_ofs + 3] = (uint8_t)(pPrev_scanline[x_ofs + 3] + lit3); + } + else + { + pCur_scanline[x_ofs] = (uint8_t)lit0; + pCur_scanline[x_ofs + 1] = (uint8_t)lit1; + pCur_scanline[x_ofs + 2] = (uint8_t)lit2; + pCur_scanline[x_ofs + 3] = (uint8_t)lit3; + } + + x_ofs += 4; + } + + prev_delta_r = (uint8_t)lit0; + prev_delta_g = (uint8_t)lit1; + prev_delta_b = (uint8_t)lit2; + prev_delta_a = (uint8_t)lit3; + } + + } while (x_ofs < dst_bpl); + + pPrev_scanline = pCur_scanline; + pCur_scanline += dst_bpl; + } // y + + // The last symbol should be EOB + assert(bit_buf_size >= FPNG_DECODER_TABLE_BITS); + uint32_t lit0 = lit_table[bit_buf & (FPNG_DECODER_TABLE_SIZE - 1)]; + uint32_t lit0_len = (lit0 >> 9) & 15; + if (!lit0_len) + return false; + lit0 &= 511; + if (lit0 != 256) + return false; + + bit_buf_size -= lit0_len; + bit_buf >>= lit0_len; + + uint32_t align_bits = bit_buf_size & 7; + bit_buf_size -= align_bits; + bit_buf >>= align_bits; + + if (src_ofs < (bit_buf_size >> 3)) + return false; + src_ofs -= (bit_buf_size >> 3); + + // We should be at the very end, because the bit buf reads ahead 32-bits (which contains the zlib adler32). + if ((src_ofs + 4) != zlib_len) + return false; + + return true; + } + +#pragma pack(push) +#pragma pack(1) + struct png_chunk_prefix + { + uint32_t m_length; + uint8_t m_type[4]; + }; + struct png_ihdr + { + png_chunk_prefix m_prefix; + uint32_t m_width; + uint32_t m_height; + uint8_t m_bitdepth; + uint8_t m_color_type; + uint8_t m_comp_method; + uint8_t m_filter_method; + uint8_t m_interlace_method; + uint32_t m_crc32; + }; + const uint32_t IHDR_EXPECTED_LENGTH = 13; + struct png_iend + { + png_chunk_prefix m_prefix; + uint32_t m_crc32; + }; +#pragma pack(pop) + + static int fpng_get_info_internal(const void* pImage, uint32_t image_size, uint32_t& width, uint32_t& height, uint32_t& channels_in_file, uint32_t &idat_ofs, uint32_t &idat_len) + { + static const uint8_t s_png_sig[8] = { 137, 80, 78, 71, 13, 10, 26, 10 }; + + if (!endian_check()) + { + assert(0); + return false; + } + + width = 0; + height = 0; + channels_in_file = 0; + idat_ofs = 0, idat_len = 0; + + // Ensure the file has at least a minimum possible size + if (image_size < (sizeof(s_png_sig) + sizeof(png_ihdr) + sizeof(png_chunk_prefix) + 1 + sizeof(uint32_t) + sizeof(png_iend))) + return FPNG_DECODE_FAILED_NOT_PNG; + + if (memcmp(pImage, s_png_sig, 8) != 0) + return FPNG_DECODE_FAILED_NOT_PNG; + + const uint8_t* pImage_u8 = static_cast(pImage) + 8; + + const png_ihdr& ihdr = *reinterpret_cast(pImage_u8); + pImage_u8 += sizeof(png_ihdr); + + if (READ_BE32(&ihdr.m_prefix.m_length) != IHDR_EXPECTED_LENGTH) + return FPNG_DECODE_FAILED_NOT_PNG; + + if (fpng_crc32(ihdr.m_prefix.m_type, 4 + IHDR_EXPECTED_LENGTH, FPNG_CRC32_INIT) != READ_BE32(&ihdr.m_crc32)) + return FPNG_DECODE_FAILED_HEADER_CRC32; + + width = READ_BE32(&ihdr.m_width); + height = READ_BE32(&ihdr.m_height); + + if (!width || !height || (width > FPNG_MAX_SUPPORTED_DIM) || (height > FPNG_MAX_SUPPORTED_DIM)) + return FPNG_DECODE_FAILED_INVALID_DIMENSIONS; + + uint64_t total_pixels = (uint64_t)width * height; + if (total_pixels > (1 << 30)) + return FPNG_DECODE_FAILED_INVALID_DIMENSIONS; + + if ((ihdr.m_comp_method) || (ihdr.m_filter_method) || (ihdr.m_interlace_method) || (ihdr.m_bitdepth != 8)) + return FPNG_DECODE_NOT_FPNG; + + if (ihdr.m_color_type == 2) + channels_in_file = 3; + else if (ihdr.m_color_type == 6) + channels_in_file = 4; + + if (!channels_in_file) + return FPNG_DECODE_NOT_FPNG; + + // Scan all the chunks. Look for one IDAT, IEND, and our custom fdEC chunk that indicates the file was compressed by us. Skip any ancillary chunks. + bool found_fdec_chunk = false; + + for (; ; ) + { + const size_t src_ofs = pImage_u8 - static_cast(pImage); + if (src_ofs >= image_size) + return FPNG_DECODE_FAILED_CHUNK_PARSING; + + const uint32_t bytes_remaining = image_size - (uint32_t)src_ofs; + if (bytes_remaining < sizeof(uint32_t) * 3) + return FPNG_DECODE_FAILED_CHUNK_PARSING; + + const png_chunk_prefix* pChunk = reinterpret_cast(pImage_u8); + + const uint32_t chunk_len = READ_BE32(&pChunk->m_length); + if ((src_ofs + sizeof(uint32_t) * 2 + (uint64_t)chunk_len + sizeof(uint32_t)) > image_size) + return FPNG_DECODE_FAILED_CHUNK_PARSING; + + for (uint32_t i = 0; i < 4; i++) + { + const uint8_t c = pChunk->m_type[i]; + const bool is_upper = (c >= 65) && (c <= 90), is_lower = (c >= 97) && (c <= 122); + if ((!is_upper) && (!is_lower)) + return FPNG_DECODE_FAILED_CHUNK_PARSING; + } + + const uint32_t expected_crc32 = READ_BE32(pImage_u8 + sizeof(uint32_t) * 2 + chunk_len); + + char chunk_type[5] = { (char)pChunk->m_type[0], (char)pChunk->m_type[1], (char)pChunk->m_type[2], (char)pChunk->m_type[3], 0 }; + const bool is_idat = strcmp(chunk_type, "IDAT") == 0; + +#if !FPNG_DISABLE_DECODE_CRC32_CHECKS + if (!is_idat) + { + uint32_t actual_crc32 = fpng_crc32(pImage_u8 + sizeof(uint32_t), sizeof(uint32_t) + chunk_len, FPNG_CRC32_INIT); + if (actual_crc32 != expected_crc32) + return FPNG_DECODE_FAILED_HEADER_CRC32; + } +#endif + + const uint8_t* pChunk_data = pImage_u8 + sizeof(uint32_t) * 2; + + if (strcmp(chunk_type, "IEND") == 0) + break; + else if (is_idat) + { + // If there were multiple IDAT's, or we didn't find the fdEC chunk, then it's not FPNG. + if ((idat_ofs) || (!found_fdec_chunk)) + return FPNG_DECODE_NOT_FPNG; + + idat_ofs = (uint32_t)src_ofs; + idat_len = chunk_len; + + // Sanity check the IDAT chunk length + if (idat_len < 7) + return FPNG_DECODE_FAILED_INVALID_IDAT; + } + else if (strcmp(chunk_type, "fdEC") == 0) + { + if (found_fdec_chunk) + return FPNG_DECODE_NOT_FPNG; + + // We've got our fdEC chunk. Now make sure it's big enough and check its contents. + if (chunk_len != 5) + return FPNG_DECODE_NOT_FPNG; + + // Check fdEC chunk sig + if ((pChunk_data[0] != 82) || (pChunk_data[1] != 36) || (pChunk_data[2] != 147) || (pChunk_data[3] != 227)) + return FPNG_DECODE_NOT_FPNG; + + // Check fdEC version + if (pChunk_data[4] != FPNG_FDEC_VERSION) + return FPNG_DECODE_NOT_FPNG; + + found_fdec_chunk = true; + } + else + { + // Bail if it's a critical chunk - can't be FPNG + if ((chunk_type[0] & 32) == 0) + return FPNG_DECODE_NOT_FPNG; + + // ancillary chunk - skip it + } + + pImage_u8 += sizeof(png_chunk_prefix) + chunk_len + sizeof(uint32_t); + } + + if ((!found_fdec_chunk) || (!idat_ofs)) + return FPNG_DECODE_NOT_FPNG; + + return FPNG_DECODE_SUCCESS; + } + + int fpng_get_info(const void* pImage, uint32_t image_size, uint32_t& width, uint32_t& height, uint32_t& channels_in_file) + { + uint32_t idat_ofs = 0, idat_len = 0; + return fpng_get_info_internal(pImage, image_size, width, height, channels_in_file, idat_ofs, idat_len); + } + + int fpng_decode_memory(const void *pImage, uint32_t image_size, std::vector &out, uint32_t& width, uint32_t& height, uint32_t &channels_in_file, uint32_t desired_channels) + { + out.resize(0); + width = 0; + height = 0; + channels_in_file = 0; + + if ((!pImage) || (!image_size) || ((desired_channels != 3) && (desired_channels != 4))) + { + assert(0); + return FPNG_DECODE_INVALID_ARG; + } + + uint32_t idat_ofs = 0, idat_len = 0; + int status = fpng_get_info_internal(pImage, image_size, width, height, channels_in_file, idat_ofs, idat_len); + if (status) + return status; + + const uint64_t mem_needed = (uint64_t)width * height * desired_channels; + if (mem_needed > UINT32_MAX) + return FPNG_DECODE_FAILED_DIMENSIONS_TOO_LARGE; + + // On 32-bit systems do a quick sanity check before we try to resize the output buffer. + if ((sizeof(size_t) == sizeof(uint32_t)) && (mem_needed >= 0x80000000)) + return FPNG_DECODE_FAILED_DIMENSIONS_TOO_LARGE; + + out.resize(mem_needed); + + const uint8_t* pIDAT_data = static_cast(pImage) + idat_ofs + sizeof(uint32_t) * 2; + const uint32_t src_len = image_size - (idat_ofs + sizeof(uint32_t) * 2); + + bool decomp_status; + if (desired_channels == 3) + { + if (channels_in_file == 3) + decomp_status = fpng_pixel_zlib_decompress_3<3>(pIDAT_data, src_len, idat_len, out.data(), width, height); + else + decomp_status = fpng_pixel_zlib_decompress_4<3>(pIDAT_data, src_len, idat_len, out.data(), width, height); + } + else + { + if (channels_in_file == 3) + decomp_status = fpng_pixel_zlib_decompress_3<4>(pIDAT_data, src_len, idat_len, out.data(), width, height); + else + decomp_status = fpng_pixel_zlib_decompress_4<4>(pIDAT_data, src_len, idat_len, out.data(), width, height); + } + if (!decomp_status) + { + // Something went wrong. Either the file data was corrupted, or it doesn't conform to one of our zlib/Deflate constraints. + // The conservative thing to do is indicate it wasn't written by us, and let the general purpose PNG decoder handle it. + return FPNG_DECODE_NOT_FPNG; + } + + return FPNG_DECODE_SUCCESS; + } + +#ifndef FPNG_NO_STDIO + int fpng_decode_file(const char* pFilename, std::vector& out, uint32_t& width, uint32_t& height, uint32_t& channels_in_file, uint32_t desired_channels) + { + FILE* pFile = nullptr; + +#ifdef _MSC_VER + fopen_s(&pFile, pFilename, "rb"); +#else + pFile = fopen(pFilename, "rb"); +#endif + + if (!pFile) + return FPNG_DECODE_FILE_OPEN_FAILED; + + if (fseek(pFile, 0, SEEK_END) != 0) + { + fclose(pFile); + return FPNG_DECODE_FILE_SEEK_FAILED; + } + +#ifdef _WIN32 + int64_t filesize = _ftelli64(pFile); +#else + int64_t filesize = ftello(pFile); +#endif + + if (fseek(pFile, 0, SEEK_SET) != 0) + { + fclose(pFile); + return FPNG_DECODE_FILE_SEEK_FAILED; + } + + if ( (filesize < 0) || (filesize > UINT32_MAX) || ( (sizeof(size_t) == sizeof(uint32_t)) && (filesize > 0x70000000) ) ) + { + fclose(pFile); + return FPNG_DECODE_FILE_TOO_LARGE; + } + + std::vector buf((size_t)filesize); + if (fread(buf.data(), 1, buf.size(), pFile) != buf.size()) + { + fclose(pFile); + return FPNG_DECODE_FILE_READ_FAILED; + } + + fclose(pFile); + + return fpng_decode_memory(buf.data(), (uint32_t)buf.size(), out, width, height, channels_in_file, desired_channels); + } +#endif + +} // namespace fpng + +/* + This is free and unencumbered software released into the public domain. + + Anyone is free to copy, modify, publish, use, compile, sell, or + distribute this software, either in source code form or as a compiled + binary, for any purpose, commercial or non-commercial, and by any + means. + + In jurisdictions that recognize copyright laws, the author or authors + of this software dedicate any and all copyright interest in the + software to the public domain. We make this dedication for the benefit + of the public at large and to the detriment of our heirs and + successors. We intend this dedication to be an overt act of + relinquishment in perpetuity of all present and future rights to this + software under copyright law. + + THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, + EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF + MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. + IN NO EVENT SHALL THE AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR + OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, + ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR + OTHER DEALINGS IN THE SOFTWARE. + + For more information, please refer to + + Richard Geldreich, Jr. + 12/30/2021 +*/ diff --git a/src/spice2x/external/fpng/fpng.h b/src/spice2x/external/fpng/fpng.h new file mode 100644 index 0000000..4d55e3a --- /dev/null +++ b/src/spice2x/external/fpng/fpng.h @@ -0,0 +1,122 @@ +// fpng.h - unlicense (see end of fpng.cpp) +#pragma once + +#include +#include +#include + +#ifndef FPNG_TRAIN_HUFFMAN_TABLES + // Set to 1 when using the -t (training) option in fpng_test to generate new opaque/alpha Huffman tables for the single pass encoder. + #define FPNG_TRAIN_HUFFMAN_TABLES (0) +#endif + +namespace fpng +{ + // ---- Library initialization - call once to identify if the processor supports SSE. + // Otherwise you'll only get scalar fallbacks. + void fpng_init(); + + // ---- Useful Utilities + + // Returns true if the CPU supports SSE 4.1, and SSE support wasn't disabled by setting FPNG_NO_SSE=1. + // fpng_init() must have been called first, or it'll assert and return false. + bool fpng_cpu_supports_sse41(); + + // Fast CRC-32 SSE4.1+pclmul or a scalar fallback (slice by 4) + const uint32_t FPNG_CRC32_INIT = 0; + uint32_t fpng_crc32(const void* pData, size_t size, uint32_t prev_crc32 = FPNG_CRC32_INIT); + + // Fast Adler32 SSE4.1 Adler-32 with a scalar fallback. + const uint32_t FPNG_ADLER32_INIT = 1; + uint32_t fpng_adler32(const void* pData, size_t size, uint32_t adler = FPNG_ADLER32_INIT); + + // ---- Compression + enum + { + // Enables computing custom Huffman tables for each file, instead of using the custom global tables. + // Results in roughly 6% smaller files on average, but compression is around 40% slower. + FPNG_ENCODE_SLOWER = 1, + + // Only use raw Deflate blocks (no compression at all). Intended for testing. + FPNG_FORCE_UNCOMPRESSED = 2, + }; + + // Fast PNG encoding. The resulting file can be decoded either using a standard PNG decoder or by the fpng_decode_memory() function below. + // pImage: pointer to RGB or RGBA image pixels, R first in memory, B/A last. + // w/h - image dimensions. Image's row pitch in bytes must is w*num_chans. + // num_chans must be 3 or 4. + bool fpng_encode_image_to_memory(const void* pImage, uint32_t w, uint32_t h, uint32_t num_chans, std::vector& out_buf, uint32_t flags = 0); + +#ifndef FPNG_NO_STDIO + // Fast PNG encoding to the specified file. + bool fpng_encode_image_to_file(const char* pFilename, const void* pImage, uint32_t w, uint32_t h, uint32_t num_chans, uint32_t flags = 0); +#endif + + // ---- Decompression + + enum + { + FPNG_DECODE_SUCCESS = 0, // file is a valid PNG file and written by FPNG and the decode succeeded + + FPNG_DECODE_NOT_FPNG, // file is a valid PNG file, but it wasn't written by FPNG so you should try decoding it with a general purpose PNG decoder + + FPNG_DECODE_INVALID_ARG, // invalid function parameter + + FPNG_DECODE_FAILED_NOT_PNG, // file cannot be a PNG file + FPNG_DECODE_FAILED_HEADER_CRC32, // a chunk CRC32 check failed, file is likely corrupted or not PNG + FPNG_DECODE_FAILED_INVALID_DIMENSIONS, // invalid image dimensions in IHDR chunk (0 or too large) + FPNG_DECODE_FAILED_DIMENSIONS_TOO_LARGE, // decoding the file fully into memory would likely require too much memory (only on 32bpp builds) + FPNG_DECODE_FAILED_CHUNK_PARSING, // failed while parsing the chunk headers, or file is corrupted + FPNG_DECODE_FAILED_INVALID_IDAT, // IDAT data length is too small and cannot be valid, file is either corrupted or it's a bug + + // fpng_decode_file() specific errors + FPNG_DECODE_FILE_OPEN_FAILED, + FPNG_DECODE_FILE_TOO_LARGE, + FPNG_DECODE_FILE_READ_FAILED, + FPNG_DECODE_FILE_SEEK_FAILED + }; + + // Fast PNG decoding of files ONLY created by fpng_encode_image_to_memory() or fpng_encode_image_to_file(). + // If fpng_get_info() or fpng_decode_memory() returns FPNG_DECODE_NOT_FPNG, you should decode the PNG by falling back to a general purpose decoder. + // + // fpng_get_info() parses the PNG header and iterates through all chunks to determine if it's a file written by FPNG, but does not decompress the actual image data so it's relatively fast. + // + // pImage, image_size: Pointer to PNG image data and its size + // width, height: output image's dimensions + // channels_in_file: will be 3 or 4 + // + // Returns FPNG_DECODE_SUCCESS on success, otherwise one of the failure codes above. + // If FPNG_DECODE_NOT_FPNG is returned, you must decompress the file with a general purpose PNG decoder. + // If another error occurs, the file is likely corrupted or invalid, but you can still try to decompress the file with another decoder (which will likely fail). + int fpng_get_info(const void* pImage, uint32_t image_size, uint32_t& width, uint32_t& height, uint32_t& channels_in_file); + + // fpng_decode_memory() decompresses 24/32bpp PNG files ONLY encoded by this module. + // If the image was written by FPNG, it will decompress the image data, otherwise it will return FPNG_DECODE_NOT_FPNG in which case you should fall back to a general purpose PNG decoder (lodepng, stb_image, libpng, etc.) + // + // pImage, image_size: Pointer to PNG image data and its size + // out: Output 24/32bpp image buffer + // width, height: output image's dimensions + // channels_in_file: will be 3 or 4 + // desired_channels: must be 3 or 4 + // + // If the image is 24bpp and 32bpp is requested, the alpha values will be set to 0xFF. + // If the image is 32bpp and 24bpp is requested, the alpha values will be discarded. + // + // Returns FPNG_DECODE_SUCCESS on success, otherwise one of the failure codes above. + // If FPNG_DECODE_NOT_FPNG is returned, you must decompress the file with a general purpose PNG decoder. + // If another error occurs, the file is likely corrupted or invalid, but you can still try to decompress the file with another decoder (which will likely fail). + int fpng_decode_memory(const void* pImage, uint32_t image_size, std::vector& out, uint32_t& width, uint32_t& height, uint32_t& channels_in_file, uint32_t desired_channels); + +#ifndef FPNG_NO_STDIO + int fpng_decode_file(const char* pFilename, std::vector& out, uint32_t& width, uint32_t& height, uint32_t& channels_in_file, uint32_t desired_channels); +#endif + + // ---- Internal API used for Huffman table training purposes + +#if FPNG_TRAIN_HUFFMAN_TABLES + const uint32_t HUFF_COUNTS_SIZE = 288; + extern uint64_t g_huff_counts[HUFF_COUNTS_SIZE]; + bool create_dynamic_block_prefix(uint64_t* pFreq, uint32_t num_chans, std::vector& prefix, uint64_t& bit_buf, int& bit_buf_size, uint32_t *pCodes, uint8_t *pCodesizes); +#endif + +} // namespace fpng diff --git a/src/spice2x/external/libjpeg-turbo/CMakeLists.txt b/src/spice2x/external/libjpeg-turbo/CMakeLists.txt new file mode 100644 index 0000000..eafc39b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/CMakeLists.txt @@ -0,0 +1,143 @@ +# Minimal build for the vendored libjpeg-turbo. +# Upstream's CMake pulls in Java bindings, tools, tests and packaging and sets +# its own global flags, none of which survive contact with this project's four +# cross toolchains, so only the static library is described here. + +set(LJT_VERSION "3.0.4") +set(VERSION ${LJT_VERSION}) +set(JPEG_LIB_VERSION 62) +set(LIBJPEG_TURBO_VERSION_NUMBER 3000004) +set(BUILD "spice2x") + +set(C_ARITH_CODING_SUPPORTED 1) +set(D_ARITH_CODING_SUPPORTED 1) + +if(CMAKE_SIZEOF_VOID_P EQUAL 8) + set(LJT_ARCH "x86_64") +else() + set(LJT_ARCH "i386") +endif() + +# SIMD is the entire reason for taking this dependency, but a missing assembler +# must not break the build; the C paths are still correct, just slower. +# ASM_NASM is enabled by the top level CMakeLists. +if(CMAKE_ASM_NASM_COMPILER) + set(WITH_SIMD 1) +else() + message(WARNING "nasm not found, building libjpeg-turbo without SIMD (screen capture will be slower)") +endif() + +if(MSVC) + set(HIDDEN "") + set(INLINE "__forceinline") + set(THREAD_LOCAL "__declspec(thread)") + set(HAVE_INTRIN_H 1) +else() + set(HIDDEN "__attribute__((visibility(\"hidden\")))") + set(INLINE "inline __attribute__((always_inline))") + set(THREAD_LOCAL "__thread") + set(HAVE_BUILTIN_CTZL 1) +endif() + +set(SIZE_T ${CMAKE_SIZEOF_VOID_P}) + +configure_file(jconfig.h.in ${CMAKE_CURRENT_BINARY_DIR}/jconfig.h) +configure_file(jconfigint.h.in ${CMAKE_CURRENT_BINARY_DIR}/jconfigint.h) +configure_file(jversion.h.in ${CMAKE_CURRENT_BINARY_DIR}/jversion.h) + +# libjpeg-turbo 3.x supports 8, 12 and 16 bit samples by compiling the +# precision dependent units once per precision; jcinit.c dispatches to all +# three at runtime, so none of them can be dropped. +# jstdhuff.c and the *ext.c units are textually included and never compiled. +set(JPEG16_SOURCES + jcapistd.c jccolor.c jcdiffct.c jclossls.c jcmainct.c jcprepct.c + jcsample.c jdapistd.c jdcolor.c jddiffct.c jdlossls.c jdmainct.c + jdpostct.c jdsample.c jutils.c) + +set(JPEG12_SOURCES ${JPEG16_SOURCES} + jccoefct.c jcdctmgr.c jdcoefct.c jddctmgr.c jdmerge.c jfdctfst.c + jfdctint.c jidctflt.c jidctfst.c jidctint.c jidctred.c jquant1.c + jquant2.c) + +set(JPEG_SOURCES ${JPEG12_SOURCES} + jcapimin.c jchuff.c jcicc.c jcinit.c jclhuff.c jcmarker.c jcmaster.c + jcomapi.c jcparam.c jcphuff.c jctrans.c jdapimin.c jdatadst.c + jdatasrc.c jdhuff.c jdicc.c jdinput.c jdlhuff.c jdmarker.c jdmaster.c + jdphuff.c jdtrans.c jerror.c jfdctflt.c jmemmgr.c jmemnobs.c + jpeg_nbits.c + jaricom.c jcarith.c jdarith.c) + +if(WITH_SIMD) + # the *ext-*.asm units are textually included by these, so the list is + # explicit rather than globbed + if(LJT_ARCH STREQUAL "x86_64") + set(LJT_SIMD_ASM + x86_64/jsimdcpu.asm x86_64/jfdctflt-sse.asm + x86_64/jccolor-sse2.asm x86_64/jcgray-sse2.asm x86_64/jchuff-sse2.asm + x86_64/jcphuff-sse2.asm x86_64/jcsample-sse2.asm x86_64/jdcolor-sse2.asm + x86_64/jdmerge-sse2.asm x86_64/jdsample-sse2.asm x86_64/jfdctfst-sse2.asm + x86_64/jfdctint-sse2.asm x86_64/jidctflt-sse2.asm x86_64/jidctfst-sse2.asm + x86_64/jidctint-sse2.asm x86_64/jidctred-sse2.asm x86_64/jquantf-sse2.asm + x86_64/jquanti-sse2.asm + x86_64/jccolor-avx2.asm x86_64/jcgray-avx2.asm x86_64/jcsample-avx2.asm + x86_64/jdcolor-avx2.asm x86_64/jdmerge-avx2.asm x86_64/jdsample-avx2.asm + x86_64/jfdctint-avx2.asm x86_64/jidctint-avx2.asm x86_64/jquanti-avx2.asm) + else() + set(LJT_SIMD_ASM + i386/jsimdcpu.asm i386/jfdctflt-3dn.asm + i386/jidctflt-3dn.asm i386/jquant-3dn.asm + i386/jccolor-mmx.asm i386/jcgray-mmx.asm i386/jcsample-mmx.asm + i386/jdcolor-mmx.asm i386/jdmerge-mmx.asm i386/jdsample-mmx.asm + i386/jfdctfst-mmx.asm i386/jfdctint-mmx.asm i386/jidctfst-mmx.asm + i386/jidctint-mmx.asm i386/jidctred-mmx.asm i386/jquant-mmx.asm + i386/jfdctflt-sse.asm i386/jidctflt-sse.asm i386/jquant-sse.asm + i386/jccolor-sse2.asm i386/jcgray-sse2.asm i386/jchuff-sse2.asm + i386/jcphuff-sse2.asm i386/jcsample-sse2.asm i386/jdcolor-sse2.asm + i386/jdmerge-sse2.asm i386/jdsample-sse2.asm i386/jfdctfst-sse2.asm + i386/jfdctint-sse2.asm i386/jidctflt-sse2.asm i386/jidctfst-sse2.asm + i386/jidctint-sse2.asm i386/jidctred-sse2.asm i386/jquantf-sse2.asm + i386/jquanti-sse2.asm + i386/jccolor-avx2.asm i386/jcgray-avx2.asm i386/jcsample-avx2.asm + i386/jdcolor-avx2.asm i386/jdmerge-avx2.asm i386/jdsample-avx2.asm + i386/jfdctint-avx2.asm i386/jidctint-avx2.asm i386/jquanti-avx2.asm) + endif() + + list(TRANSFORM LJT_SIMD_ASM PREPEND "${CMAKE_CURRENT_SOURCE_DIR}/simd/") + list(APPEND JPEG_SOURCES simd/${LJT_ARCH}/jsimd.c ${LJT_SIMD_ASM}) + + # nasm resolves %include against these, and requires the trailing separator + set(CMAKE_ASM_NASM_FLAGS + "${CMAKE_ASM_NASM_FLAGS} -I${CMAKE_CURRENT_SOURCE_DIR}/simd/nasm/ -I${CMAKE_CURRENT_SOURCE_DIR}/simd/${LJT_ARCH}/") + + if(LJT_ARCH STREQUAL "x86_64") + set(CMAKE_ASM_NASM_OBJECT_FORMAT win64) + set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -D__x86_64__ -DWIN64") + else() + set(CMAKE_ASM_NASM_OBJECT_FORMAT win32) + set(CMAKE_ASM_NASM_FLAGS "${CMAKE_ASM_NASM_FLAGS} -DWIN32") + endif() +endif() + +add_library(jpeg12-static OBJECT ${JPEG12_SOURCES}) +set_property(TARGET jpeg12-static PROPERTY COMPILE_FLAGS "-DBITS_IN_JSAMPLE=12") + +add_library(jpeg16-static OBJECT ${JPEG16_SOURCES}) +set_property(TARGET jpeg16-static PROPERTY COMPILE_FLAGS "-DBITS_IN_JSAMPLE=16") + +add_library(jpeg-static STATIC ${JPEG_SOURCES} + $ + $) + +foreach(ljt_target jpeg-static jpeg12-static jpeg16-static) + target_include_directories(${ljt_target} PUBLIC + ${CMAKE_CURRENT_SOURCE_DIR} + ${CMAKE_CURRENT_BINARY_DIR}) + + target_include_directories(${ljt_target} PRIVATE + ${CMAKE_CURRENT_SOURCE_DIR}/simd + ${CMAKE_CURRENT_SOURCE_DIR}/simd/${LJT_ARCH}) + + if(NOT MSVC) + target_compile_options(${ljt_target} PRIVATE -Wno-unused-parameter -Wno-sign-compare) + endif() +endforeach() diff --git a/src/spice2x/external/libjpeg-turbo/LICENSE.md b/src/spice2x/external/libjpeg-turbo/LICENSE.md new file mode 100644 index 0000000..2204864 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/LICENSE.md @@ -0,0 +1,135 @@ +libjpeg-turbo Licenses +====================== + +libjpeg-turbo is covered by two compatible BSD-style open source licenses: + +- The IJG (Independent JPEG Group) License, which is listed in + [README.ijg](README.ijg) + + This license applies to the libjpeg API library and associated programs, + including any code inherited from libjpeg and any modifications to that + code. Note that the libjpeg-turbo SIMD source code bears the + [zlib License](https://opensource.org/licenses/Zlib), but in the context of + the overall libjpeg API library, the terms of the zlib License are subsumed + by the terms of the IJG License. + +- The Modified (3-clause) BSD License, which is listed below + + This license applies to the TurboJPEG API library and associated programs, as + well as the build system. Note that the TurboJPEG API library wraps the + libjpeg API library, so in the context of the overall TurboJPEG API library, + both the terms of the IJG License and the terms of the Modified (3-clause) + BSD License apply. + + +Complying with the libjpeg-turbo Licenses +========================================= + +This section provides a roll-up of the libjpeg-turbo licensing terms, to the +best of our understanding. This is not a license in and of itself. It is +intended solely for clarification. + +1. If you are distributing a modified version of the libjpeg-turbo source, + then: + + 1. You cannot alter or remove any existing copyright or license notices + from the source. + + **Origin** + - Clause 1 of the IJG License + - Clause 1 of the Modified BSD License + - Clauses 1 and 3 of the zlib License + + 2. You must add your own copyright notice to the header of each source + file you modified, so others can tell that you modified that file. (If + there is not an existing copyright header in that file, then you can + simply add a notice stating that you modified the file.) + + **Origin** + - Clause 1 of the IJG License + - Clause 2 of the zlib License + + 3. You must include the IJG README file, and you must not alter any of the + copyright or license text in that file. + + **Origin** + - Clause 1 of the IJG License + +2. If you are distributing only libjpeg-turbo binaries without the source, or + if you are distributing an application that statically links with + libjpeg-turbo, then: + + 1. Your product documentation must include a message stating: + + This software is based in part on the work of the Independent JPEG + Group. + + **Origin** + - Clause 2 of the IJG license + + 2. If your binary distribution includes or uses the TurboJPEG API, then + your product documentation must include the text of the Modified BSD + License (see below.) + + **Origin** + - Clause 2 of the Modified BSD License + +3. You cannot use the name of the IJG or The libjpeg-turbo Project or the + contributors thereof in advertising, publicity, etc. + + **Origin** + - IJG License + - Clause 3 of the Modified BSD License + +4. The IJG and The libjpeg-turbo Project do not warrant libjpeg-turbo to be + free of defects, nor do we accept any liability for undesirable + consequences resulting from your use of the software. + + **Origin** + - IJG License + - Modified BSD License + - zlib License + + +The Modified (3-clause) BSD License +=================================== + +Copyright (C)2009-2023 D. R. Commander. All Rights Reserved.
+Copyright (C)2015 Viktor Szathmáry. All Rights Reserved. + +Redistribution and use in source and binary forms, with or without +modification, are permitted provided that the following conditions are met: + +- Redistributions of source code must retain the above copyright notice, + this list of conditions and the following disclaimer. +- Redistributions in binary form must reproduce the above copyright notice, + this list of conditions and the following disclaimer in the documentation + and/or other materials provided with the distribution. +- Neither the name of the libjpeg-turbo Project nor the names of its + contributors may be used to endorse or promote products derived from this + software without specific prior written permission. + +THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS", +AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE +IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE +ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDERS OR CONTRIBUTORS BE +LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR +CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF +SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS +INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN +CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) +ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE +POSSIBILITY OF SUCH DAMAGE. + + +Why Two Licenses? +================= + +The zlib License could have been used instead of the Modified (3-clause) BSD +License, and since the IJG License effectively subsumes the distribution +conditions of the zlib License, this would have effectively placed +libjpeg-turbo binary distributions under the IJG License. However, the IJG +License specifically refers to the Independent JPEG Group and does not extend +attribution and endorsement protections to other entities. Thus, it was +desirable to choose a license that granted us the same protections for new code +that were granted to the IJG for code derived from their software. diff --git a/src/spice2x/external/libjpeg-turbo/README.ijg b/src/spice2x/external/libjpeg-turbo/README.ijg new file mode 100644 index 0000000..1be3595 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/README.ijg @@ -0,0 +1,258 @@ +libjpeg-turbo note: This file has been modified by The libjpeg-turbo Project +to include only information relevant to libjpeg-turbo, to wordsmith certain +sections, and to remove impolitic language that existed in the libjpeg v8 +README. It is included only for reference. Please see README.md for +information specific to libjpeg-turbo. + + +The Independent JPEG Group's JPEG software +========================================== + +This distribution contains a release of the Independent JPEG Group's free JPEG +software. You are welcome to redistribute this software and to use it for any +purpose, subject to the conditions under LEGAL ISSUES, below. + +This software is the work of Tom Lane, Guido Vollbeding, Philip Gladstone, +Bill Allombert, Jim Boucher, Lee Crocker, Bob Friesenhahn, Ben Jackson, +Julian Minguillon, Luis Ortiz, George Phillips, Davide Rossi, Ge' Weijers, +and other members of the Independent JPEG Group. + +IJG is not affiliated with the ISO/IEC JTC1/SC29/WG1 standards committee +(also known as JPEG, together with ITU-T SG16). + + +DOCUMENTATION ROADMAP +===================== + +This file contains the following sections: + +OVERVIEW General description of JPEG and the IJG software. +LEGAL ISSUES Copyright, lack of warranty, terms of distribution. +REFERENCES Where to learn more about JPEG. +ARCHIVE LOCATIONS Where to find newer versions of this software. +FILE FORMAT WARS Software *not* to get. +TO DO Plans for future IJG releases. + +Other documentation files in the distribution are: + +User documentation: + usage.txt Usage instructions for cjpeg, djpeg, jpegtran, + rdjpgcom, and wrjpgcom. + *.1 Unix-style man pages for programs (same info as usage.txt). + wizard.txt Advanced usage instructions for JPEG wizards only. + change.log Version-to-version change highlights. +Programmer and internal documentation: + libjpeg.txt How to use the JPEG library in your own programs. + example.c Sample code for calling the JPEG library. + structure.txt Overview of the JPEG library's internal structure. + coderules.txt Coding style rules --- please read if you contribute code. + +Please read at least usage.txt. Some information can also be found in the JPEG +FAQ (Frequently Asked Questions) article. See ARCHIVE LOCATIONS below to find +out where to obtain the FAQ article. + +If you want to understand how the JPEG code works, we suggest reading one or +more of the REFERENCES, then looking at the documentation files (in roughly +the order listed) before diving into the code. + + +OVERVIEW +======== + +This package contains C software to implement JPEG image encoding, decoding, +and transcoding. JPEG (pronounced "jay-peg") is a standardized compression +method for full-color and grayscale images. JPEG's strong suit is compressing +photographic images or other types of images that have smooth color and +brightness transitions between neighboring pixels. Images with sharp lines or +other abrupt features may not compress well with JPEG, and a higher JPEG +quality may have to be used to avoid visible compression artifacts with such +images. + +JPEG is normally lossy, meaning that the output pixels are not necessarily +identical to the input pixels. However, on photographic content and other +"smooth" images, very good compression ratios can be obtained with no visible +compression artifacts, and extremely high compression ratios are possible if +you are willing to sacrifice image quality (by reducing the "quality" setting +in the compressor.) + +This software implements JPEG baseline, extended-sequential, progressive, and +lossless compression processes. Provision is made for supporting all variants +of these processes, although some uncommon parameter settings aren't +implemented yet. We have made no provision for supporting the hierarchical +processes defined in the standard. + +We provide a set of library routines for reading and writing JPEG image files, +plus two sample applications "cjpeg" and "djpeg", which use the library to +perform conversion between JPEG and some other popular image file formats. +The library is intended to be reused in other applications. + +In order to support file conversion and viewing software, we have included +considerable functionality beyond the bare JPEG coding/decoding capability; +for example, the color quantization modules are not strictly part of JPEG +decoding, but they are essential for output to colormapped file formats. These +extra functions can be compiled out of the library if not required for a +particular application. + +We have also included "jpegtran", a utility for lossless transcoding between +different JPEG processes, and "rdjpgcom" and "wrjpgcom", two simple +applications for inserting and extracting textual comments in JFIF files. + +The emphasis in designing this software has been on achieving portability and +flexibility, while also making it fast enough to be useful. In particular, +the software is not intended to be read as a tutorial on JPEG. (See the +REFERENCES section for introductory material.) Rather, it is intended to +be reliable, portable, industrial-strength code. We do not claim to have +achieved that goal in every aspect of the software, but we strive for it. + +We welcome the use of this software as a component of commercial products. +No royalty is required, but we do ask for an acknowledgement in product +documentation, as described under LEGAL ISSUES. + + +LEGAL ISSUES +============ + +In plain English: + +1. We don't promise that this software works. (But if you find any bugs, + please let us know!) +2. You can use this software for whatever you want. You don't have to pay us. +3. You may not pretend that you wrote this software. If you use it in a + program, you must acknowledge somewhere in your documentation that + you've used the IJG code. + +In legalese: + +The authors make NO WARRANTY or representation, either express or implied, +with respect to this software, its quality, accuracy, merchantability, or +fitness for a particular purpose. This software is provided "AS IS", and you, +its user, assume the entire risk as to its quality and accuracy. + +This software is copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding. +All Rights Reserved except as specified below. + +Permission is hereby granted to use, copy, modify, and distribute this +software (or portions thereof) for any purpose, without fee, subject to these +conditions: +(1) If any part of the source code for this software is distributed, then this +README file must be included, with this copyright and no-warranty notice +unaltered; and any additions, deletions, or changes to the original files +must be clearly indicated in accompanying documentation. +(2) If only executable code is distributed, then the accompanying +documentation must state that "this software is based in part on the work of +the Independent JPEG Group". +(3) Permission for use of this software is granted only if the user accepts +full responsibility for any undesirable consequences; the authors accept +NO LIABILITY for damages of any kind. + +These conditions apply to any software derived from or based on the IJG code, +not just to the unmodified library. If you use our work, you ought to +acknowledge us. + +Permission is NOT granted for the use of any IJG author's name or company name +in advertising or publicity relating to this software or products derived from +it. This software may be referred to only as "the Independent JPEG Group's +software". + +We specifically permit and encourage the use of this software as the basis of +commercial products, provided that all warranty or liability claims are +assumed by the product vendor. + + +REFERENCES +========== + +We recommend reading one or more of these references before trying to +understand the innards of the JPEG software. + +The best short technical introduction to the JPEG compression algorithm is + Wallace, Gregory K. "The JPEG Still Picture Compression Standard", + Communications of the ACM, April 1991 (vol. 34 no. 4), pp. 30-44. +(Adjacent articles in that issue discuss MPEG motion picture compression, +applications of JPEG, and related topics.) If you don't have the CACM issue +handy, a PDF file containing a revised version of Wallace's article is +available at http://www.ijg.org/files/Wallace.JPEG.pdf. The file (actually +a preprint for an article that appeared in IEEE Trans. Consumer Electronics) +omits the sample images that appeared in CACM, but it includes corrections +and some added material. Note: the Wallace article is copyright ACM and IEEE, +and it may not be used for commercial purposes. + +A somewhat less technical, more leisurely introduction to JPEG can be found in +"The Data Compression Book" by Mark Nelson and Jean-loup Gailly, published by +M&T Books (New York), 2nd ed. 1996, ISBN 1-55851-434-1. This book provides +good explanations and example C code for a multitude of compression methods +including JPEG. It is an excellent source if you are comfortable reading C +code but don't know much about data compression in general. The book's JPEG +sample code is far from industrial-strength, but when you are ready to look +at a full implementation, you've got one here... + +The best currently available description of JPEG is the textbook "JPEG Still +Image Data Compression Standard" by William B. Pennebaker and Joan L. +Mitchell, published by Van Nostrand Reinhold, 1993, ISBN 0-442-01272-1. +Price US$59.95, 638 pp. The book includes the complete text of the ISO JPEG +standards (DIS 10918-1 and draft DIS 10918-2). + +The original JPEG standard is divided into two parts, Part 1 being the actual +specification, while Part 2 covers compliance testing methods. Part 1 is +titled "Digital Compression and Coding of Continuous-tone Still Images, +Part 1: Requirements and guidelines" and has document numbers ISO/IEC IS +10918-1, ITU-T T.81. Part 2 is titled "Digital Compression and Coding of +Continuous-tone Still Images, Part 2: Compliance testing" and has document +numbers ISO/IEC IS 10918-2, ITU-T T.83. + +The JPEG standard does not specify all details of an interchangeable file +format. For the omitted details, we follow the "JFIF" conventions, revision +1.02. JFIF version 1 has been adopted as ISO/IEC 10918-5 (05/2013) and +Recommendation ITU-T T.871 (05/2011): Information technology - Digital +compression and coding of continuous-tone still images: JPEG File Interchange +Format (JFIF). It is available as a free download in PDF file format from +https://www.iso.org/standard/54989.html and http://www.itu.int/rec/T-REC-T.871. +A PDF file of the older JFIF 1.02 specification is available at +http://www.w3.org/Graphics/JPEG/jfif3.pdf. + +The TIFF 6.0 file format specification can be obtained from +http://mirrors.ctan.org/graphics/tiff/TIFF6.ps.gz. The JPEG incorporation +scheme found in the TIFF 6.0 spec of 3-June-92 has a number of serious +problems. IJG does not recommend use of the TIFF 6.0 design (TIFF Compression +tag 6). Instead, we recommend the JPEG design proposed by TIFF Technical Note +#2 (Compression tag 7). Copies of this Note can be obtained from +http://www.ijg.org/files/. It is expected that the next revision +of the TIFF spec will replace the 6.0 JPEG design with the Note's design. +Although IJG's own code does not support TIFF/JPEG, the free libtiff library +uses our library to implement TIFF/JPEG per the Note. + + +ARCHIVE LOCATIONS +================= + +The "official" archive site for this software is www.ijg.org. +The most recent released version can always be found there in +directory "files". + +The JPEG FAQ (Frequently Asked Questions) article is a source of some +general information about JPEG. It is available at +http://www.faqs.org/faqs/jpeg-faq. + + +FILE FORMAT COMPATIBILITY +========================= + +This software implements ITU T.81 | ISO/IEC 10918 with some extensions from +ITU T.871 | ISO/IEC 10918-5 (JPEG File Interchange Format-- see REFERENCES). +Informally, the term "JPEG image" or "JPEG file" most often refers to JFIF or +a subset thereof, but there are other formats containing the name "JPEG" that +are incompatible with the original JPEG standard or with JFIF (for instance, +JPEG 2000 and JPEG XR). This software therefore does not support these +formats. Indeed, one of the original reasons for developing this free software +was to help force convergence on a common, interoperable format standard for +JPEG files. + +JFIF is a minimal or "low end" representation. TIFF/JPEG (TIFF revision 6.0 as +modified by TIFF Technical Note #2) can be used for "high end" applications +that need to record a lot of additional data about an image. + + +TO DO +===== + +Please send bug reports, offers of help, etc. to jpeg-info@jpegclub.org. diff --git a/src/spice2x/external/libjpeg-turbo/jaricom.c b/src/spice2x/external/libjpeg-turbo/jaricom.c new file mode 100644 index 0000000..215640c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jaricom.c @@ -0,0 +1,157 @@ +/* + * jaricom.c + * + * This file was part of the Independent JPEG Group's software: + * Developed 1997-2009 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2018, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains probability estimation tables for common use in + * arithmetic entropy encoding and decoding routines. + * + * This data represents Table D.2 in + * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994 and Table 24 in + * Recommendation ITU-T T.82 (1993) | ISO/IEC 11544:1993. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" + +/* The following #define specifies the packing of the four components + * into the compact JLONG representation. + * Note that this formula must match the actual arithmetic encoder + * and decoder implementation. The implementation has to be changed + * if this formula is changed. + * The current organization is leaned on Markus Kuhn's JBIG + * implementation (jbig_tab.c). + */ + +#define V(i, a, b, c, d) \ + (((JLONG)a << 16) | ((JLONG)c << 8) | ((JLONG)d << 7) | b) + +const JLONG jpeg_aritab[113 + 1] = { +/* + * Index, Qe_Value, Next_Index_LPS, Next_Index_MPS, Switch_MPS + */ + V( 0, 0x5a1d, 1, 1, 1 ), + V( 1, 0x2586, 14, 2, 0 ), + V( 2, 0x1114, 16, 3, 0 ), + V( 3, 0x080b, 18, 4, 0 ), + V( 4, 0x03d8, 20, 5, 0 ), + V( 5, 0x01da, 23, 6, 0 ), + V( 6, 0x00e5, 25, 7, 0 ), + V( 7, 0x006f, 28, 8, 0 ), + V( 8, 0x0036, 30, 9, 0 ), + V( 9, 0x001a, 33, 10, 0 ), + V( 10, 0x000d, 35, 11, 0 ), + V( 11, 0x0006, 9, 12, 0 ), + V( 12, 0x0003, 10, 13, 0 ), + V( 13, 0x0001, 12, 13, 0 ), + V( 14, 0x5a7f, 15, 15, 1 ), + V( 15, 0x3f25, 36, 16, 0 ), + V( 16, 0x2cf2, 38, 17, 0 ), + V( 17, 0x207c, 39, 18, 0 ), + V( 18, 0x17b9, 40, 19, 0 ), + V( 19, 0x1182, 42, 20, 0 ), + V( 20, 0x0cef, 43, 21, 0 ), + V( 21, 0x09a1, 45, 22, 0 ), + V( 22, 0x072f, 46, 23, 0 ), + V( 23, 0x055c, 48, 24, 0 ), + V( 24, 0x0406, 49, 25, 0 ), + V( 25, 0x0303, 51, 26, 0 ), + V( 26, 0x0240, 52, 27, 0 ), + V( 27, 0x01b1, 54, 28, 0 ), + V( 28, 0x0144, 56, 29, 0 ), + V( 29, 0x00f5, 57, 30, 0 ), + V( 30, 0x00b7, 59, 31, 0 ), + V( 31, 0x008a, 60, 32, 0 ), + V( 32, 0x0068, 62, 33, 0 ), + V( 33, 0x004e, 63, 34, 0 ), + V( 34, 0x003b, 32, 35, 0 ), + V( 35, 0x002c, 33, 9, 0 ), + V( 36, 0x5ae1, 37, 37, 1 ), + V( 37, 0x484c, 64, 38, 0 ), + V( 38, 0x3a0d, 65, 39, 0 ), + V( 39, 0x2ef1, 67, 40, 0 ), + V( 40, 0x261f, 68, 41, 0 ), + V( 41, 0x1f33, 69, 42, 0 ), + V( 42, 0x19a8, 70, 43, 0 ), + V( 43, 0x1518, 72, 44, 0 ), + V( 44, 0x1177, 73, 45, 0 ), + V( 45, 0x0e74, 74, 46, 0 ), + V( 46, 0x0bfb, 75, 47, 0 ), + V( 47, 0x09f8, 77, 48, 0 ), + V( 48, 0x0861, 78, 49, 0 ), + V( 49, 0x0706, 79, 50, 0 ), + V( 50, 0x05cd, 48, 51, 0 ), + V( 51, 0x04de, 50, 52, 0 ), + V( 52, 0x040f, 50, 53, 0 ), + V( 53, 0x0363, 51, 54, 0 ), + V( 54, 0x02d4, 52, 55, 0 ), + V( 55, 0x025c, 53, 56, 0 ), + V( 56, 0x01f8, 54, 57, 0 ), + V( 57, 0x01a4, 55, 58, 0 ), + V( 58, 0x0160, 56, 59, 0 ), + V( 59, 0x0125, 57, 60, 0 ), + V( 60, 0x00f6, 58, 61, 0 ), + V( 61, 0x00cb, 59, 62, 0 ), + V( 62, 0x00ab, 61, 63, 0 ), + V( 63, 0x008f, 61, 32, 0 ), + V( 64, 0x5b12, 65, 65, 1 ), + V( 65, 0x4d04, 80, 66, 0 ), + V( 66, 0x412c, 81, 67, 0 ), + V( 67, 0x37d8, 82, 68, 0 ), + V( 68, 0x2fe8, 83, 69, 0 ), + V( 69, 0x293c, 84, 70, 0 ), + V( 70, 0x2379, 86, 71, 0 ), + V( 71, 0x1edf, 87, 72, 0 ), + V( 72, 0x1aa9, 87, 73, 0 ), + V( 73, 0x174e, 72, 74, 0 ), + V( 74, 0x1424, 72, 75, 0 ), + V( 75, 0x119c, 74, 76, 0 ), + V( 76, 0x0f6b, 74, 77, 0 ), + V( 77, 0x0d51, 75, 78, 0 ), + V( 78, 0x0bb6, 77, 79, 0 ), + V( 79, 0x0a40, 77, 48, 0 ), + V( 80, 0x5832, 80, 81, 1 ), + V( 81, 0x4d1c, 88, 82, 0 ), + V( 82, 0x438e, 89, 83, 0 ), + V( 83, 0x3bdd, 90, 84, 0 ), + V( 84, 0x34ee, 91, 85, 0 ), + V( 85, 0x2eae, 92, 86, 0 ), + V( 86, 0x299a, 93, 87, 0 ), + V( 87, 0x2516, 86, 71, 0 ), + V( 88, 0x5570, 88, 89, 1 ), + V( 89, 0x4ca9, 95, 90, 0 ), + V( 90, 0x44d9, 96, 91, 0 ), + V( 91, 0x3e22, 97, 92, 0 ), + V( 92, 0x3824, 99, 93, 0 ), + V( 93, 0x32b4, 99, 94, 0 ), + V( 94, 0x2e17, 93, 86, 0 ), + V( 95, 0x56a8, 95, 96, 1 ), + V( 96, 0x4f46, 101, 97, 0 ), + V( 97, 0x47e5, 102, 98, 0 ), + V( 98, 0x41cf, 103, 99, 0 ), + V( 99, 0x3c3d, 104, 100, 0 ), + V( 100, 0x375e, 99, 93, 0 ), + V( 101, 0x5231, 105, 102, 0 ), + V( 102, 0x4c0f, 106, 103, 0 ), + V( 103, 0x4639, 107, 104, 0 ), + V( 104, 0x415e, 103, 99, 0 ), + V( 105, 0x5627, 105, 106, 1 ), + V( 106, 0x50e7, 108, 107, 0 ), + V( 107, 0x4b85, 109, 103, 0 ), + V( 108, 0x5597, 110, 109, 0 ), + V( 109, 0x504f, 111, 107, 0 ), + V( 110, 0x5a10, 110, 111, 1 ), + V( 111, 0x5522, 112, 109, 0 ), + V( 112, 0x59eb, 112, 111, 1 ), +/* + * This last entry is used for fixed probability estimate of 0.5 + * as recommended in Section 10.3 Table 5 of ITU-T Rec. T.851. + */ + V( 113, 0x5a1d, 113, 113, 0 ) +}; diff --git a/src/spice2x/external/libjpeg-turbo/jcapimin.c b/src/spice2x/external/libjpeg-turbo/jcapimin.c new file mode 100644 index 0000000..cbb3d13 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcapimin.c @@ -0,0 +1,318 @@ +/* + * jcapimin.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1998, Thomas G. Lane. + * Modified 2003-2010 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains application interface code for the compression half + * of the JPEG library. These are the "minimum" API routines that may be + * needed in either the normal full-compression case or the transcoding-only + * case. + * + * Most of the routines intended to be called directly by an application + * are in this file or in jcapistd.c. But also see jcparam.c for + * parameter-setup helper routines, jcomapi.c for routines shared by + * compression and decompression, and jctrans.c for the transcoding case. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jcmaster.h" + + +/* + * Initialization of a JPEG compression object. + * The error manager must already be set up (in case memory manager fails). + */ + +GLOBAL(void) +jpeg_CreateCompress(j_compress_ptr cinfo, int version, size_t structsize) +{ + int i; + + /* Guard against version mismatches between library and caller. */ + cinfo->mem = NULL; /* so jpeg_destroy knows mem mgr not called */ + if (version != JPEG_LIB_VERSION) + ERREXIT2(cinfo, JERR_BAD_LIB_VERSION, JPEG_LIB_VERSION, version); + if (structsize != sizeof(struct jpeg_compress_struct)) + ERREXIT2(cinfo, JERR_BAD_STRUCT_SIZE, + (int)sizeof(struct jpeg_compress_struct), (int)structsize); + + /* For debugging purposes, we zero the whole master structure. + * But the application has already set the err pointer, and may have set + * client_data, so we have to save and restore those fields. + * Note: if application hasn't set client_data, tools like Purify may + * complain here. + */ + { + struct jpeg_error_mgr *err = cinfo->err; + void *client_data = cinfo->client_data; /* ignore Purify complaint here */ + memset(cinfo, 0, sizeof(struct jpeg_compress_struct)); + cinfo->err = err; + cinfo->client_data = client_data; + } + cinfo->is_decompressor = FALSE; + + /* Initialize a memory manager instance for this object */ + jinit_memory_mgr((j_common_ptr)cinfo); + + /* Zero out pointers to permanent structures. */ + cinfo->progress = NULL; + cinfo->dest = NULL; + + cinfo->comp_info = NULL; + + for (i = 0; i < NUM_QUANT_TBLS; i++) { + cinfo->quant_tbl_ptrs[i] = NULL; +#if JPEG_LIB_VERSION >= 70 + cinfo->q_scale_factor[i] = 100; +#endif + } + + for (i = 0; i < NUM_HUFF_TBLS; i++) { + cinfo->dc_huff_tbl_ptrs[i] = NULL; + cinfo->ac_huff_tbl_ptrs[i] = NULL; + } + +#if JPEG_LIB_VERSION >= 80 + /* Must do it here for emit_dqt in case jpeg_write_tables is used */ + cinfo->block_size = DCTSIZE; + cinfo->natural_order = jpeg_natural_order; + cinfo->lim_Se = DCTSIZE2 - 1; +#endif + + cinfo->script_space = NULL; + + cinfo->input_gamma = 1.0; /* in case application forgets */ + + cinfo->data_precision = BITS_IN_JSAMPLE; + + /* OK, I'm ready */ + cinfo->global_state = CSTATE_START; + + /* The master struct is used to store extension parameters, so we allocate it + * here. + */ + cinfo->master = (struct jpeg_comp_master *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(my_comp_master)); + memset(cinfo->master, 0, sizeof(my_comp_master)); +} + + +/* + * Destruction of a JPEG compression object + */ + +GLOBAL(void) +jpeg_destroy_compress(j_compress_ptr cinfo) +{ + jpeg_destroy((j_common_ptr)cinfo); /* use common routine */ +} + + +/* + * Abort processing of a JPEG compression operation, + * but don't destroy the object itself. + */ + +GLOBAL(void) +jpeg_abort_compress(j_compress_ptr cinfo) +{ + jpeg_abort((j_common_ptr)cinfo); /* use common routine */ +} + + +/* + * Forcibly suppress or un-suppress all quantization and Huffman tables. + * Marks all currently defined tables as already written (if suppress) + * or not written (if !suppress). This will control whether they get emitted + * by a subsequent jpeg_start_compress call. + * + * This routine is exported for use by applications that want to produce + * abbreviated JPEG datastreams. It logically belongs in jcparam.c, but + * since it is called by jpeg_start_compress, we put it here --- otherwise + * jcparam.o would be linked whether the application used it or not. + */ + +GLOBAL(void) +jpeg_suppress_tables(j_compress_ptr cinfo, boolean suppress) +{ + int i; + JQUANT_TBL *qtbl; + JHUFF_TBL *htbl; + + for (i = 0; i < NUM_QUANT_TBLS; i++) { + if ((qtbl = cinfo->quant_tbl_ptrs[i]) != NULL) + qtbl->sent_table = suppress; + } + + for (i = 0; i < NUM_HUFF_TBLS; i++) { + if ((htbl = cinfo->dc_huff_tbl_ptrs[i]) != NULL) + htbl->sent_table = suppress; + if ((htbl = cinfo->ac_huff_tbl_ptrs[i]) != NULL) + htbl->sent_table = suppress; + } +} + + +/* + * Finish JPEG compression. + * + * If a multipass operating mode was selected, this may do a great deal of + * work including most of the actual output. + */ + +GLOBAL(void) +jpeg_finish_compress(j_compress_ptr cinfo) +{ + JDIMENSION iMCU_row; + + if (cinfo->global_state == CSTATE_SCANNING || + cinfo->global_state == CSTATE_RAW_OK) { + /* Terminate first pass */ + if (cinfo->next_scanline < cinfo->image_height) + ERREXIT(cinfo, JERR_TOO_LITTLE_DATA); + (*cinfo->master->finish_pass) (cinfo); + } else if (cinfo->global_state != CSTATE_WRCOEFS) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + /* Perform any remaining passes */ + while (!cinfo->master->is_last_pass) { + (*cinfo->master->prepare_for_pass) (cinfo); + for (iMCU_row = 0; iMCU_row < cinfo->total_iMCU_rows; iMCU_row++) { + if (cinfo->progress != NULL) { + cinfo->progress->pass_counter = (long)iMCU_row; + cinfo->progress->pass_limit = (long)cinfo->total_iMCU_rows; + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + } + /* We bypass the main controller and invoke coef controller directly; + * all work is being done from the coefficient buffer. + */ + if (cinfo->data_precision == 16) { +#ifdef C_LOSSLESS_SUPPORTED + if (!(*cinfo->coef->compress_data_16) (cinfo, (J16SAMPIMAGE)NULL)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); +#endif + } else if (cinfo->data_precision == 12) { + if (!(*cinfo->coef->compress_data_12) (cinfo, (J12SAMPIMAGE)NULL)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); + } else { + if (!(*cinfo->coef->compress_data) (cinfo, (JSAMPIMAGE)NULL)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); + } + } + (*cinfo->master->finish_pass) (cinfo); + } + /* Write EOI, do final cleanup */ + (*cinfo->marker->write_file_trailer) (cinfo); + (*cinfo->dest->term_destination) (cinfo); + /* We can use jpeg_abort to release memory and reset global_state */ + jpeg_abort((j_common_ptr)cinfo); +} + + +/* + * Write a special marker. + * This is only recommended for writing COM or APPn markers. + * Must be called after jpeg_start_compress() and before + * first call to jpeg_write_scanlines() or jpeg_write_raw_data(). + */ + +GLOBAL(void) +jpeg_write_marker(j_compress_ptr cinfo, int marker, const JOCTET *dataptr, + unsigned int datalen) +{ + void (*write_marker_byte) (j_compress_ptr info, int val); + + if (cinfo->next_scanline != 0 || + (cinfo->global_state != CSTATE_SCANNING && + cinfo->global_state != CSTATE_RAW_OK && + cinfo->global_state != CSTATE_WRCOEFS)) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + (*cinfo->marker->write_marker_header) (cinfo, marker, datalen); + write_marker_byte = cinfo->marker->write_marker_byte; /* copy for speed */ + while (datalen--) { + (*write_marker_byte) (cinfo, *dataptr); + dataptr++; + } +} + +/* Same, but piecemeal. */ + +GLOBAL(void) +jpeg_write_m_header(j_compress_ptr cinfo, int marker, unsigned int datalen) +{ + if (cinfo->next_scanline != 0 || + (cinfo->global_state != CSTATE_SCANNING && + cinfo->global_state != CSTATE_RAW_OK && + cinfo->global_state != CSTATE_WRCOEFS)) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + (*cinfo->marker->write_marker_header) (cinfo, marker, datalen); +} + +GLOBAL(void) +jpeg_write_m_byte(j_compress_ptr cinfo, int val) +{ + (*cinfo->marker->write_marker_byte) (cinfo, val); +} + + +/* + * Alternate compression function: just write an abbreviated table file. + * Before calling this, all parameters and a data destination must be set up. + * + * To produce a pair of files containing abbreviated tables and abbreviated + * image data, one would proceed as follows: + * + * initialize JPEG object + * set JPEG parameters + * set destination to table file + * jpeg_write_tables(cinfo); + * set destination to image file + * jpeg_start_compress(cinfo, FALSE); + * write data... + * jpeg_finish_compress(cinfo); + * + * jpeg_write_tables has the side effect of marking all tables written + * (same as jpeg_suppress_tables(..., TRUE)). Thus a subsequent start_compress + * will not re-emit the tables unless it is passed write_all_tables=TRUE. + */ + +GLOBAL(void) +jpeg_write_tables(j_compress_ptr cinfo) +{ + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + /* (Re)initialize error mgr and destination modules */ + (*cinfo->err->reset_error_mgr) ((j_common_ptr)cinfo); + (*cinfo->dest->init_destination) (cinfo); + /* Initialize the marker writer ... bit of a crock to do it here. */ + jinit_marker_writer(cinfo); + /* Write them tables! */ + (*cinfo->marker->write_tables_only) (cinfo); + /* And clean up. */ + (*cinfo->dest->term_destination) (cinfo); + /* + * In library releases up through v6a, we called jpeg_abort() here to free + * any working memory allocated by the destination manager and marker + * writer. Some applications had a problem with that: they allocated space + * of their own from the library memory manager, and didn't want it to go + * away during write_tables. So now we do nothing. This will cause a + * memory leak if an app calls write_tables repeatedly without doing a full + * compression cycle or otherwise resetting the JPEG object. However, that + * seems less bad than unexpectedly freeing memory in the normal case. + * An app that prefers the old behavior can call jpeg_abort for itself after + * each call to jpeg_write_tables(). + */ +} diff --git a/src/spice2x/external/libjpeg-turbo/jcapistd.c b/src/spice2x/external/libjpeg-turbo/jcapistd.c new file mode 100644 index 0000000..2053028 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcapistd.c @@ -0,0 +1,187 @@ +/* + * jcapistd.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains application interface code for the compression half + * of the JPEG library. These are the "standard" API routines that are + * used in the normal full-compression case. They are not used by a + * transcoding-only application. Note that if an application links in + * jpeg_start_compress, it will end up linking in the entire compressor. + * We thus must separate this file from jcapimin.c to avoid linking the + * whole compression library into a transcoder. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE == 8 + +/* + * Compression initialization. + * Before calling this, all parameters and a data destination must be set up. + * + * We require a write_all_tables parameter as a failsafe check when writing + * multiple datastreams from the same compression object. Since prior runs + * will have left all the tables marked sent_table=TRUE, a subsequent run + * would emit an abbreviated stream (no tables) by default. This may be what + * is wanted, but for safety's sake it should not be the default behavior: + * programmers should have to make a deliberate choice to emit abbreviated + * images. Therefore the documentation and examples should encourage people + * to pass write_all_tables=TRUE; then it will take active thought to do the + * wrong thing. + */ + +GLOBAL(void) +jpeg_start_compress(j_compress_ptr cinfo, boolean write_all_tables) +{ + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + if (write_all_tables) + jpeg_suppress_tables(cinfo, FALSE); /* mark all tables to be written */ + + /* (Re)initialize error mgr and destination modules */ + (*cinfo->err->reset_error_mgr) ((j_common_ptr)cinfo); + (*cinfo->dest->init_destination) (cinfo); + /* Perform master selection of active modules */ + jinit_compress_master(cinfo); + /* Set up for the first pass */ + (*cinfo->master->prepare_for_pass) (cinfo); + /* Ready for application to drive first pass through _jpeg_write_scanlines + * or _jpeg_write_raw_data. + */ + cinfo->next_scanline = 0; + cinfo->global_state = (cinfo->raw_data_in ? CSTATE_RAW_OK : CSTATE_SCANNING); +} + +#endif + + +/* + * Write some scanlines of data to the JPEG compressor. + * + * The return value will be the number of lines actually written. + * This should be less than the supplied num_lines only in case that + * the data destination module has requested suspension of the compressor, + * or if more than image_height scanlines are passed in. + * + * Note: we warn about excess calls to _jpeg_write_scanlines() since + * this likely signals an application programmer error. However, + * excess scanlines passed in the last valid call are *silently* ignored, + * so that the application need not adjust num_lines for end-of-image + * when using a multiple-scanline buffer. + */ + +GLOBAL(JDIMENSION) +_jpeg_write_scanlines(j_compress_ptr cinfo, _JSAMPARRAY scanlines, + JDIMENSION num_lines) +{ +#if BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) + JDIMENSION row_ctr, rows_left; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (cinfo->global_state != CSTATE_SCANNING) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + if (cinfo->next_scanline >= cinfo->image_height) + WARNMS(cinfo, JWRN_TOO_MUCH_DATA); + + /* Call progress monitor hook if present */ + if (cinfo->progress != NULL) { + cinfo->progress->pass_counter = (long)cinfo->next_scanline; + cinfo->progress->pass_limit = (long)cinfo->image_height; + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + } + + /* Give master control module another chance if this is first call to + * _jpeg_write_scanlines. This lets output of the frame/scan headers be + * delayed so that application can write COM, etc, markers between + * jpeg_start_compress and _jpeg_write_scanlines. + */ + if (cinfo->master->call_pass_startup) + (*cinfo->master->pass_startup) (cinfo); + + /* Ignore any extra scanlines at bottom of image. */ + rows_left = cinfo->image_height - cinfo->next_scanline; + if (num_lines > rows_left) + num_lines = rows_left; + + row_ctr = 0; + (*cinfo->main->_process_data) (cinfo, scanlines, &row_ctr, num_lines); + cinfo->next_scanline += row_ctr; + return row_ctr; +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + return 0; +#endif +} + + +#if BITS_IN_JSAMPLE != 16 + +/* + * Alternate entry point to write raw data. + * Processes exactly one iMCU row per call, unless suspended. + */ + +GLOBAL(JDIMENSION) +_jpeg_write_raw_data(j_compress_ptr cinfo, _JSAMPIMAGE data, + JDIMENSION num_lines) +{ + JDIMENSION lines_per_iMCU_row; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + if (cinfo->global_state != CSTATE_RAW_OK) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + if (cinfo->next_scanline >= cinfo->image_height) { + WARNMS(cinfo, JWRN_TOO_MUCH_DATA); + return 0; + } + + /* Call progress monitor hook if present */ + if (cinfo->progress != NULL) { + cinfo->progress->pass_counter = (long)cinfo->next_scanline; + cinfo->progress->pass_limit = (long)cinfo->image_height; + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + } + + /* Give master control module another chance if this is first call to + * _jpeg_write_raw_data. This lets output of the frame/scan headers be + * delayed so that application can write COM, etc, markers between + * jpeg_start_compress and _jpeg_write_raw_data. + */ + if (cinfo->master->call_pass_startup) + (*cinfo->master->pass_startup) (cinfo); + + /* Verify that at least one iMCU row has been passed. */ + lines_per_iMCU_row = cinfo->max_v_samp_factor * DCTSIZE; + if (num_lines < lines_per_iMCU_row) + ERREXIT(cinfo, JERR_BUFFER_SIZE); + + /* Directly compress the row. */ + if (!(*cinfo->coef->_compress_data) (cinfo, data)) { + /* If compressor did not consume the whole row, suspend processing. */ + return 0; + } + + /* OK, we processed one iMCU row. */ + cinfo->next_scanline += lines_per_iMCU_row; + return lines_per_iMCU_row; +} + +#endif /* BITS_IN_JSAMPLE != 16 */ diff --git a/src/spice2x/external/libjpeg-turbo/jcarith.c b/src/spice2x/external/libjpeg-turbo/jcarith.c new file mode 100644 index 0000000..b172052 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcarith.c @@ -0,0 +1,932 @@ +/* + * jcarith.c + * + * This file was part of the Independent JPEG Group's software: + * Developed 1997-2009 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2018, 2021-2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains portable arithmetic entropy encoding routines for JPEG + * (implementing Recommendation ITU-T T.81 | ISO/IEC 10918-1). + * + * Both sequential and progressive modes are supported in this single module. + * + * Suspension is not currently supported in this module. + * + * NOTE: All referenced figures are from + * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" + + +/* Expanded entropy encoder object for arithmetic encoding. */ + +typedef struct { + struct jpeg_entropy_encoder pub; /* public fields */ + + JLONG c; /* C register, base of coding interval, layout as in sec. D.1.3 */ + JLONG a; /* A register, normalized size of coding interval */ + JLONG sc; /* counter for stacked 0xFF values which might overflow */ + JLONG zc; /* counter for pending 0x00 output values which might * + * be discarded at the end ("Pacman" termination) */ + int ct; /* bit shift counter, determines when next byte will be written */ + int buffer; /* buffer for most recent output byte != 0xFF */ + + int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */ + int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */ + + unsigned int restarts_to_go; /* MCUs left in this restart interval */ + int next_restart_num; /* next restart number to write (0-7) */ + + /* Pointers to statistics areas (these workspaces have image lifespan) */ + unsigned char *dc_stats[NUM_ARITH_TBLS]; + unsigned char *ac_stats[NUM_ARITH_TBLS]; + + /* Statistics bin for coding with fixed probability 0.5 */ + unsigned char fixed_bin[4]; +} arith_entropy_encoder; + +typedef arith_entropy_encoder *arith_entropy_ptr; + +/* The following two definitions specify the allocation chunk size + * for the statistics area. + * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least + * 49 statistics bins for DC, and 245 statistics bins for AC coding. + * + * We use a compact representation with 1 byte per statistics bin, + * thus the numbers directly represent byte sizes. + * This 1 byte per statistics bin contains the meaning of the MPS + * (more probable symbol) in the highest bit (mask 0x80), and the + * index into the probability estimation state machine table + * in the lower bits (mask 0x7F). + */ + +#define DC_STAT_BINS 64 +#define AC_STAT_BINS 256 + +/* NOTE: Uncomment the following #define if you want to use the + * given formula for calculating the AC conditioning parameter Kx + * for spectral selection progressive coding in section G.1.3.2 + * of the spec (Kx = Kmin + SRL (8 + Se - Kmin) 4). + * Although the spec and P&M authors claim that this "has proven + * to give good results for 8 bit precision samples", I'm not + * convinced yet that this is really beneficial. + * Early tests gave only very marginal compression enhancements + * (a few - around 5 or so - bytes even for very large files), + * which would turn out rather negative if we'd suppress the + * DAC (Define Arithmetic Conditioning) marker segments for + * the default parameters in the future. + * Note that currently the marker writing module emits 12-byte + * DAC segments for a full-component scan in a color image. + * This is not worth worrying about IMHO. However, since the + * spec defines the default values to be used if the tables + * are omitted (unlike Huffman tables, which are required + * anyway), one might optimize this behaviour in the future, + * and then it would be disadvantageous to use custom tables if + * they don't provide sufficient gain to exceed the DAC size. + * + * On the other hand, I'd consider it as a reasonable result + * that the conditioning has no significant influence on the + * compression performance. This means that the basic + * statistical model is already rather stable. + * + * Thus, at the moment, we use the default conditioning values + * anyway, and do not use the custom formula. + * +#define CALCULATE_SPECTRAL_CONDITIONING + */ + +/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than JLONG. + * We assume that int right shift is unsigned if JLONG right shift is, + * which should be safe. + */ + +#ifdef RIGHT_SHIFT_IS_UNSIGNED +#define ISHIFT_TEMPS int ishift_temp; +#define IRIGHT_SHIFT(x, shft) \ + ((ishift_temp = (x)) < 0 ? \ + (ishift_temp >> (shft)) | ((~0) << (16 - (shft))) : \ + (ishift_temp >> (shft))) +#else +#define ISHIFT_TEMPS +#define IRIGHT_SHIFT(x, shft) ((x) >> (shft)) +#endif + + +LOCAL(void) +emit_byte(int val, j_compress_ptr cinfo) +/* Write next output byte; we do not support suspension in this module. */ +{ + struct jpeg_destination_mgr *dest = cinfo->dest; + + *dest->next_output_byte++ = (JOCTET)val; + if (--dest->free_in_buffer == 0) + if (!(*dest->empty_output_buffer) (cinfo)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); +} + + +/* + * Finish up at the end of an arithmetic-compressed scan. + */ + +METHODDEF(void) +finish_pass(j_compress_ptr cinfo) +{ + arith_entropy_ptr e = (arith_entropy_ptr)cinfo->entropy; + JLONG temp; + + /* Section D.1.8: Termination of encoding */ + + /* Find the e->c in the coding interval with the largest + * number of trailing zero bits */ + if ((temp = (e->a - 1 + e->c) & 0xFFFF0000UL) < e->c) + e->c = temp + 0x8000L; + else + e->c = temp; + /* Send remaining bytes to output */ + e->c <<= e->ct; + if (e->c & 0xF8000000UL) { + /* One final overflow has to be handled */ + if (e->buffer >= 0) { + if (e->zc) + do emit_byte(0x00, cinfo); + while (--e->zc); + emit_byte(e->buffer + 1, cinfo); + if (e->buffer + 1 == 0xFF) + emit_byte(0x00, cinfo); + } + e->zc += e->sc; /* carry-over converts stacked 0xFF bytes to 0x00 */ + e->sc = 0; + } else { + if (e->buffer == 0) + ++e->zc; + else if (e->buffer >= 0) { + if (e->zc) + do emit_byte(0x00, cinfo); + while (--e->zc); + emit_byte(e->buffer, cinfo); + } + if (e->sc) { + if (e->zc) + do emit_byte(0x00, cinfo); + while (--e->zc); + do { + emit_byte(0xFF, cinfo); + emit_byte(0x00, cinfo); + } while (--e->sc); + } + } + /* Output final bytes only if they are not 0x00 */ + if (e->c & 0x7FFF800L) { + if (e->zc) /* output final pending zero bytes */ + do emit_byte(0x00, cinfo); + while (--e->zc); + emit_byte((e->c >> 19) & 0xFF, cinfo); + if (((e->c >> 19) & 0xFF) == 0xFF) + emit_byte(0x00, cinfo); + if (e->c & 0x7F800L) { + emit_byte((e->c >> 11) & 0xFF, cinfo); + if (((e->c >> 11) & 0xFF) == 0xFF) + emit_byte(0x00, cinfo); + } + } +} + + +/* + * The core arithmetic encoding routine (common in JPEG and JBIG). + * This needs to go as fast as possible. + * Machine-dependent optimization facilities + * are not utilized in this portable implementation. + * However, this code should be fairly efficient and + * may be a good base for further optimizations anyway. + * + * Parameter 'val' to be encoded may be 0 or 1 (binary decision). + * + * Note: I've added full "Pacman" termination support to the + * byte output routines, which is equivalent to the optional + * Discard_final_zeros procedure (Figure D.15) in the spec. + * Thus, we always produce the shortest possible output + * stream compliant to the spec (no trailing zero bytes, + * except for FF stuffing). + * + * I've also introduced a new scheme for accessing + * the probability estimation state machine table, + * derived from Markus Kuhn's JBIG implementation. + */ + +LOCAL(void) +arith_encode(j_compress_ptr cinfo, unsigned char *st, int val) +{ + register arith_entropy_ptr e = (arith_entropy_ptr)cinfo->entropy; + register unsigned char nl, nm; + register JLONG qe, temp; + register int sv; + + /* Fetch values from our compact representation of Table D.2: + * Qe values and probability estimation state machine + */ + sv = *st; + qe = jpeg_aritab[sv & 0x7F]; /* => Qe_Value */ + nl = qe & 0xFF; qe >>= 8; /* Next_Index_LPS + Switch_MPS */ + nm = qe & 0xFF; qe >>= 8; /* Next_Index_MPS */ + + /* Encode & estimation procedures per sections D.1.4 & D.1.5 */ + e->a -= qe; + if (val != (sv >> 7)) { + /* Encode the less probable symbol */ + if (e->a >= qe) { + /* If the interval size (qe) for the less probable symbol (LPS) + * is larger than the interval size for the MPS, then exchange + * the two symbols for coding efficiency, otherwise code the LPS + * as usual: */ + e->c += e->a; + e->a = qe; + } + *st = (sv & 0x80) ^ nl; /* Estimate_after_LPS */ + } else { + /* Encode the more probable symbol */ + if (e->a >= 0x8000L) + return; /* A >= 0x8000 -> ready, no renormalization required */ + if (e->a < qe) { + /* If the interval size (qe) for the less probable symbol (LPS) + * is larger than the interval size for the MPS, then exchange + * the two symbols for coding efficiency: */ + e->c += e->a; + e->a = qe; + } + *st = (sv & 0x80) ^ nm; /* Estimate_after_MPS */ + } + + /* Renormalization & data output per section D.1.6 */ + do { + e->a <<= 1; + e->c <<= 1; + if (--e->ct == 0) { + /* Another byte is ready for output */ + temp = e->c >> 19; + if (temp > 0xFF) { + /* Handle overflow over all stacked 0xFF bytes */ + if (e->buffer >= 0) { + if (e->zc) + do emit_byte(0x00, cinfo); + while (--e->zc); + emit_byte(e->buffer + 1, cinfo); + if (e->buffer + 1 == 0xFF) + emit_byte(0x00, cinfo); + } + e->zc += e->sc; /* carry-over converts stacked 0xFF bytes to 0x00 */ + e->sc = 0; + /* Note: The 3 spacer bits in the C register guarantee + * that the new buffer byte can't be 0xFF here + * (see page 160 in the P&M JPEG book). */ + e->buffer = temp & 0xFF; /* new output byte, might overflow later */ + } else if (temp == 0xFF) { + ++e->sc; /* stack 0xFF byte (which might overflow later) */ + } else { + /* Output all stacked 0xFF bytes, they will not overflow any more */ + if (e->buffer == 0) + ++e->zc; + else if (e->buffer >= 0) { + if (e->zc) + do emit_byte(0x00, cinfo); + while (--e->zc); + emit_byte(e->buffer, cinfo); + } + if (e->sc) { + if (e->zc) + do emit_byte(0x00, cinfo); + while (--e->zc); + do { + emit_byte(0xFF, cinfo); + emit_byte(0x00, cinfo); + } while (--e->sc); + } + e->buffer = temp & 0xFF; /* new output byte (can still overflow) */ + } + e->c &= 0x7FFFFL; + e->ct += 8; + } + } while (e->a < 0x8000L); +} + + +/* + * Emit a restart marker & resynchronize predictions. + */ + +LOCAL(void) +emit_restart(j_compress_ptr cinfo, int restart_num) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + int ci; + jpeg_component_info *compptr; + + finish_pass(cinfo); + + emit_byte(0xFF, cinfo); + emit_byte(JPEG_RST0 + restart_num, cinfo); + + /* Re-initialize statistics areas */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* DC needs no table for refinement scan */ + if (cinfo->progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) { + memset(entropy->dc_stats[compptr->dc_tbl_no], 0, DC_STAT_BINS); + /* Reset DC predictions to 0 */ + entropy->last_dc_val[ci] = 0; + entropy->dc_context[ci] = 0; + } + /* AC needs no table when not present */ + if (cinfo->progressive_mode == 0 || cinfo->Se) { + memset(entropy->ac_stats[compptr->ac_tbl_no], 0, AC_STAT_BINS); + } + } + + /* Reset arithmetic encoding variables */ + entropy->c = 0; + entropy->a = 0x10000L; + entropy->sc = 0; + entropy->zc = 0; + entropy->ct = 11; + entropy->buffer = -1; /* empty */ +} + + +/* + * MCU encoding for DC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +METHODDEF(boolean) +encode_mcu_DC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + JBLOCKROW block; + unsigned char *st; + int blkn, ci, tbl; + int v, v2, m; + ISHIFT_TEMPS + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + emit_restart(cinfo, entropy->next_restart_num); + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + /* Encode the MCU data blocks */ + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data[blkn]; + ci = cinfo->MCU_membership[blkn]; + tbl = cinfo->cur_comp_info[ci]->dc_tbl_no; + + /* Compute the DC value after the required point transform by Al. + * This is simply an arithmetic right shift. + */ + m = IRIGHT_SHIFT((int)((*block)[0]), cinfo->Al); + + /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */ + + /* Table F.4: Point to statistics bin S0 for DC coefficient coding */ + st = entropy->dc_stats[tbl] + entropy->dc_context[ci]; + + /* Figure F.4: Encode_DC_DIFF */ + if ((v = m - entropy->last_dc_val[ci]) == 0) { + arith_encode(cinfo, st, 0); + entropy->dc_context[ci] = 0; /* zero diff category */ + } else { + entropy->last_dc_val[ci] = m; + arith_encode(cinfo, st, 1); + /* Figure F.6: Encoding nonzero value v */ + /* Figure F.7: Encoding the sign of v */ + if (v > 0) { + arith_encode(cinfo, st + 1, 0); /* Table F.4: SS = S0 + 1 */ + st += 2; /* Table F.4: SP = S0 + 2 */ + entropy->dc_context[ci] = 4; /* small positive diff category */ + } else { + v = -v; + arith_encode(cinfo, st + 1, 1); /* Table F.4: SS = S0 + 1 */ + st += 3; /* Table F.4: SN = S0 + 3 */ + entropy->dc_context[ci] = 8; /* small negative diff category */ + } + /* Figure F.8: Encoding the magnitude category of v */ + m = 0; + if (v -= 1) { + arith_encode(cinfo, st, 1); + m = 1; + v2 = v; + st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */ + while (v2 >>= 1) { + arith_encode(cinfo, st, 1); + m <<= 1; + st += 1; + } + } + arith_encode(cinfo, st, 0); + /* Section F.1.4.4.1.2: Establish dc_context conditioning category */ + if (m < (int)((1L << cinfo->arith_dc_L[tbl]) >> 1)) + entropy->dc_context[ci] = 0; /* zero diff category */ + else if (m > (int)((1L << cinfo->arith_dc_U[tbl]) >> 1)) + entropy->dc_context[ci] += 8; /* large diff category */ + /* Figure F.9: Encoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + arith_encode(cinfo, st, (m & v) ? 1 : 0); + } + } + + return TRUE; +} + + +/* + * MCU encoding for AC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +METHODDEF(boolean) +encode_mcu_AC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + JBLOCKROW block; + unsigned char *st; + int tbl, k, ke; + int v, v2, m; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + emit_restart(cinfo, entropy->next_restart_num); + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + /* Encode the MCU data block */ + block = MCU_data[0]; + tbl = cinfo->cur_comp_info[0]->ac_tbl_no; + + /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */ + + /* Establish EOB (end-of-block) index */ + for (ke = cinfo->Se; ke > 0; ke--) + /* We must apply the point transform by Al. For AC coefficients this + * is an integer division with rounding towards 0. To do this portably + * in C, we shift after obtaining the absolute value. + */ + if ((v = (*block)[jpeg_natural_order[ke]]) >= 0) { + if (v >>= cinfo->Al) break; + } else { + v = -v; + if (v >>= cinfo->Al) break; + } + + /* Figure F.5: Encode_AC_Coefficients */ + for (k = cinfo->Ss; k <= ke; k++) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + arith_encode(cinfo, st, 0); /* EOB decision */ + for (;;) { + if ((v = (*block)[jpeg_natural_order[k]]) >= 0) { + if (v >>= cinfo->Al) { + arith_encode(cinfo, st + 1, 1); + arith_encode(cinfo, entropy->fixed_bin, 0); + break; + } + } else { + v = -v; + if (v >>= cinfo->Al) { + arith_encode(cinfo, st + 1, 1); + arith_encode(cinfo, entropy->fixed_bin, 1); + break; + } + } + arith_encode(cinfo, st + 1, 0); st += 3; k++; + } + st += 2; + /* Figure F.8: Encoding the magnitude category of v */ + m = 0; + if (v -= 1) { + arith_encode(cinfo, st, 1); + m = 1; + v2 = v; + if (v2 >>= 1) { + arith_encode(cinfo, st, 1); + m <<= 1; + st = entropy->ac_stats[tbl] + + (k <= cinfo->arith_ac_K[tbl] ? 189 : 217); + while (v2 >>= 1) { + arith_encode(cinfo, st, 1); + m <<= 1; + st += 1; + } + } + } + arith_encode(cinfo, st, 0); + /* Figure F.9: Encoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + arith_encode(cinfo, st, (m & v) ? 1 : 0); + } + /* Encode EOB decision only if k <= cinfo->Se */ + if (k <= cinfo->Se) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + arith_encode(cinfo, st, 1); + } + + return TRUE; +} + + +/* + * MCU encoding for DC successive approximation refinement scan. + */ + +METHODDEF(boolean) +encode_mcu_DC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + unsigned char *st; + int Al, blkn; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + emit_restart(cinfo, entropy->next_restart_num); + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + st = entropy->fixed_bin; /* use fixed probability estimation */ + Al = cinfo->Al; + + /* Encode the MCU data blocks */ + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + /* We simply emit the Al'th bit of the DC coefficient value. */ + arith_encode(cinfo, st, (MCU_data[blkn][0][0] >> Al) & 1); + } + + return TRUE; +} + + +/* + * MCU encoding for AC successive approximation refinement scan. + */ + +METHODDEF(boolean) +encode_mcu_AC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + JBLOCKROW block; + unsigned char *st; + int tbl, k, ke, kex; + int v; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + emit_restart(cinfo, entropy->next_restart_num); + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + /* Encode the MCU data block */ + block = MCU_data[0]; + tbl = cinfo->cur_comp_info[0]->ac_tbl_no; + + /* Section G.1.3.3: Encoding of AC coefficients */ + + /* Establish EOB (end-of-block) index */ + for (ke = cinfo->Se; ke > 0; ke--) + /* We must apply the point transform by Al. For AC coefficients this + * is an integer division with rounding towards 0. To do this portably + * in C, we shift after obtaining the absolute value. + */ + if ((v = (*block)[jpeg_natural_order[ke]]) >= 0) { + if (v >>= cinfo->Al) break; + } else { + v = -v; + if (v >>= cinfo->Al) break; + } + + /* Establish EOBx (previous stage end-of-block) index */ + for (kex = ke; kex > 0; kex--) + if ((v = (*block)[jpeg_natural_order[kex]]) >= 0) { + if (v >>= cinfo->Ah) break; + } else { + v = -v; + if (v >>= cinfo->Ah) break; + } + + /* Figure G.10: Encode_AC_Coefficients_SA */ + for (k = cinfo->Ss; k <= ke; k++) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + if (k > kex) + arith_encode(cinfo, st, 0); /* EOB decision */ + for (;;) { + if ((v = (*block)[jpeg_natural_order[k]]) >= 0) { + if (v >>= cinfo->Al) { + if (v >> 1) /* previously nonzero coef */ + arith_encode(cinfo, st + 2, (v & 1)); + else { /* newly nonzero coef */ + arith_encode(cinfo, st + 1, 1); + arith_encode(cinfo, entropy->fixed_bin, 0); + } + break; + } + } else { + v = -v; + if (v >>= cinfo->Al) { + if (v >> 1) /* previously nonzero coef */ + arith_encode(cinfo, st + 2, (v & 1)); + else { /* newly nonzero coef */ + arith_encode(cinfo, st + 1, 1); + arith_encode(cinfo, entropy->fixed_bin, 1); + } + break; + } + } + arith_encode(cinfo, st + 1, 0); st += 3; k++; + } + } + /* Encode EOB decision only if k <= cinfo->Se */ + if (k <= cinfo->Se) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + arith_encode(cinfo, st, 1); + } + + return TRUE; +} + + +/* + * Encode and output one MCU's worth of arithmetic-compressed coefficients. + */ + +METHODDEF(boolean) +encode_mcu(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + jpeg_component_info *compptr; + JBLOCKROW block; + unsigned char *st; + int blkn, ci, tbl, k, ke; + int v, v2, m; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + emit_restart(cinfo, entropy->next_restart_num); + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + /* Encode the MCU data blocks */ + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data[blkn]; + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + + /* Sections F.1.4.1 & F.1.4.4.1: Encoding of DC coefficients */ + + tbl = compptr->dc_tbl_no; + + /* Table F.4: Point to statistics bin S0 for DC coefficient coding */ + st = entropy->dc_stats[tbl] + entropy->dc_context[ci]; + + /* Figure F.4: Encode_DC_DIFF */ + if ((v = (*block)[0] - entropy->last_dc_val[ci]) == 0) { + arith_encode(cinfo, st, 0); + entropy->dc_context[ci] = 0; /* zero diff category */ + } else { + entropy->last_dc_val[ci] = (*block)[0]; + arith_encode(cinfo, st, 1); + /* Figure F.6: Encoding nonzero value v */ + /* Figure F.7: Encoding the sign of v */ + if (v > 0) { + arith_encode(cinfo, st + 1, 0); /* Table F.4: SS = S0 + 1 */ + st += 2; /* Table F.4: SP = S0 + 2 */ + entropy->dc_context[ci] = 4; /* small positive diff category */ + } else { + v = -v; + arith_encode(cinfo, st + 1, 1); /* Table F.4: SS = S0 + 1 */ + st += 3; /* Table F.4: SN = S0 + 3 */ + entropy->dc_context[ci] = 8; /* small negative diff category */ + } + /* Figure F.8: Encoding the magnitude category of v */ + m = 0; + if (v -= 1) { + arith_encode(cinfo, st, 1); + m = 1; + v2 = v; + st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */ + while (v2 >>= 1) { + arith_encode(cinfo, st, 1); + m <<= 1; + st += 1; + } + } + arith_encode(cinfo, st, 0); + /* Section F.1.4.4.1.2: Establish dc_context conditioning category */ + if (m < (int)((1L << cinfo->arith_dc_L[tbl]) >> 1)) + entropy->dc_context[ci] = 0; /* zero diff category */ + else if (m > (int)((1L << cinfo->arith_dc_U[tbl]) >> 1)) + entropy->dc_context[ci] += 8; /* large diff category */ + /* Figure F.9: Encoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + arith_encode(cinfo, st, (m & v) ? 1 : 0); + } + + /* Sections F.1.4.2 & F.1.4.4.2: Encoding of AC coefficients */ + + tbl = compptr->ac_tbl_no; + + /* Establish EOB (end-of-block) index */ + for (ke = DCTSIZE2 - 1; ke > 0; ke--) + if ((*block)[jpeg_natural_order[ke]]) break; + + /* Figure F.5: Encode_AC_Coefficients */ + for (k = 1; k <= ke; k++) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + arith_encode(cinfo, st, 0); /* EOB decision */ + while ((v = (*block)[jpeg_natural_order[k]]) == 0) { + arith_encode(cinfo, st + 1, 0); st += 3; k++; + } + arith_encode(cinfo, st + 1, 1); + /* Figure F.6: Encoding nonzero value v */ + /* Figure F.7: Encoding the sign of v */ + if (v > 0) { + arith_encode(cinfo, entropy->fixed_bin, 0); + } else { + v = -v; + arith_encode(cinfo, entropy->fixed_bin, 1); + } + st += 2; + /* Figure F.8: Encoding the magnitude category of v */ + m = 0; + if (v -= 1) { + arith_encode(cinfo, st, 1); + m = 1; + v2 = v; + if (v2 >>= 1) { + arith_encode(cinfo, st, 1); + m <<= 1; + st = entropy->ac_stats[tbl] + + (k <= cinfo->arith_ac_K[tbl] ? 189 : 217); + while (v2 >>= 1) { + arith_encode(cinfo, st, 1); + m <<= 1; + st += 1; + } + } + } + arith_encode(cinfo, st, 0); + /* Figure F.9: Encoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + arith_encode(cinfo, st, (m & v) ? 1 : 0); + } + /* Encode EOB decision only if k <= DCTSIZE2 - 1 */ + if (k <= DCTSIZE2 - 1) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + arith_encode(cinfo, st, 1); + } + } + + return TRUE; +} + + +/* + * Initialize for an arithmetic-compressed scan. + */ + +METHODDEF(void) +start_pass(j_compress_ptr cinfo, boolean gather_statistics) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + int ci, tbl; + jpeg_component_info *compptr; + + if (gather_statistics) + /* Make sure to avoid that in the master control logic! + * We are fully adaptive here and need no extra + * statistics gathering pass! + */ + ERREXIT(cinfo, JERR_NOTIMPL); + + /* We assume jcmaster.c already validated the progressive scan parameters. */ + + /* Select execution routines */ + if (cinfo->progressive_mode) { + if (cinfo->Ah == 0) { + if (cinfo->Ss == 0) + entropy->pub.encode_mcu = encode_mcu_DC_first; + else + entropy->pub.encode_mcu = encode_mcu_AC_first; + } else { + if (cinfo->Ss == 0) + entropy->pub.encode_mcu = encode_mcu_DC_refine; + else + entropy->pub.encode_mcu = encode_mcu_AC_refine; + } + } else + entropy->pub.encode_mcu = encode_mcu; + + /* Allocate & initialize requested statistics areas */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* DC needs no table for refinement scan */ + if (cinfo->progressive_mode == 0 || (cinfo->Ss == 0 && cinfo->Ah == 0)) { + tbl = compptr->dc_tbl_no; + if (tbl < 0 || tbl >= NUM_ARITH_TBLS) + ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl); + if (entropy->dc_stats[tbl] == NULL) + entropy->dc_stats[tbl] = (unsigned char *)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, DC_STAT_BINS); + memset(entropy->dc_stats[tbl], 0, DC_STAT_BINS); + /* Initialize DC predictions to 0 */ + entropy->last_dc_val[ci] = 0; + entropy->dc_context[ci] = 0; + } + /* AC needs no table when not present */ + if (cinfo->progressive_mode == 0 || cinfo->Se) { + tbl = compptr->ac_tbl_no; + if (tbl < 0 || tbl >= NUM_ARITH_TBLS) + ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl); + if (entropy->ac_stats[tbl] == NULL) + entropy->ac_stats[tbl] = (unsigned char *)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, AC_STAT_BINS); + memset(entropy->ac_stats[tbl], 0, AC_STAT_BINS); +#ifdef CALCULATE_SPECTRAL_CONDITIONING + if (cinfo->progressive_mode) + /* Section G.1.3.2: Set appropriate arithmetic conditioning value Kx */ + cinfo->arith_ac_K[tbl] = cinfo->Ss + + ((8 + cinfo->Se - cinfo->Ss) >> 4); +#endif + } + } + + /* Initialize arithmetic encoding variables */ + entropy->c = 0; + entropy->a = 0x10000L; + entropy->sc = 0; + entropy->zc = 0; + entropy->ct = 11; + entropy->buffer = -1; /* empty */ + + /* Initialize restart stuff */ + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num = 0; +} + + +/* + * Module initialization routine for arithmetic entropy encoding. + */ + +GLOBAL(void) +jinit_arith_encoder(j_compress_ptr cinfo) +{ + arith_entropy_ptr entropy; + int i; + + entropy = (arith_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(arith_entropy_encoder)); + cinfo->entropy = (struct jpeg_entropy_encoder *)entropy; + entropy->pub.start_pass = start_pass; + entropy->pub.finish_pass = finish_pass; + + /* Mark tables unallocated */ + for (i = 0; i < NUM_ARITH_TBLS; i++) { + entropy->dc_stats[i] = NULL; + entropy->ac_stats[i] = NULL; + } + + /* Initialize index for fixed probability estimation */ + entropy->fixed_bin[0] = 113; +} diff --git a/src/spice2x/external/libjpeg-turbo/jccoefct.c b/src/spice2x/external/libjpeg-turbo/jccoefct.c new file mode 100644 index 0000000..2a5dde2 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jccoefct.c @@ -0,0 +1,454 @@ +/* + * jccoefct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the coefficient buffer controller for compression. + * This controller is the top level of the lossy JPEG compressor proper. + * The coefficient buffer lies between forward-DCT and entropy encoding steps. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + + +/* We use a full-image coefficient buffer when doing Huffman optimization, + * and also for writing multiple-scan JPEG files. In all cases, the DCT + * step is run during the first pass, and subsequent passes need only read + * the buffered coefficients. + */ +#ifdef ENTROPY_OPT_SUPPORTED +#define FULL_COEF_BUFFER_SUPPORTED +#else +#ifdef C_MULTISCAN_FILES_SUPPORTED +#define FULL_COEF_BUFFER_SUPPORTED +#endif +#endif + + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_c_coef_controller pub; /* public fields */ + + JDIMENSION iMCU_row_num; /* iMCU row # within image */ + JDIMENSION mcu_ctr; /* counts MCUs processed in current row */ + int MCU_vert_offset; /* counts MCU rows within iMCU row */ + int MCU_rows_per_iMCU_row; /* number of such rows needed */ + + /* For single-pass compression, it's sufficient to buffer just one MCU + * (although this may prove a bit slow in practice). We allocate a + * workspace of C_MAX_BLOCKS_IN_MCU coefficient blocks, and reuse it for each + * MCU constructed and sent. In multi-pass modes, this array points to the + * current MCU's blocks within the virtual arrays. + */ + JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU]; + + /* In multi-pass modes, we need a virtual block array for each component. */ + jvirt_barray_ptr whole_image[MAX_COMPONENTS]; +} my_coef_controller; + +typedef my_coef_controller *my_coef_ptr; + + +/* Forward declarations */ +METHODDEF(boolean) compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf); +#ifdef FULL_COEF_BUFFER_SUPPORTED +METHODDEF(boolean) compress_first_pass(j_compress_ptr cinfo, + _JSAMPIMAGE input_buf); +METHODDEF(boolean) compress_output(j_compress_ptr cinfo, + _JSAMPIMAGE input_buf); +#endif + + +LOCAL(void) +start_iMCU_row(j_compress_ptr cinfo) +/* Reset within-iMCU-row counters for a new row */ +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + + /* In an interleaved scan, an MCU row is the same as an iMCU row. + * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. + * But at the bottom of the image, process only what's left. + */ + if (cinfo->comps_in_scan > 1) { + coef->MCU_rows_per_iMCU_row = 1; + } else { + if (coef->iMCU_row_num < (cinfo->total_iMCU_rows - 1)) + coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; + else + coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; + } + + coef->mcu_ctr = 0; + coef->MCU_vert_offset = 0; +} + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_coef(j_compress_ptr cinfo, J_BUF_MODE pass_mode) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + + coef->iMCU_row_num = 0; + start_iMCU_row(cinfo); + + switch (pass_mode) { + case JBUF_PASS_THRU: + if (coef->whole_image[0] != NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + coef->pub._compress_data = compress_data; + break; +#ifdef FULL_COEF_BUFFER_SUPPORTED + case JBUF_SAVE_AND_PASS: + if (coef->whole_image[0] == NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + coef->pub._compress_data = compress_first_pass; + break; + case JBUF_CRANK_DEST: + if (coef->whole_image[0] == NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + coef->pub._compress_data = compress_output; + break; +#endif + default: + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + break; + } +} + + +/* + * Process some data in the single-pass case. + * We process the equivalent of one fully interleaved MCU row ("iMCU" row) + * per call, ie, v_samp_factor block rows for each component in the image. + * Returns TRUE if the iMCU row is completed, FALSE if suspended. + * + * NB: input_buf contains a plane for each component in image, + * which we index according to the component's SOF position. + */ + +METHODDEF(boolean) +compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION MCU_col_num; /* index of current MCU within row */ + JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + int blkn, bi, ci, yindex, yoffset, blockcnt; + JDIMENSION ypos, xpos; + jpeg_component_info *compptr; + + /* Loop to write as much as one whole iMCU row */ + for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; + yoffset++) { + for (MCU_col_num = coef->mcu_ctr; MCU_col_num <= last_MCU_col; + MCU_col_num++) { + /* Determine where data comes from in input_buf and do the DCT thing. + * Each call on forward_DCT processes a horizontal row of DCT blocks + * as wide as an MCU; we rely on having allocated the MCU_buffer[] blocks + * sequentially. Dummy blocks at the right or bottom edge are filled in + * specially. The data in them does not matter for image reconstruction, + * so we fill them with values that will encode to the smallest amount of + * data, viz: all zeroes in the AC entries, DC entries equal to previous + * block's DC value. (Thanks to Thomas Kinsman for this idea.) + */ + blkn = 0; + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width : + compptr->last_col_width; + xpos = MCU_col_num * compptr->MCU_sample_width; + ypos = yoffset * DCTSIZE; /* ypos == (yoffset+yindex) * DCTSIZE */ + for (yindex = 0; yindex < compptr->MCU_height; yindex++) { + if (coef->iMCU_row_num < last_iMCU_row || + yoffset + yindex < compptr->last_row_height) { + (*cinfo->fdct->_forward_DCT) (cinfo, compptr, + input_buf[compptr->component_index], + coef->MCU_buffer[blkn], + ypos, xpos, (JDIMENSION)blockcnt); + if (blockcnt < compptr->MCU_width) { + /* Create some dummy blocks at the right edge of the image. */ + jzero_far((void *)coef->MCU_buffer[blkn + blockcnt], + (compptr->MCU_width - blockcnt) * sizeof(JBLOCK)); + for (bi = blockcnt; bi < compptr->MCU_width; bi++) { + coef->MCU_buffer[blkn + bi][0][0] = + coef->MCU_buffer[blkn + bi - 1][0][0]; + } + } + } else { + /* Create a row of dummy blocks at the bottom of the image. */ + jzero_far((void *)coef->MCU_buffer[blkn], + compptr->MCU_width * sizeof(JBLOCK)); + for (bi = 0; bi < compptr->MCU_width; bi++) { + coef->MCU_buffer[blkn + bi][0][0] = + coef->MCU_buffer[blkn - 1][0][0]; + } + } + blkn += compptr->MCU_width; + ypos += DCTSIZE; + } + } + /* Try to write the MCU. In event of a suspension failure, we will + * re-DCT the MCU on restart (a bit inefficient, could be fixed...) + */ + if (!(*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { + /* Suspension forced; update state counters and exit */ + coef->MCU_vert_offset = yoffset; + coef->mcu_ctr = MCU_col_num; + return FALSE; + } + } + /* Completed an MCU row, but perhaps not an iMCU row */ + coef->mcu_ctr = 0; + } + /* Completed the iMCU row, advance counters for next one */ + coef->iMCU_row_num++; + start_iMCU_row(cinfo); + return TRUE; +} + + +#ifdef FULL_COEF_BUFFER_SUPPORTED + +/* + * Process some data in the first pass of a multi-pass case. + * We process the equivalent of one fully interleaved MCU row ("iMCU" row) + * per call, ie, v_samp_factor block rows for each component in the image. + * This amount of data is read from the source buffer, DCT'd and quantized, + * and saved into the virtual arrays. We also generate suitable dummy blocks + * as needed at the right and lower edges. (The dummy blocks are constructed + * in the virtual arrays, which have been padded appropriately.) This makes + * it possible for subsequent passes not to worry about real vs. dummy blocks. + * + * We must also emit the data to the entropy encoder. This is conveniently + * done by calling compress_output() after we've loaded the current strip + * of the virtual arrays. + * + * NB: input_buf contains a plane for each component in image. All + * components are DCT'd and loaded into the virtual arrays in this pass. + * However, it may be that only a subset of the components are emitted to + * the entropy encoder during this first pass; be careful about looking + * at the scan-dependent variables (MCU dimensions, etc). + */ + +METHODDEF(boolean) +compress_first_pass(j_compress_ptr cinfo, _JSAMPIMAGE input_buf) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + JDIMENSION blocks_across, MCUs_across, MCUindex; + int bi, ci, h_samp_factor, block_row, block_rows, ndummy; + JCOEF lastDC; + jpeg_component_info *compptr; + JBLOCKARRAY buffer; + JBLOCKROW thisblockrow, lastblockrow; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Align the virtual buffer for this component. */ + buffer = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[ci], + coef->iMCU_row_num * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, TRUE); + /* Count non-dummy DCT block rows in this iMCU row. */ + if (coef->iMCU_row_num < last_iMCU_row) + block_rows = compptr->v_samp_factor; + else { + /* NB: can't use last_row_height here, since may not be set! */ + block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (block_rows == 0) block_rows = compptr->v_samp_factor; + } + blocks_across = compptr->width_in_blocks; + h_samp_factor = compptr->h_samp_factor; + /* Count number of dummy blocks to be added at the right margin. */ + ndummy = (int)(blocks_across % h_samp_factor); + if (ndummy > 0) + ndummy = h_samp_factor - ndummy; + /* Perform DCT for all non-dummy blocks in this iMCU row. Each call + * on forward_DCT processes a complete horizontal row of DCT blocks. + */ + for (block_row = 0; block_row < block_rows; block_row++) { + thisblockrow = buffer[block_row]; + (*cinfo->fdct->_forward_DCT) (cinfo, compptr, + input_buf[ci], thisblockrow, + (JDIMENSION)(block_row * DCTSIZE), + (JDIMENSION)0, blocks_across); + if (ndummy > 0) { + /* Create dummy blocks at the right edge of the image. */ + thisblockrow += blocks_across; /* => first dummy block */ + jzero_far((void *)thisblockrow, ndummy * sizeof(JBLOCK)); + lastDC = thisblockrow[-1][0]; + for (bi = 0; bi < ndummy; bi++) { + thisblockrow[bi][0] = lastDC; + } + } + } + /* If at end of image, create dummy block rows as needed. + * The tricky part here is that within each MCU, we want the DC values + * of the dummy blocks to match the last real block's DC value. + * This squeezes a few more bytes out of the resulting file... + */ + if (coef->iMCU_row_num == last_iMCU_row) { + blocks_across += ndummy; /* include lower right corner */ + MCUs_across = blocks_across / h_samp_factor; + for (block_row = block_rows; block_row < compptr->v_samp_factor; + block_row++) { + thisblockrow = buffer[block_row]; + lastblockrow = buffer[block_row - 1]; + jzero_far((void *)thisblockrow, + (size_t)(blocks_across * sizeof(JBLOCK))); + for (MCUindex = 0; MCUindex < MCUs_across; MCUindex++) { + lastDC = lastblockrow[h_samp_factor - 1][0]; + for (bi = 0; bi < h_samp_factor; bi++) { + thisblockrow[bi][0] = lastDC; + } + thisblockrow += h_samp_factor; /* advance to next MCU in row */ + lastblockrow += h_samp_factor; + } + } + } + } + /* NB: compress_output will increment iMCU_row_num if successful. + * A suspension return will result in redoing all the work above next time. + */ + + /* Emit data to the entropy encoder, sharing code with subsequent passes */ + return compress_output(cinfo, input_buf); +} + + +/* + * Process some data in subsequent passes of a multi-pass case. + * We process the equivalent of one fully interleaved MCU row ("iMCU" row) + * per call, ie, v_samp_factor block rows for each component in the scan. + * The data is obtained from the virtual arrays and fed to the entropy coder. + * Returns TRUE if the iMCU row is completed, FALSE if suspended. + * + * NB: input_buf is ignored; it is likely to be a NULL pointer. + */ + +METHODDEF(boolean) +compress_output(j_compress_ptr cinfo, _JSAMPIMAGE input_buf) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION MCU_col_num; /* index of current MCU within row */ + int blkn, ci, xindex, yindex, yoffset; + JDIMENSION start_col; + JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; + JBLOCKROW buffer_ptr; + jpeg_component_info *compptr; + + /* Align the virtual buffers for the components used in this scan. + * NB: during first pass, this is safe only because the buffers will + * already be aligned properly, so jmemmgr.c won't need to do any I/O. + */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + buffer[ci] = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[compptr->component_index], + coef->iMCU_row_num * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, FALSE); + } + + /* Loop to process one whole iMCU row */ + for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; + yoffset++) { + for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row; + MCU_col_num++) { + /* Construct list of pointers to DCT blocks belonging to this MCU */ + blkn = 0; /* index of current DCT block within MCU */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + start_col = MCU_col_num * compptr->MCU_width; + for (yindex = 0; yindex < compptr->MCU_height; yindex++) { + buffer_ptr = buffer[ci][yindex + yoffset] + start_col; + for (xindex = 0; xindex < compptr->MCU_width; xindex++) { + coef->MCU_buffer[blkn++] = buffer_ptr++; + } + } + } + /* Try to write the MCU. */ + if (!(*cinfo->entropy->encode_mcu) (cinfo, coef->MCU_buffer)) { + /* Suspension forced; update state counters and exit */ + coef->MCU_vert_offset = yoffset; + coef->mcu_ctr = MCU_col_num; + return FALSE; + } + } + /* Completed an MCU row, but perhaps not an iMCU row */ + coef->mcu_ctr = 0; + } + /* Completed the iMCU row, advance counters for next one */ + coef->iMCU_row_num++; + start_iMCU_row(cinfo); + return TRUE; +} + +#endif /* FULL_COEF_BUFFER_SUPPORTED */ + + +/* + * Initialize coefficient buffer controller. + */ + +GLOBAL(void) +_jinit_c_coef_controller(j_compress_ptr cinfo, boolean need_full_buffer) +{ + my_coef_ptr coef; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + coef = (my_coef_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_coef_controller)); + cinfo->coef = (struct jpeg_c_coef_controller *)coef; + coef->pub.start_pass = start_pass_coef; + + /* Create the coefficient buffer. */ + if (need_full_buffer) { +#ifdef FULL_COEF_BUFFER_SUPPORTED + /* Allocate a full-image virtual array for each component, */ + /* padded to a multiple of samp_factor DCT blocks in each direction. */ + int ci; + jpeg_component_info *compptr; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + coef->whole_image[ci] = (*cinfo->mem->request_virt_barray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, FALSE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)jround_up((long)compptr->height_in_blocks, + (long)compptr->v_samp_factor), + (JDIMENSION)compptr->v_samp_factor); + } +#else + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); +#endif + } else { + /* We only need a single-MCU buffer. */ + JBLOCKROW buffer; + int i; + + buffer = (JBLOCKROW) + (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, + C_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK)); + for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) { + coef->MCU_buffer[i] = buffer + i; + } + coef->whole_image[0] = NULL; /* flag for no virtual arrays */ + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jccolext.c b/src/spice2x/external/libjpeg-turbo/jccolext.c new file mode 100644 index 0000000..8eba36c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jccolext.c @@ -0,0 +1,152 @@ +/* + * jccolext.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2009-2012, 2015, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains input colorspace conversion routines. + */ + + +/* This file is included by jccolor.c */ + + +/* + * Convert some rows of samples to the JPEG colorspace. + * + * Note that we change from the application's interleaved-pixel format + * to our internal noninterleaved, one-plane-per-component format. + * The input buffer is therefore three times as wide as the output buffer. + * + * A starting row offset is provided only for the output buffer. The caller + * can easily adjust the passed input_buf value to accommodate any row + * offset required on that side. + */ + +INLINE +LOCAL(void) +rgb_ycc_convert_internal(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int r, g, b; + register JLONG *ctab = cconvert->rgb_ycc_tab; + register _JSAMPROW inptr; + register _JSAMPROW outptr0, outptr1, outptr2; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->image_width; + + while (--num_rows >= 0) { + inptr = *input_buf++; + outptr0 = output_buf[0][output_row]; + outptr1 = output_buf[1][output_row]; + outptr2 = output_buf[2][output_row]; + output_row++; + for (col = 0; col < num_cols; col++) { + r = RANGE_LIMIT(inptr[RGB_RED]); + g = RANGE_LIMIT(inptr[RGB_GREEN]); + b = RANGE_LIMIT(inptr[RGB_BLUE]); + inptr += RGB_PIXELSIZE; + /* If the inputs are 0.._MAXJSAMPLE, the outputs of these equations + * must be too; we do not need an explicit range-limiting operation. + * Hence the value being shifted is never negative, and we don't + * need the general RIGHT_SHIFT macro. + */ + /* Y */ + outptr0[col] = (_JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] + + ctab[b + B_Y_OFF]) >> SCALEBITS); + /* Cb */ + outptr1[col] = (_JSAMPLE)((ctab[r + R_CB_OFF] + ctab[g + G_CB_OFF] + + ctab[b + B_CB_OFF]) >> SCALEBITS); + /* Cr */ + outptr2[col] = (_JSAMPLE)((ctab[r + R_CR_OFF] + ctab[g + G_CR_OFF] + + ctab[b + B_CR_OFF]) >> SCALEBITS); + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/**************** Cases other than RGB -> YCbCr **************/ + + +/* + * Convert some rows of samples to the JPEG colorspace. + * This version handles RGB->grayscale conversion, which is the same + * as the RGB->Y portion of RGB->YCbCr. + * We assume rgb_ycc_start has been called (we only use the Y tables). + */ + +INLINE +LOCAL(void) +rgb_gray_convert_internal(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int r, g, b; + register JLONG *ctab = cconvert->rgb_ycc_tab; + register _JSAMPROW inptr; + register _JSAMPROW outptr; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->image_width; + + while (--num_rows >= 0) { + inptr = *input_buf++; + outptr = output_buf[0][output_row]; + output_row++; + for (col = 0; col < num_cols; col++) { + r = RANGE_LIMIT(inptr[RGB_RED]); + g = RANGE_LIMIT(inptr[RGB_GREEN]); + b = RANGE_LIMIT(inptr[RGB_BLUE]); + inptr += RGB_PIXELSIZE; + /* Y */ + outptr[col] = (_JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] + + ctab[b + B_Y_OFF]) >> SCALEBITS); + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * Convert some rows of samples to the JPEG colorspace. + * This version handles extended RGB->plain RGB conversion + */ + +INLINE +LOCAL(void) +rgb_rgb_convert_internal(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows) +{ + register _JSAMPROW inptr; + register _JSAMPROW outptr0, outptr1, outptr2; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->image_width; + + while (--num_rows >= 0) { + inptr = *input_buf++; + outptr0 = output_buf[0][output_row]; + outptr1 = output_buf[1][output_row]; + outptr2 = output_buf[2][output_row]; + output_row++; + for (col = 0; col < num_cols; col++) { + outptr0[col] = inptr[RGB_RED]; + outptr1[col] = inptr[RGB_GREEN]; + outptr2[col] = inptr[RGB_BLUE]; + inptr += RGB_PIXELSIZE; + } + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jccolor.c b/src/spice2x/external/libjpeg-turbo/jccolor.c new file mode 100644 index 0000000..cd3a6a7 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jccolor.c @@ -0,0 +1,732 @@ +/* + * jccolor.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2009-2012, 2015, 2022, D. R. Commander. + * Copyright (C) 2014, MIPS Technologies, Inc., California. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains input colorspace conversion routines. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsimd.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) + +/* Private subobject */ + +typedef struct { + struct jpeg_color_converter pub; /* public fields */ + +#if BITS_IN_JSAMPLE != 16 + /* Private state for RGB->YCC conversion */ + JLONG *rgb_ycc_tab; /* => table for RGB to YCbCr conversion */ +#endif +} my_color_converter; + +typedef my_color_converter *my_cconvert_ptr; + + +/**************** RGB -> YCbCr conversion: most common case **************/ + +/* + * YCbCr is defined per CCIR 601-1, except that Cb and Cr are + * normalized to the range 0.._MAXJSAMPLE rather than -0.5 .. 0.5. + * The conversion equations to be implemented are therefore + * Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + * Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + _CENTERJSAMPLE + * Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + _CENTERJSAMPLE + * (These numbers are derived from TIFF 6.0 section 21, dated 3-June-92.) + * Note: older versions of the IJG code used a zero offset of _MAXJSAMPLE/2, + * rather than _CENTERJSAMPLE, for Cb and Cr. This gave equal positive and + * negative swings for Cb/Cr, but meant that grayscale values (Cb=Cr=0) + * were not represented exactly. Now we sacrifice exact representation of + * maximum red and maximum blue in order to get exact grayscales. + * + * To avoid floating-point arithmetic, we represent the fractional constants + * as integers scaled up by 2^16 (about 4 digits precision); we have to divide + * the products by 2^16, with appropriate rounding, to get the correct answer. + * + * For even more speed, we avoid doing any multiplications in the inner loop + * by precalculating the constants times R,G,B for all possible values. + * For 8-bit samples this is very reasonable (only 256 entries per table); + * for 12-bit samples it is still acceptable. It's not very reasonable for + * 16-bit samples, but if you want lossless storage you shouldn't be changing + * colorspace anyway. + * The _CENTERJSAMPLE offsets and the rounding fudge-factor of 0.5 are included + * in the tables to save adding them separately in the inner loop. + */ + +#define SCALEBITS 16 /* speediest right-shift on some machines */ +#define CBCR_OFFSET ((JLONG)_CENTERJSAMPLE << SCALEBITS) +#define ONE_HALF ((JLONG)1 << (SCALEBITS - 1)) +#define FIX(x) ((JLONG)((x) * (1L << SCALEBITS) + 0.5)) + +/* We allocate one big table and divide it up into eight parts, instead of + * doing eight alloc_small requests. This lets us use a single table base + * address, which can be held in a register in the inner loops on many + * machines (more than can hold all eight addresses, anyway). + */ + +#define R_Y_OFF 0 /* offset to R => Y section */ +#define G_Y_OFF (1 * (_MAXJSAMPLE + 1)) /* offset to G => Y section */ +#define B_Y_OFF (2 * (_MAXJSAMPLE + 1)) /* etc. */ +#define R_CB_OFF (3 * (_MAXJSAMPLE + 1)) +#define G_CB_OFF (4 * (_MAXJSAMPLE + 1)) +#define B_CB_OFF (5 * (_MAXJSAMPLE + 1)) +#define R_CR_OFF B_CB_OFF /* B=>Cb, R=>Cr are the same */ +#define G_CR_OFF (6 * (_MAXJSAMPLE + 1)) +#define B_CR_OFF (7 * (_MAXJSAMPLE + 1)) +#define TABLE_SIZE (8 * (_MAXJSAMPLE + 1)) + +/* 12-bit samples use a 16-bit data type, so it is possible to pass + * out-of-range sample values (< 0 or > 4095) to jpeg_write_scanlines(). + * Thus, we mask the incoming 12-bit samples to guard against overrunning + * or underrunning the conversion tables. + */ + +#if BITS_IN_JSAMPLE == 12 +#define RANGE_LIMIT(value) ((value) & 0xFFF) +#else +#define RANGE_LIMIT(value) (value) +#endif + + +/* Include inline routines for colorspace extensions */ + +#include "jccolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE + +#define RGB_RED EXT_RGB_RED +#define RGB_GREEN EXT_RGB_GREEN +#define RGB_BLUE EXT_RGB_BLUE +#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +#define rgb_ycc_convert_internal extrgb_ycc_convert_internal +#define rgb_gray_convert_internal extrgb_gray_convert_internal +#define rgb_rgb_convert_internal extrgb_rgb_convert_internal +#include "jccolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef rgb_ycc_convert_internal +#undef rgb_gray_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_RGBX_RED +#define RGB_GREEN EXT_RGBX_GREEN +#define RGB_BLUE EXT_RGBX_BLUE +#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +#define rgb_ycc_convert_internal extrgbx_ycc_convert_internal +#define rgb_gray_convert_internal extrgbx_gray_convert_internal +#define rgb_rgb_convert_internal extrgbx_rgb_convert_internal +#include "jccolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef rgb_ycc_convert_internal +#undef rgb_gray_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_BGR_RED +#define RGB_GREEN EXT_BGR_GREEN +#define RGB_BLUE EXT_BGR_BLUE +#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +#define rgb_ycc_convert_internal extbgr_ycc_convert_internal +#define rgb_gray_convert_internal extbgr_gray_convert_internal +#define rgb_rgb_convert_internal extbgr_rgb_convert_internal +#include "jccolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef rgb_ycc_convert_internal +#undef rgb_gray_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_BGRX_RED +#define RGB_GREEN EXT_BGRX_GREEN +#define RGB_BLUE EXT_BGRX_BLUE +#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +#define rgb_ycc_convert_internal extbgrx_ycc_convert_internal +#define rgb_gray_convert_internal extbgrx_gray_convert_internal +#define rgb_rgb_convert_internal extbgrx_rgb_convert_internal +#include "jccolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef rgb_ycc_convert_internal +#undef rgb_gray_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_XBGR_RED +#define RGB_GREEN EXT_XBGR_GREEN +#define RGB_BLUE EXT_XBGR_BLUE +#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +#define rgb_ycc_convert_internal extxbgr_ycc_convert_internal +#define rgb_gray_convert_internal extxbgr_gray_convert_internal +#define rgb_rgb_convert_internal extxbgr_rgb_convert_internal +#include "jccolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef rgb_ycc_convert_internal +#undef rgb_gray_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_XRGB_RED +#define RGB_GREEN EXT_XRGB_GREEN +#define RGB_BLUE EXT_XRGB_BLUE +#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +#define rgb_ycc_convert_internal extxrgb_ycc_convert_internal +#define rgb_gray_convert_internal extxrgb_gray_convert_internal +#define rgb_rgb_convert_internal extxrgb_rgb_convert_internal +#include "jccolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef rgb_ycc_convert_internal +#undef rgb_gray_convert_internal +#undef rgb_rgb_convert_internal + + +/* + * Initialize for RGB->YCC colorspace conversion. + */ + +METHODDEF(void) +rgb_ycc_start(j_compress_ptr cinfo) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + JLONG *rgb_ycc_tab; + JLONG i; + + /* Allocate and fill in the conversion tables. */ + cconvert->rgb_ycc_tab = rgb_ycc_tab = (JLONG *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (TABLE_SIZE * sizeof(JLONG))); + + for (i = 0; i <= _MAXJSAMPLE; i++) { + rgb_ycc_tab[i + R_Y_OFF] = FIX(0.29900) * i; + rgb_ycc_tab[i + G_Y_OFF] = FIX(0.58700) * i; + rgb_ycc_tab[i + B_Y_OFF] = FIX(0.11400) * i + ONE_HALF; + rgb_ycc_tab[i + R_CB_OFF] = (-FIX(0.16874)) * i; + rgb_ycc_tab[i + G_CB_OFF] = (-FIX(0.33126)) * i; + /* We use a rounding fudge-factor of 0.5-epsilon for Cb and Cr. + * This ensures that the maximum output will round to _MAXJSAMPLE + * not _MAXJSAMPLE+1, and thus that we don't have to range-limit. + */ + rgb_ycc_tab[i + B_CB_OFF] = FIX(0.50000) * i + CBCR_OFFSET + ONE_HALF - 1; +/* B=>Cb and R=>Cr tables are the same + rgb_ycc_tab[i + R_CR_OFF] = FIX(0.50000) * i + CBCR_OFFSET + ONE_HALF - 1; +*/ + rgb_ycc_tab[i + G_CR_OFF] = (-FIX(0.41869)) * i; + rgb_ycc_tab[i + B_CR_OFF] = (-FIX(0.08131)) * i; + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * Convert some rows of samples to the JPEG colorspace. + */ + +METHODDEF(void) +rgb_ycc_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows) +{ + switch (cinfo->in_color_space) { + case JCS_EXT_RGB: + extrgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + extrgbx_ycc_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_BGR: + extbgr_ycc_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + extbgrx_ycc_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + extxbgr_ycc_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + extxrgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + default: + rgb_ycc_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + } +} + + +/**************** Cases other than RGB -> YCbCr **************/ + + +/* + * Convert some rows of samples to the JPEG colorspace. + */ + +METHODDEF(void) +rgb_gray_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows) +{ + switch (cinfo->in_color_space) { + case JCS_EXT_RGB: + extrgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + extrgbx_gray_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_BGR: + extbgr_gray_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + extbgrx_gray_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + extxbgr_gray_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + extxrgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + default: + rgb_gray_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + } +} + + +/* + * Extended RGB to plain RGB conversion + */ + +METHODDEF(void) +rgb_rgb_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows) +{ + switch (cinfo->in_color_space) { + case JCS_EXT_RGB: + extrgb_rgb_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + extrgbx_rgb_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_BGR: + extbgr_rgb_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + extbgrx_rgb_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + extxbgr_rgb_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + extxrgb_rgb_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + default: + rgb_rgb_convert_internal(cinfo, input_buf, output_buf, output_row, + num_rows); + break; + } +} + + +/* + * Convert some rows of samples to the JPEG colorspace. + * This version handles Adobe-style CMYK->YCCK conversion, + * where we convert R=1-C, G=1-M, and B=1-Y to YCbCr using the same + * conversion as above, while passing K (black) unchanged. + * We assume rgb_ycc_start has been called. + */ + +METHODDEF(void) +cmyk_ycck_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int r, g, b; + register JLONG *ctab = cconvert->rgb_ycc_tab; + register _JSAMPROW inptr; + register _JSAMPROW outptr0, outptr1, outptr2, outptr3; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->image_width; + + while (--num_rows >= 0) { + inptr = *input_buf++; + outptr0 = output_buf[0][output_row]; + outptr1 = output_buf[1][output_row]; + outptr2 = output_buf[2][output_row]; + outptr3 = output_buf[3][output_row]; + output_row++; + for (col = 0; col < num_cols; col++) { + r = _MAXJSAMPLE - RANGE_LIMIT(inptr[0]); + g = _MAXJSAMPLE - RANGE_LIMIT(inptr[1]); + b = _MAXJSAMPLE - RANGE_LIMIT(inptr[2]); + /* K passes through as-is */ + outptr3[col] = inptr[3]; + inptr += 4; + /* If the inputs are 0.._MAXJSAMPLE, the outputs of these equations + * must be too; we do not need an explicit range-limiting operation. + * Hence the value being shifted is never negative, and we don't + * need the general RIGHT_SHIFT macro. + */ + /* Y */ + outptr0[col] = (_JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] + + ctab[b + B_Y_OFF]) >> SCALEBITS); + /* Cb */ + outptr1[col] = (_JSAMPLE)((ctab[r + R_CB_OFF] + ctab[g + G_CB_OFF] + + ctab[b + B_CB_OFF]) >> SCALEBITS); + /* Cr */ + outptr2[col] = (_JSAMPLE)((ctab[r + R_CR_OFF] + ctab[g + G_CR_OFF] + + ctab[b + B_CR_OFF]) >> SCALEBITS); + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * Convert some rows of samples to the JPEG colorspace. + * This version handles grayscale output with no conversion. + * The source can be either plain grayscale or YCbCr (since Y == gray). + */ + +METHODDEF(void) +grayscale_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows) +{ + register _JSAMPROW inptr; + register _JSAMPROW outptr; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->image_width; + int instride = cinfo->input_components; + + while (--num_rows >= 0) { + inptr = *input_buf++; + outptr = output_buf[0][output_row]; + output_row++; + for (col = 0; col < num_cols; col++) { + outptr[col] = inptr[0]; + inptr += instride; + } + } +} + + +/* + * Convert some rows of samples to the JPEG colorspace. + * This version handles multi-component colorspaces without conversion. + * We assume input_components == num_components. + */ + +METHODDEF(void) +null_convert(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPIMAGE output_buf, JDIMENSION output_row, int num_rows) +{ + register _JSAMPROW inptr; + register _JSAMPROW outptr, outptr0, outptr1, outptr2, outptr3; + register JDIMENSION col; + register int ci; + int nc = cinfo->num_components; + JDIMENSION num_cols = cinfo->image_width; + + if (nc == 3) { + while (--num_rows >= 0) { + inptr = *input_buf++; + outptr0 = output_buf[0][output_row]; + outptr1 = output_buf[1][output_row]; + outptr2 = output_buf[2][output_row]; + output_row++; + for (col = 0; col < num_cols; col++) { + outptr0[col] = *inptr++; + outptr1[col] = *inptr++; + outptr2[col] = *inptr++; + } + } + } else if (nc == 4) { + while (--num_rows >= 0) { + inptr = *input_buf++; + outptr0 = output_buf[0][output_row]; + outptr1 = output_buf[1][output_row]; + outptr2 = output_buf[2][output_row]; + outptr3 = output_buf[3][output_row]; + output_row++; + for (col = 0; col < num_cols; col++) { + outptr0[col] = *inptr++; + outptr1[col] = *inptr++; + outptr2[col] = *inptr++; + outptr3[col] = *inptr++; + } + } + } else { + while (--num_rows >= 0) { + /* It seems fastest to make a separate pass for each component. */ + for (ci = 0; ci < nc; ci++) { + inptr = *input_buf; + outptr = output_buf[ci][output_row]; + for (col = 0; col < num_cols; col++) { + outptr[col] = inptr[ci]; + inptr += nc; + } + } + input_buf++; + output_row++; + } + } +} + + +/* + * Empty method for start_pass. + */ + +METHODDEF(void) +null_method(j_compress_ptr cinfo) +{ + /* no work needed */ +} + + +/* + * Module initialization routine for input colorspace conversion. + */ + +GLOBAL(void) +_jinit_color_converter(j_compress_ptr cinfo) +{ + my_cconvert_ptr cconvert; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + cconvert = (my_cconvert_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_color_converter)); + cinfo->cconvert = (struct jpeg_color_converter *)cconvert; + /* set start_pass to null method until we find out differently */ + cconvert->pub.start_pass = null_method; + + /* Make sure input_components agrees with in_color_space */ + switch (cinfo->in_color_space) { + case JCS_GRAYSCALE: + if (cinfo->input_components != 1) + ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE); + break; + + case JCS_RGB: + case JCS_EXT_RGB: + case JCS_EXT_RGBX: + case JCS_EXT_BGR: + case JCS_EXT_BGRX: + case JCS_EXT_XBGR: + case JCS_EXT_XRGB: + case JCS_EXT_RGBA: + case JCS_EXT_BGRA: + case JCS_EXT_ABGR: + case JCS_EXT_ARGB: + if (cinfo->input_components != rgb_pixelsize[cinfo->in_color_space]) + ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE); + break; + + case JCS_YCbCr: + if (cinfo->input_components != 3) + ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE); + break; + + case JCS_CMYK: + case JCS_YCCK: + if (cinfo->input_components != 4) + ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE); + break; + + default: /* JCS_UNKNOWN can be anything */ + if (cinfo->input_components < 1) + ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE); + break; + } + + /* Check num_components, set conversion method based on requested space. + * NOTE: We do not allow any lossy color conversion algorithms in lossless + * mode. + */ + switch (cinfo->jpeg_color_space) { + case JCS_GRAYSCALE: + if (cinfo->master->lossless && + cinfo->in_color_space != cinfo->jpeg_color_space) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + if (cinfo->num_components != 1) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + if (cinfo->in_color_space == JCS_GRAYSCALE) + cconvert->pub._color_convert = grayscale_convert; + else if (IsExtRGB(cinfo->in_color_space)) { +#ifdef WITH_SIMD + if (jsimd_can_rgb_gray()) + cconvert->pub._color_convert = jsimd_rgb_gray_convert; + else +#endif + { + cconvert->pub.start_pass = rgb_ycc_start; + cconvert->pub._color_convert = rgb_gray_convert; + } + } else if (cinfo->in_color_space == JCS_YCbCr) + cconvert->pub._color_convert = grayscale_convert; + else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + case JCS_RGB: + if (cinfo->master->lossless && !IsExtRGB(cinfo->in_color_space)) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + if (cinfo->num_components != 3) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + if (rgb_red[cinfo->in_color_space] == 0 && + rgb_green[cinfo->in_color_space] == 1 && + rgb_blue[cinfo->in_color_space] == 2 && + rgb_pixelsize[cinfo->in_color_space] == 3) { +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_c_can_null_convert()) + cconvert->pub._color_convert = jsimd_c_null_convert; + else +#endif + cconvert->pub._color_convert = null_convert; + } else if (IsExtRGB(cinfo->in_color_space)) + cconvert->pub._color_convert = rgb_rgb_convert; + else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + case JCS_YCbCr: + if (cinfo->master->lossless && + cinfo->in_color_space != cinfo->jpeg_color_space) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + if (cinfo->num_components != 3) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + if (IsExtRGB(cinfo->in_color_space)) { +#ifdef WITH_SIMD + if (jsimd_can_rgb_ycc()) + cconvert->pub._color_convert = jsimd_rgb_ycc_convert; + else +#endif + { + cconvert->pub.start_pass = rgb_ycc_start; + cconvert->pub._color_convert = rgb_ycc_convert; + } + } else if (cinfo->in_color_space == JCS_YCbCr) { +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_c_can_null_convert()) + cconvert->pub._color_convert = jsimd_c_null_convert; + else +#endif + cconvert->pub._color_convert = null_convert; + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + case JCS_CMYK: + if (cinfo->master->lossless && + cinfo->in_color_space != cinfo->jpeg_color_space) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + if (cinfo->num_components != 4) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + if (cinfo->in_color_space == JCS_CMYK) { +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_c_can_null_convert()) + cconvert->pub._color_convert = jsimd_c_null_convert; + else +#endif + cconvert->pub._color_convert = null_convert; + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + case JCS_YCCK: + if (cinfo->master->lossless && + cinfo->in_color_space != cinfo->jpeg_color_space) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + if (cinfo->num_components != 4) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + if (cinfo->in_color_space == JCS_CMYK) { + cconvert->pub.start_pass = rgb_ycc_start; + cconvert->pub._color_convert = cmyk_ycck_convert; + } else if (cinfo->in_color_space == JCS_YCCK) { +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_c_can_null_convert()) + cconvert->pub._color_convert = jsimd_c_null_convert; + else +#endif + cconvert->pub._color_convert = null_convert; + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + default: /* allow null conversion of JCS_UNKNOWN */ + if (cinfo->jpeg_color_space != cinfo->in_color_space || + cinfo->num_components != cinfo->input_components) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_c_can_null_convert()) + cconvert->pub._color_convert = jsimd_c_null_convert; + else +#endif + cconvert->pub._color_convert = null_convert; + break; + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jcdctmgr.c b/src/spice2x/external/libjpeg-turbo/jcdctmgr.c new file mode 100644 index 0000000..dbdbad6 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcdctmgr.c @@ -0,0 +1,748 @@ +/* + * jcdctmgr.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 1999-2006, MIYASAKA Masaru. + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2011, 2014-2015, 2022, 2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the forward-DCT management logic. + * This code selects a particular DCT implementation to be used, + * and it performs related housekeeping chores including coefficient + * quantization. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ +#include "jsimddct.h" + + +/* Private subobject for this module */ + +typedef void (*forward_DCT_method_ptr) (DCTELEM *data); +typedef void (*float_DCT_method_ptr) (FAST_FLOAT *data); + +typedef void (*convsamp_method_ptr) (_JSAMPARRAY sample_data, + JDIMENSION start_col, + DCTELEM *workspace); +typedef void (*float_convsamp_method_ptr) (_JSAMPARRAY sample_data, + JDIMENSION start_col, + FAST_FLOAT *workspace); + +typedef void (*quantize_method_ptr) (JCOEFPTR coef_block, DCTELEM *divisors, + DCTELEM *workspace); +typedef void (*float_quantize_method_ptr) (JCOEFPTR coef_block, + FAST_FLOAT *divisors, + FAST_FLOAT *workspace); + +METHODDEF(void) quantize(JCOEFPTR, DCTELEM *, DCTELEM *); + +typedef struct { + struct jpeg_forward_dct pub; /* public fields */ + + /* Pointer to the DCT routine actually in use */ + forward_DCT_method_ptr dct; + convsamp_method_ptr convsamp; + quantize_method_ptr quantize; + + /* The actual post-DCT divisors --- not identical to the quant table + * entries, because of scaling (especially for an unnormalized DCT). + * Each table is given in normal array order. + */ + DCTELEM *divisors[NUM_QUANT_TBLS]; + + /* work area for FDCT subroutine */ + DCTELEM *workspace; + +#ifdef DCT_FLOAT_SUPPORTED + /* Same as above for the floating-point case. */ + float_DCT_method_ptr float_dct; + float_convsamp_method_ptr float_convsamp; + float_quantize_method_ptr float_quantize; + FAST_FLOAT *float_divisors[NUM_QUANT_TBLS]; + FAST_FLOAT *float_workspace; +#endif +} my_fdct_controller; + +typedef my_fdct_controller *my_fdct_ptr; + + +#if BITS_IN_JSAMPLE == 8 + +/* + * Find the highest bit in an integer through binary search. + */ + +LOCAL(int) +flss(UINT16 val) +{ + int bit; + + bit = 16; + + if (!val) + return 0; + + if (!(val & 0xff00)) { + bit -= 8; + val <<= 8; + } + if (!(val & 0xf000)) { + bit -= 4; + val <<= 4; + } + if (!(val & 0xc000)) { + bit -= 2; + val <<= 2; + } + if (!(val & 0x8000)) { + bit -= 1; + val <<= 1; + } + + return bit; +} + + +/* + * Compute values to do a division using reciprocal. + * + * This implementation is based on an algorithm described in + * "Optimizing subroutines in assembly language: + * An optimization guide for x86 platforms" (https://agner.org/optimize). + * More information about the basic algorithm can be found in + * the paper "Integer Division Using Reciprocals" by Robert Alverson. + * + * The basic idea is to replace x/d by x * d^-1. In order to store + * d^-1 with enough precision we shift it left a few places. It turns + * out that this algoright gives just enough precision, and also fits + * into DCTELEM: + * + * b = (the number of significant bits in divisor) - 1 + * r = (word size) + b + * f = 2^r / divisor + * + * f will not be an integer for most cases, so we need to compensate + * for the rounding error introduced: + * + * no fractional part: + * + * result = input >> r + * + * fractional part of f < 0.5: + * + * round f down to nearest integer + * result = ((input + 1) * f) >> r + * + * fractional part of f > 0.5: + * + * round f up to nearest integer + * result = (input * f) >> r + * + * This is the original algorithm that gives truncated results. But we + * want properly rounded results, so we replace "input" with + * "input + divisor/2". + * + * In order to allow SIMD implementations we also tweak the values to + * allow the same calculation to be made at all times: + * + * dctbl[0] = f rounded to nearest integer + * dctbl[1] = divisor / 2 (+ 1 if fractional part of f < 0.5) + * dctbl[2] = 1 << ((word size) * 2 - r) + * dctbl[3] = r - (word size) + * + * dctbl[2] is for stupid instruction sets where the shift operation + * isn't member wise (e.g. MMX). + * + * The reason dctbl[2] and dctbl[3] reduce the shift with (word size) + * is that most SIMD implementations have a "multiply and store top + * half" operation. + * + * Lastly, we store each of the values in their own table instead + * of in a consecutive manner, yet again in order to allow SIMD + * routines. + */ + +LOCAL(int) +compute_reciprocal(UINT16 divisor, DCTELEM *dtbl) +{ + UDCTELEM2 fq, fr; + UDCTELEM c; + int b, r; + + if (divisor == 1) { + /* divisor == 1 means unquantized, so these reciprocal/correction/shift + * values will cause the C quantization algorithm to act like the + * identity function. Since only the C quantization algorithm is used in + * these cases, the scale value is irrelevant. + */ + dtbl[DCTSIZE2 * 0] = (DCTELEM)1; /* reciprocal */ + dtbl[DCTSIZE2 * 1] = (DCTELEM)0; /* correction */ + dtbl[DCTSIZE2 * 2] = (DCTELEM)1; /* scale */ + dtbl[DCTSIZE2 * 3] = -(DCTELEM)(sizeof(DCTELEM) * 8); /* shift */ + return 0; + } + + b = flss(divisor) - 1; + r = sizeof(DCTELEM) * 8 + b; + + fq = ((UDCTELEM2)1 << r) / divisor; + fr = ((UDCTELEM2)1 << r) % divisor; + + c = divisor / 2; /* for rounding */ + + if (fr == 0) { /* divisor is power of two */ + /* fq will be one bit too large to fit in DCTELEM, so adjust */ + fq >>= 1; + r--; + } else if (fr <= (divisor / 2U)) { /* fractional part is < 0.5 */ + c++; + } else { /* fractional part is > 0.5 */ + fq++; + } + + dtbl[DCTSIZE2 * 0] = (DCTELEM)fq; /* reciprocal */ + dtbl[DCTSIZE2 * 1] = (DCTELEM)c; /* correction + roundfactor */ +#ifdef WITH_SIMD + dtbl[DCTSIZE2 * 2] = (DCTELEM)(1 << (sizeof(DCTELEM) * 8 * 2 - r)); /* scale */ +#else + dtbl[DCTSIZE2 * 2] = 1; +#endif + dtbl[DCTSIZE2 * 3] = (DCTELEM)r - sizeof(DCTELEM) * 8; /* shift */ + + if (r <= 16) return 0; + else return 1; +} + +#endif + + +/* + * Initialize for a processing pass. + * Verify that all referenced Q-tables are present, and set up + * the divisor table for each one. + * In the current implementation, DCT of all components is done during + * the first pass, even if only some components will be output in the + * first scan. Hence all components should be examined here. + */ + +METHODDEF(void) +start_pass_fdctmgr(j_compress_ptr cinfo) +{ + my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct; + int ci, qtblno, i; + jpeg_component_info *compptr; + JQUANT_TBL *qtbl; + DCTELEM *dtbl; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + qtblno = compptr->quant_tbl_no; + /* Make sure specified quantization table is present */ + if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS || + cinfo->quant_tbl_ptrs[qtblno] == NULL) + ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno); + qtbl = cinfo->quant_tbl_ptrs[qtblno]; + /* Compute divisors for this quant table */ + /* We may do this more than once for same table, but it's not a big deal */ + switch (cinfo->dct_method) { +#ifdef DCT_ISLOW_SUPPORTED + case JDCT_ISLOW: + /* For LL&M IDCT method, divisors are equal to raw quantization + * coefficients multiplied by 8 (to counteract scaling). + */ + if (fdct->divisors[qtblno] == NULL) { + fdct->divisors[qtblno] = (DCTELEM *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (DCTSIZE2 * 4) * sizeof(DCTELEM)); + } + dtbl = fdct->divisors[qtblno]; + for (i = 0; i < DCTSIZE2; i++) { +#if BITS_IN_JSAMPLE == 8 +#ifdef WITH_SIMD + if (!compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i]) && + fdct->quantize == jsimd_quantize) + fdct->quantize = quantize; +#else + compute_reciprocal(qtbl->quantval[i] << 3, &dtbl[i]); +#endif +#else + dtbl[i] = ((DCTELEM)qtbl->quantval[i]) << 3; +#endif + } + break; +#endif +#ifdef DCT_IFAST_SUPPORTED + case JDCT_IFAST: + { + /* For AA&N IDCT method, divisors are equal to quantization + * coefficients scaled by scalefactor[row]*scalefactor[col], where + * scalefactor[0] = 1 + * scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7 + * We apply a further scale factor of 8. + */ +#define CONST_BITS 14 + static const INT16 aanscales[DCTSIZE2] = { + /* precomputed values scaled up by 14 bits */ + 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, + 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, + 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, + 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, + 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, + 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, + 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, + 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 + }; + SHIFT_TEMPS + + if (fdct->divisors[qtblno] == NULL) { + fdct->divisors[qtblno] = (DCTELEM *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (DCTSIZE2 * 4) * sizeof(DCTELEM)); + } + dtbl = fdct->divisors[qtblno]; + for (i = 0; i < DCTSIZE2; i++) { +#if BITS_IN_JSAMPLE == 8 +#ifdef WITH_SIMD + if (!compute_reciprocal( + DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i], + (JLONG)aanscales[i]), + CONST_BITS - 3), &dtbl[i]) && + fdct->quantize == jsimd_quantize) + fdct->quantize = quantize; +#else + compute_reciprocal( + DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i], + (JLONG)aanscales[i]), + CONST_BITS-3), &dtbl[i]); +#endif +#else + dtbl[i] = (DCTELEM) + DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i], + (JLONG)aanscales[i]), + CONST_BITS - 3); +#endif + } + } + break; +#endif +#ifdef DCT_FLOAT_SUPPORTED + case JDCT_FLOAT: + { + /* For float AA&N IDCT method, divisors are equal to quantization + * coefficients scaled by scalefactor[row]*scalefactor[col], where + * scalefactor[0] = 1 + * scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7 + * We apply a further scale factor of 8. + * What's actually stored is 1/divisor so that the inner loop can + * use a multiplication rather than a division. + */ + FAST_FLOAT *fdtbl; + int row, col; + static const double aanscalefactor[DCTSIZE] = { + 1.0, 1.387039845, 1.306562965, 1.175875602, + 1.0, 0.785694958, 0.541196100, 0.275899379 + }; + + if (fdct->float_divisors[qtblno] == NULL) { + fdct->float_divisors[qtblno] = (FAST_FLOAT *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + DCTSIZE2 * sizeof(FAST_FLOAT)); + } + fdtbl = fdct->float_divisors[qtblno]; + i = 0; + for (row = 0; row < DCTSIZE; row++) { + for (col = 0; col < DCTSIZE; col++) { + fdtbl[i] = (FAST_FLOAT) + (1.0 / (((double)qtbl->quantval[i] * + aanscalefactor[row] * aanscalefactor[col] * 8.0))); + i++; + } + } + } + break; +#endif + default: + ERREXIT(cinfo, JERR_NOT_COMPILED); + break; + } + } +} + + +/* + * Load data into workspace, applying unsigned->signed conversion. + */ + +METHODDEF(void) +convsamp(_JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace) +{ + register DCTELEM *workspaceptr; + register _JSAMPROW elemptr; + register int elemr; + + workspaceptr = workspace; + for (elemr = 0; elemr < DCTSIZE; elemr++) { + elemptr = sample_data[elemr] + start_col; + +#if DCTSIZE == 8 /* unroll the inner loop */ + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; +#else + { + register int elemc; + for (elemc = DCTSIZE; elemc > 0; elemc--) + *workspaceptr++ = (*elemptr++) - _CENTERJSAMPLE; + } +#endif + } +} + + +/* + * Quantize/descale the coefficients, and store into coef_blocks[]. + */ + +METHODDEF(void) +quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace) +{ + int i; + DCTELEM temp; + JCOEFPTR output_ptr = coef_block; + +#if BITS_IN_JSAMPLE == 8 + + UDCTELEM recip, corr; + int shift; + UDCTELEM2 product; + + for (i = 0; i < DCTSIZE2; i++) { + temp = workspace[i]; + recip = divisors[i + DCTSIZE2 * 0]; + corr = divisors[i + DCTSIZE2 * 1]; + shift = divisors[i + DCTSIZE2 * 3]; + + if (temp < 0) { + temp = -temp; + product = (UDCTELEM2)(temp + corr) * recip; + product >>= shift + sizeof(DCTELEM) * 8; + temp = (DCTELEM)product; + temp = -temp; + } else { + product = (UDCTELEM2)(temp + corr) * recip; + product >>= shift + sizeof(DCTELEM) * 8; + temp = (DCTELEM)product; + } + output_ptr[i] = (JCOEF)temp; + } + +#else + + register DCTELEM qval; + + for (i = 0; i < DCTSIZE2; i++) { + qval = divisors[i]; + temp = workspace[i]; + /* Divide the coefficient value by qval, ensuring proper rounding. + * Since C does not specify the direction of rounding for negative + * quotients, we have to force the dividend positive for portability. + * + * In most files, at least half of the output values will be zero + * (at default quantization settings, more like three-quarters...) + * so we should ensure that this case is fast. On many machines, + * a comparison is enough cheaper than a divide to make a special test + * a win. Since both inputs will be nonnegative, we need only test + * for a < b to discover whether a/b is 0. + * If your machine's division is fast enough, define FAST_DIVIDE. + */ +#ifdef FAST_DIVIDE +#define DIVIDE_BY(a, b) a /= b +#else +#define DIVIDE_BY(a, b) if (a >= b) a /= b; else a = 0 +#endif + if (temp < 0) { + temp = -temp; + temp += qval >> 1; /* for rounding */ + DIVIDE_BY(temp, qval); + temp = -temp; + } else { + temp += qval >> 1; /* for rounding */ + DIVIDE_BY(temp, qval); + } + output_ptr[i] = (JCOEF)temp; + } + +#endif + +} + + +/* + * Perform forward DCT on one or more blocks of a component. + * + * The input samples are taken from the sample_data[] array starting at + * position start_row/start_col, and moving to the right for any additional + * blocks. The quantized coefficients are returned in coef_blocks[]. + */ + +METHODDEF(void) +forward_DCT(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY sample_data, JBLOCKROW coef_blocks, + JDIMENSION start_row, JDIMENSION start_col, JDIMENSION num_blocks) +/* This version is used for integer DCT implementations. */ +{ + /* This routine is heavily used, so it's worth coding it tightly. */ + my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct; + DCTELEM *divisors = fdct->divisors[compptr->quant_tbl_no]; + DCTELEM *workspace; + JDIMENSION bi; + + /* Make sure the compiler doesn't look up these every pass */ + forward_DCT_method_ptr do_dct = fdct->dct; + convsamp_method_ptr do_convsamp = fdct->convsamp; + quantize_method_ptr do_quantize = fdct->quantize; + workspace = fdct->workspace; + + sample_data += start_row; /* fold in the vertical offset once */ + + for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) { + /* Load data into workspace, applying unsigned->signed conversion */ + (*do_convsamp) (sample_data, start_col, workspace); + + /* Perform the DCT */ + (*do_dct) (workspace); + + /* Quantize/descale the coefficients, and store into coef_blocks[] */ + (*do_quantize) (coef_blocks[bi], divisors, workspace); + } +} + + +#ifdef DCT_FLOAT_SUPPORTED + +METHODDEF(void) +convsamp_float(_JSAMPARRAY sample_data, JDIMENSION start_col, + FAST_FLOAT *workspace) +{ + register FAST_FLOAT *workspaceptr; + register _JSAMPROW elemptr; + register int elemr; + + workspaceptr = workspace; + for (elemr = 0; elemr < DCTSIZE; elemr++) { + elemptr = sample_data[elemr] + start_col; +#if DCTSIZE == 8 /* unroll the inner loop */ + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); +#else + { + register int elemc; + for (elemc = DCTSIZE; elemc > 0; elemc--) + *workspaceptr++ = (FAST_FLOAT)((*elemptr++) - _CENTERJSAMPLE); + } +#endif + } +} + + +METHODDEF(void) +quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors, + FAST_FLOAT *workspace) +{ + register FAST_FLOAT temp; + register int i; + register JCOEFPTR output_ptr = coef_block; + + for (i = 0; i < DCTSIZE2; i++) { + /* Apply the quantization and scaling factor */ + temp = workspace[i] * divisors[i]; + + /* Round to nearest integer. + * Since C does not specify the direction of rounding for negative + * quotients, we have to force the dividend positive for portability. + * The maximum coefficient size is +-16K (for 12-bit data), so this + * code should work for either 16-bit or 32-bit ints. + */ + output_ptr[i] = (JCOEF)((int)(temp + (FAST_FLOAT)16384.5) - 16384); + } +} + + +METHODDEF(void) +forward_DCT_float(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY sample_data, JBLOCKROW coef_blocks, + JDIMENSION start_row, JDIMENSION start_col, + JDIMENSION num_blocks) +/* This version is used for floating-point DCT implementations. */ +{ + /* This routine is heavily used, so it's worth coding it tightly. */ + my_fdct_ptr fdct = (my_fdct_ptr)cinfo->fdct; + FAST_FLOAT *divisors = fdct->float_divisors[compptr->quant_tbl_no]; + FAST_FLOAT *workspace; + JDIMENSION bi; + + + /* Make sure the compiler doesn't look up these every pass */ + float_DCT_method_ptr do_dct = fdct->float_dct; + float_convsamp_method_ptr do_convsamp = fdct->float_convsamp; + float_quantize_method_ptr do_quantize = fdct->float_quantize; + workspace = fdct->float_workspace; + + sample_data += start_row; /* fold in the vertical offset once */ + + for (bi = 0; bi < num_blocks; bi++, start_col += DCTSIZE) { + /* Load data into workspace, applying unsigned->signed conversion */ + (*do_convsamp) (sample_data, start_col, workspace); + + /* Perform the DCT */ + (*do_dct) (workspace); + + /* Quantize/descale the coefficients, and store into coef_blocks[] */ + (*do_quantize) (coef_blocks[bi], divisors, workspace); + } +} + +#endif /* DCT_FLOAT_SUPPORTED */ + + +/* + * Initialize FDCT manager. + */ + +GLOBAL(void) +_jinit_forward_dct(j_compress_ptr cinfo) +{ + my_fdct_ptr fdct; + int i; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + fdct = (my_fdct_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_fdct_controller)); + cinfo->fdct = (struct jpeg_forward_dct *)fdct; + fdct->pub.start_pass = start_pass_fdctmgr; + + /* First determine the DCT... */ + switch (cinfo->dct_method) { +#ifdef DCT_ISLOW_SUPPORTED + case JDCT_ISLOW: + fdct->pub._forward_DCT = forward_DCT; +#ifdef WITH_SIMD + if (jsimd_can_fdct_islow()) + fdct->dct = jsimd_fdct_islow; + else +#endif + fdct->dct = _jpeg_fdct_islow; + break; +#endif +#ifdef DCT_IFAST_SUPPORTED + case JDCT_IFAST: + fdct->pub._forward_DCT = forward_DCT; +#ifdef WITH_SIMD + if (jsimd_can_fdct_ifast()) + fdct->dct = jsimd_fdct_ifast; + else +#endif + fdct->dct = _jpeg_fdct_ifast; + break; +#endif +#ifdef DCT_FLOAT_SUPPORTED + case JDCT_FLOAT: + fdct->pub._forward_DCT = forward_DCT_float; +#ifdef WITH_SIMD + if (jsimd_can_fdct_float()) + fdct->float_dct = jsimd_fdct_float; + else +#endif + fdct->float_dct = jpeg_fdct_float; + break; +#endif + default: + ERREXIT(cinfo, JERR_NOT_COMPILED); + break; + } + + /* ...then the supporting stages. */ + switch (cinfo->dct_method) { +#ifdef DCT_ISLOW_SUPPORTED + case JDCT_ISLOW: +#endif +#ifdef DCT_IFAST_SUPPORTED + case JDCT_IFAST: +#endif +#if defined(DCT_ISLOW_SUPPORTED) || defined(DCT_IFAST_SUPPORTED) +#ifdef WITH_SIMD + if (jsimd_can_convsamp()) + fdct->convsamp = jsimd_convsamp; + else +#endif + fdct->convsamp = convsamp; +#ifdef WITH_SIMD + if (jsimd_can_quantize()) + fdct->quantize = jsimd_quantize; + else +#endif + fdct->quantize = quantize; + break; +#endif +#ifdef DCT_FLOAT_SUPPORTED + case JDCT_FLOAT: +#ifdef WITH_SIMD + if (jsimd_can_convsamp_float()) + fdct->float_convsamp = jsimd_convsamp_float; + else +#endif + fdct->float_convsamp = convsamp_float; +#ifdef WITH_SIMD + if (jsimd_can_quantize_float()) + fdct->float_quantize = jsimd_quantize_float; + else +#endif + fdct->float_quantize = quantize_float; + break; +#endif + default: + ERREXIT(cinfo, JERR_NOT_COMPILED); + break; + } + + /* Allocate workspace memory */ +#ifdef DCT_FLOAT_SUPPORTED + if (cinfo->dct_method == JDCT_FLOAT) + fdct->float_workspace = (FAST_FLOAT *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(FAST_FLOAT) * DCTSIZE2); + else +#endif + fdct->workspace = (DCTELEM *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(DCTELEM) * DCTSIZE2); + + /* Mark divisor tables unallocated */ + for (i = 0; i < NUM_QUANT_TBLS; i++) { + fdct->divisors[i] = NULL; +#ifdef DCT_FLOAT_SUPPORTED + fdct->float_divisors[i] = NULL; +#endif + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jcdiffct.c b/src/spice2x/external/libjpeg-turbo/jcdiffct.c new file mode 100644 index 0000000..0bae068 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcdiffct.c @@ -0,0 +1,411 @@ +/* + * jcdiffct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the difference buffer controller for compression. + * This controller is the top level of the lossless JPEG compressor proper. + * The difference buffer lies between the prediction/differencing and entropy + * encoding steps. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jlossls.h" /* Private declarations for lossless codec */ + + +#ifdef C_LOSSLESS_SUPPORTED + +/* We use a full-image sample buffer when doing Huffman optimization, + * and also for writing multiple-scan JPEG files. In all cases, the + * full-image buffer is filled during the first pass, and the scaling, + * prediction and differencing steps are run during subsequent passes. + */ +#ifdef ENTROPY_OPT_SUPPORTED +#define FULL_SAMP_BUFFER_SUPPORTED +#else +#ifdef C_MULTISCAN_FILES_SUPPORTED +#define FULL_SAMP_BUFFER_SUPPORTED +#endif +#endif + + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_c_coef_controller pub; /* public fields */ + + JDIMENSION iMCU_row_num; /* iMCU row # within image */ + JDIMENSION mcu_ctr; /* counts MCUs processed in current row */ + int MCU_vert_offset; /* counts MCU rows within iMCU row */ + int MCU_rows_per_iMCU_row; /* number of such rows needed */ + + _JSAMPROW cur_row[MAX_COMPONENTS]; /* row of point-transformed samples */ + _JSAMPROW prev_row[MAX_COMPONENTS]; /* previous row of Pt'd samples */ + JDIFFARRAY diff_buf[MAX_COMPONENTS]; /* iMCU row of differences */ + + /* In multi-pass modes, we need a virtual sample array for each component. */ + jvirt_sarray_ptr whole_image[MAX_COMPONENTS]; +} my_diff_controller; + +typedef my_diff_controller *my_diff_ptr; + + +/* Forward declarations */ +METHODDEF(boolean) compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf); +#ifdef FULL_SAMP_BUFFER_SUPPORTED +METHODDEF(boolean) compress_first_pass(j_compress_ptr cinfo, + _JSAMPIMAGE input_buf); +METHODDEF(boolean) compress_output(j_compress_ptr cinfo, + _JSAMPIMAGE input_buf); +#endif + + +LOCAL(void) +start_iMCU_row(j_compress_ptr cinfo) +/* Reset within-iMCU-row counters for a new row */ +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + + /* In an interleaved scan, an MCU row is the same as an iMCU row. + * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. + * But at the bottom of the image, process only what's left. + */ + if (cinfo->comps_in_scan > 1) { + diff->MCU_rows_per_iMCU_row = 1; + } else { + if (diff->iMCU_row_num < (cinfo->total_iMCU_rows-1)) + diff->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; + else + diff->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; + } + + diff->mcu_ctr = 0; + diff->MCU_vert_offset = 0; +} + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_diff(j_compress_ptr cinfo, J_BUF_MODE pass_mode) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + + /* Because it is hitching a ride on the jpeg_forward_dct struct, + * start_pass_lossless() will be called at the start of the initial pass. + * This ensures that it will be called at the start of the Huffman + * optimization and output passes as well. + */ + if (pass_mode == JBUF_CRANK_DEST) + (*cinfo->fdct->start_pass) (cinfo); + + diff->iMCU_row_num = 0; + start_iMCU_row(cinfo); + + switch (pass_mode) { + case JBUF_PASS_THRU: + if (diff->whole_image[0] != NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + diff->pub._compress_data = compress_data; + break; +#ifdef FULL_SAMP_BUFFER_SUPPORTED + case JBUF_SAVE_AND_PASS: + if (diff->whole_image[0] == NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + diff->pub._compress_data = compress_first_pass; + break; + case JBUF_CRANK_DEST: + if (diff->whole_image[0] == NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + diff->pub._compress_data = compress_output; + break; +#endif + default: + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + break; + } +} + + +#define SWAP_ROWS(rowa, rowb) { \ + _JSAMPROW temp = rowa; \ + rowa = rowb; rowb = temp; \ +} + +/* + * Process some data in the single-pass case. + * We process the equivalent of one fully interleaved MCU row ("iMCU" row) + * per call, ie, v_samp_factor rows for each component in the image. + * Returns TRUE if the iMCU row is completed, FALSE if suspended. + * + * NB: input_buf contains a plane for each component in image, + * which we index according to the component's SOF position. + */ + +METHODDEF(boolean) +compress_data(j_compress_ptr cinfo, _JSAMPIMAGE input_buf) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + lossless_comp_ptr losslessc = (lossless_comp_ptr)cinfo->fdct; + JDIMENSION MCU_col_num; /* index of current MCU within row */ + JDIMENSION MCU_count; /* number of MCUs encoded */ + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + int ci, compi, yoffset, samp_row, samp_rows, samps_across; + jpeg_component_info *compptr; + + /* Loop to write as much as one whole iMCU row */ + for (yoffset = diff->MCU_vert_offset; yoffset < diff->MCU_rows_per_iMCU_row; + yoffset++) { + + MCU_col_num = diff->mcu_ctr; + + /* Scale and predict each scanline of the MCU row separately. + * + * Note: We only do this if we are at the start of an MCU row, ie, + * we don't want to reprocess a row suspended by the output. + */ + if (MCU_col_num == 0) { + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + compi = compptr->component_index; + if (diff->iMCU_row_num < last_iMCU_row) + samp_rows = compptr->v_samp_factor; + else { + /* NB: can't use last_row_height here, since may not be set! */ + samp_rows = + (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (samp_rows == 0) samp_rows = compptr->v_samp_factor; + else { + /* Fill dummy difference rows at the bottom edge with zeros, which + * will encode to the smallest amount of data. + */ + for (samp_row = samp_rows; samp_row < compptr->v_samp_factor; + samp_row++) + memset(diff->diff_buf[compi][samp_row], 0, + jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor) * sizeof(JDIFF)); + } + } + samps_across = compptr->width_in_blocks; + + for (samp_row = 0; samp_row < samp_rows; samp_row++) { + (*losslessc->scaler_scale) (cinfo, + input_buf[compi][samp_row], + diff->cur_row[compi], + samps_across); + (*losslessc->predict_difference[compi]) + (cinfo, compi, diff->cur_row[compi], diff->prev_row[compi], + diff->diff_buf[compi][samp_row], samps_across); + SWAP_ROWS(diff->cur_row[compi], diff->prev_row[compi]); + } + } + } + /* Try to write the MCU row (or remaining portion of suspended MCU row). */ + MCU_count = + (*cinfo->entropy->encode_mcus) (cinfo, + diff->diff_buf, yoffset, MCU_col_num, + cinfo->MCUs_per_row - MCU_col_num); + if (MCU_count != cinfo->MCUs_per_row - MCU_col_num) { + /* Suspension forced; update state counters and exit */ + diff->MCU_vert_offset = yoffset; + diff->mcu_ctr += MCU_col_num; + return FALSE; + } + /* Completed an MCU row, but perhaps not an iMCU row */ + diff->mcu_ctr = 0; + } + /* Completed the iMCU row, advance counters for next one */ + diff->iMCU_row_num++; + start_iMCU_row(cinfo); + return TRUE; +} + + +#ifdef FULL_SAMP_BUFFER_SUPPORTED + +/* + * Process some data in the first pass of a multi-pass case. + * We process the equivalent of one fully interleaved MCU row ("iMCU" row) + * per call, ie, v_samp_factor rows for each component in the image. + * This amount of data is read from the source buffer and saved into the + * virtual arrays. + * + * We must also emit the data to the compressor. This is conveniently + * done by calling compress_output() after we've loaded the current strip + * of the virtual arrays. + * + * NB: input_buf contains a plane for each component in image. All components + * are loaded into the virtual arrays in this pass. However, it may be that + * only a subset of the components are emitted to the compressor during + * this first pass; be careful about looking at the scan-dependent variables + * (MCU dimensions, etc). + */ + +METHODDEF(boolean) +compress_first_pass(j_compress_ptr cinfo, _JSAMPIMAGE input_buf) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + JDIMENSION samps_across; + int ci, samp_row, samp_rows; + _JSAMPARRAY buffer; + jpeg_component_info *compptr; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Align the virtual buffer for this component. */ + buffer = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray) + ((j_common_ptr)cinfo, diff->whole_image[ci], + diff->iMCU_row_num * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, TRUE); + + /* Count non-dummy sample rows in this iMCU row. */ + if (diff->iMCU_row_num < last_iMCU_row) + samp_rows = compptr->v_samp_factor; + else { + /* NB: can't use last_row_height here, since may not be set! */ + samp_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (samp_rows == 0) samp_rows = compptr->v_samp_factor; + } + samps_across = compptr->width_in_blocks; + + /* Perform point transform scaling and prediction/differencing for all + * non-dummy rows in this iMCU row. Each call on these functions + * processes a complete row of samples. + */ + for (samp_row = 0; samp_row < samp_rows; samp_row++) { + memcpy(buffer[samp_row], input_buf[ci][samp_row], + samps_across * sizeof(_JSAMPLE)); + } + } + /* NB: compress_output will increment iMCU_row_num if successful. + * A suspension return will result in redoing all the work above next time. + */ + + /* Emit data to the compressor, sharing code with subsequent passes */ + return compress_output(cinfo, input_buf); +} + + +/* + * Process some data in subsequent passes of a multi-pass case. + * We process the equivalent of one fully interleaved MCU row ("iMCU" row) + * per call, ie, v_samp_factor rows for each component in the scan. + * The data is obtained from the virtual arrays and fed to the compressor. + * Returns TRUE if the iMCU row is completed, FALSE if suspended. + * + * NB: input_buf is ignored; it is likely to be a NULL pointer. + */ + +METHODDEF(boolean) +compress_output(j_compress_ptr cinfo, _JSAMPIMAGE input_buf) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + int ci, compi; + _JSAMPARRAY buffer[MAX_COMPS_IN_SCAN]; + jpeg_component_info *compptr; + + /* Align the virtual buffers for the components used in this scan. + * NB: during first pass, this is safe only because the buffers will + * already be aligned properly, so jmemmgr.c won't need to do any I/O. + */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + compi = compptr->component_index; + buffer[compi] = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray) + ((j_common_ptr)cinfo, diff->whole_image[compi], + diff->iMCU_row_num * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, FALSE); + } + + return compress_data(cinfo, buffer); +} + +#endif /* FULL_SAMP_BUFFER_SUPPORTED */ + + +/* + * Initialize difference buffer controller. + */ + +GLOBAL(void) +_jinit_c_diff_controller(j_compress_ptr cinfo, boolean need_full_buffer) +{ + my_diff_ptr diff; + int ci, row; + jpeg_component_info *compptr; + + diff = (my_diff_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_diff_controller)); + cinfo->coef = (struct jpeg_c_coef_controller *)diff; + diff->pub.start_pass = start_pass_diff; + + /* Create the prediction row buffers. */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + diff->cur_row[ci] = *(_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)1); + diff->prev_row[ci] = *(_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)1); + } + + /* Create the difference buffer. */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + diff->diff_buf[ci] = + ALLOC_DARRAY(JPOOL_IMAGE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)compptr->v_samp_factor); + /* Prefill difference rows with zeros. We do this because only actual + * data is placed in the buffers during prediction/differencing, leaving + * any dummy differences at the right edge as zeros, which will encode + * to the smallest amount of data. + */ + for (row = 0; row < compptr->v_samp_factor; row++) + memset(diff->diff_buf[ci][row], 0, + jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor) * sizeof(JDIFF)); + } + + /* Create the sample buffer. */ + if (need_full_buffer) { +#ifdef FULL_SAMP_BUFFER_SUPPORTED + /* Allocate a full-image virtual array for each component, */ + /* padded to a multiple of samp_factor differences in each direction. */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + diff->whole_image[ci] = (*cinfo->mem->request_virt_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, FALSE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)jround_up((long)compptr->height_in_blocks, + (long)compptr->v_samp_factor), + (JDIMENSION)compptr->v_samp_factor); + } +#else + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); +#endif + } else + diff->whole_image[0] = NULL; /* flag for no virtual arrays */ +} + +#endif /* C_LOSSLESS_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jchuff.c b/src/spice2x/external/libjpeg-turbo/jchuff.c new file mode 100644 index 0000000..8cdd5bd --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jchuff.c @@ -0,0 +1,1175 @@ +/* + * jchuff.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2009-2011, 2014-2016, 2018-2024, D. R. Commander. + * Copyright (C) 2015, Matthieu Darbois. + * Copyright (C) 2018, Matthias Räncker. + * Copyright (C) 2020, Arm Limited. + * Copyright (C) 2022, Felix Hanau. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains Huffman entropy encoding routines. + * + * Much of the complexity here has to do with supporting output suspension. + * If the data destination module demands suspension, we want to be able to + * back up to the start of the current MCU. To do this, we copy state + * variables into local working storage, and update them back to the + * permanent JPEG objects only upon successful completion of an MCU. + * + * NOTE: All referenced figures are from + * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#ifdef WITH_SIMD +#include "jsimd.h" +#else +#include "jchuff.h" /* Declarations shared with jc*huff.c */ +#endif +#include +#include "jpeg_nbits.h" + + +/* Expanded entropy encoder object for Huffman encoding. + * + * The savable_state subrecord contains fields that change within an MCU, + * but must not be updated permanently until we complete the MCU. + */ + +#if defined(__x86_64__) && defined(__ILP32__) +typedef unsigned long long bit_buf_type; +#else +typedef size_t bit_buf_type; +#endif + +/* NOTE: The more optimal Huffman encoding algorithm is only used by the + * intrinsics implementation of the Arm Neon SIMD extensions, which is why we + * retain the old Huffman encoder behavior when using the GAS implementation. + */ +#if defined(WITH_SIMD) && !(defined(__arm__) || defined(__aarch64__) || \ + defined(_M_ARM) || defined(_M_ARM64)) +typedef unsigned long long simd_bit_buf_type; +#else +typedef bit_buf_type simd_bit_buf_type; +#endif + +#if (defined(SIZEOF_SIZE_T) && SIZEOF_SIZE_T == 8) || defined(_WIN64) || \ + (defined(__x86_64__) && defined(__ILP32__)) +#define BIT_BUF_SIZE 64 +#elif (defined(SIZEOF_SIZE_T) && SIZEOF_SIZE_T == 4) || defined(_WIN32) +#define BIT_BUF_SIZE 32 +#else +#error Cannot determine word size +#endif +#define SIMD_BIT_BUF_SIZE (sizeof(simd_bit_buf_type) * 8) + +typedef struct { + union { + bit_buf_type c; +#ifdef WITH_SIMD + simd_bit_buf_type simd; +#endif + } put_buffer; /* current bit accumulation buffer */ + int free_bits; /* # of bits available in it */ + /* (Neon GAS: # of bits now in it) */ + int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */ +} savable_state; + +typedef struct { + struct jpeg_entropy_encoder pub; /* public fields */ + + savable_state saved; /* Bit buffer & DC state at start of MCU */ + + /* These fields are NOT loaded into local working state. */ + unsigned int restarts_to_go; /* MCUs left in this restart interval */ + int next_restart_num; /* next restart number to write (0-7) */ + + /* Pointers to derived tables (these workspaces have image lifespan) */ + c_derived_tbl *dc_derived_tbls[NUM_HUFF_TBLS]; + c_derived_tbl *ac_derived_tbls[NUM_HUFF_TBLS]; + +#ifdef ENTROPY_OPT_SUPPORTED /* Statistics tables for optimization */ + long *dc_count_ptrs[NUM_HUFF_TBLS]; + long *ac_count_ptrs[NUM_HUFF_TBLS]; +#endif + +#ifdef WITH_SIMD + int simd; +#endif +} huff_entropy_encoder; + +typedef huff_entropy_encoder *huff_entropy_ptr; + +/* Working state while writing an MCU. + * This struct contains all the fields that are needed by subroutines. + */ + +typedef struct { + JOCTET *next_output_byte; /* => next byte to write in buffer */ + size_t free_in_buffer; /* # of byte spaces remaining in buffer */ + savable_state cur; /* Current bit buffer & DC state */ + j_compress_ptr cinfo; /* dump_buffer needs access to this */ +#ifdef WITH_SIMD + int simd; +#endif +} working_state; + + +/* Forward declarations */ +METHODDEF(boolean) encode_mcu_huff(j_compress_ptr cinfo, JBLOCKROW *MCU_data); +METHODDEF(void) finish_pass_huff(j_compress_ptr cinfo); +#ifdef ENTROPY_OPT_SUPPORTED +METHODDEF(boolean) encode_mcu_gather(j_compress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(void) finish_pass_gather(j_compress_ptr cinfo); +#endif + + +/* + * Initialize for a Huffman-compressed scan. + * If gather_statistics is TRUE, we do not output anything during the scan, + * just count the Huffman symbols used and generate Huffman code tables. + */ + +METHODDEF(void) +start_pass_huff(j_compress_ptr cinfo, boolean gather_statistics) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + int ci, dctbl, actbl; + jpeg_component_info *compptr; + + if (gather_statistics) { +#ifdef ENTROPY_OPT_SUPPORTED + entropy->pub.encode_mcu = encode_mcu_gather; + entropy->pub.finish_pass = finish_pass_gather; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + entropy->pub.encode_mcu = encode_mcu_huff; + entropy->pub.finish_pass = finish_pass_huff; + } + +#ifdef WITH_SIMD + entropy->simd = jsimd_can_huff_encode_one_block(); +#endif + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + dctbl = compptr->dc_tbl_no; + actbl = compptr->ac_tbl_no; + if (gather_statistics) { +#ifdef ENTROPY_OPT_SUPPORTED + /* Check for invalid table indexes */ + /* (make_c_derived_tbl does this in the other path) */ + if (dctbl < 0 || dctbl >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, dctbl); + if (actbl < 0 || actbl >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, actbl); + /* Allocate and zero the statistics tables */ + /* Note that jpeg_gen_optimal_table expects 257 entries in each table! */ + if (entropy->dc_count_ptrs[dctbl] == NULL) + entropy->dc_count_ptrs[dctbl] = (long *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + 257 * sizeof(long)); + memset(entropy->dc_count_ptrs[dctbl], 0, 257 * sizeof(long)); + if (entropy->ac_count_ptrs[actbl] == NULL) + entropy->ac_count_ptrs[actbl] = (long *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + 257 * sizeof(long)); + memset(entropy->ac_count_ptrs[actbl], 0, 257 * sizeof(long)); +#endif + } else { + /* Compute derived values for Huffman tables */ + /* We may do this more than once for a table, but it's not expensive */ + jpeg_make_c_derived_tbl(cinfo, TRUE, dctbl, + &entropy->dc_derived_tbls[dctbl]); + jpeg_make_c_derived_tbl(cinfo, FALSE, actbl, + &entropy->ac_derived_tbls[actbl]); + } + /* Initialize DC predictions to 0 */ + entropy->saved.last_dc_val[ci] = 0; + } + + /* Initialize bit buffer to empty */ +#ifdef WITH_SIMD + if (entropy->simd) { + entropy->saved.put_buffer.simd = 0; +#if defined(__aarch64__) && !defined(NEON_INTRINSICS) + entropy->saved.free_bits = 0; +#else + entropy->saved.free_bits = SIMD_BIT_BUF_SIZE; +#endif + } else +#endif + { + entropy->saved.put_buffer.c = 0; + entropy->saved.free_bits = BIT_BUF_SIZE; + } + + /* Initialize restart stuff */ + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num = 0; +} + + +/* + * Compute the derived values for a Huffman table. + * This routine also performs some validation checks on the table. + * + * Note this is also used by jcphuff.c and jclhuff.c. + */ + +GLOBAL(void) +jpeg_make_c_derived_tbl(j_compress_ptr cinfo, boolean isDC, int tblno, + c_derived_tbl **pdtbl) +{ + JHUFF_TBL *htbl; + c_derived_tbl *dtbl; + int p, i, l, lastp, si, maxsymbol; + char huffsize[257]; + unsigned int huffcode[257]; + unsigned int code; + + /* Note that huffsize[] and huffcode[] are filled in code-length order, + * paralleling the order of the symbols themselves in htbl->huffval[]. + */ + + /* Find the input Huffman table */ + if (tblno < 0 || tblno >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno); + htbl = + isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno]; + if (htbl == NULL) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno); + + /* Allocate a workspace if we haven't already done so. */ + if (*pdtbl == NULL) + *pdtbl = (c_derived_tbl *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(c_derived_tbl)); + dtbl = *pdtbl; + + /* Figure C.1: make table of Huffman code length for each symbol */ + + p = 0; + for (l = 1; l <= 16; l++) { + i = (int)htbl->bits[l]; + if (i < 0 || p + i > 256) /* protect against table overrun */ + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + while (i--) + huffsize[p++] = (char)l; + } + huffsize[p] = 0; + lastp = p; + + /* Figure C.2: generate the codes themselves */ + /* We also validate that the counts represent a legal Huffman code tree. */ + + code = 0; + si = huffsize[0]; + p = 0; + while (huffsize[p]) { + while (((int)huffsize[p]) == si) { + huffcode[p++] = code; + code++; + } + /* code is now 1 more than the last code used for codelength si; but + * it must still fit in si bits, since no code is allowed to be all ones. + */ + if (((JLONG)code) >= (((JLONG)1) << si)) + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + code <<= 1; + si++; + } + + /* Figure C.3: generate encoding tables */ + /* These are code and size indexed by symbol value */ + + /* Set all codeless symbols to have code length 0; + * this lets us detect duplicate VAL entries here, and later + * allows emit_bits to detect any attempt to emit such symbols. + */ + memset(dtbl->ehufco, 0, sizeof(dtbl->ehufco)); + memset(dtbl->ehufsi, 0, sizeof(dtbl->ehufsi)); + + /* This is also a convenient place to check for out-of-range and duplicated + * VAL entries. We allow 0..255 for AC symbols but only 0..15 for DC in + * lossy mode and 0..16 for DC in lossless mode. (We could constrain them + * further based on data depth and mode, but this seems enough.) + */ + maxsymbol = isDC ? (cinfo->master->lossless ? 16 : 15) : 255; + + for (p = 0; p < lastp; p++) { + i = htbl->huffval[p]; + if (i < 0 || i > maxsymbol || dtbl->ehufsi[i]) + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + dtbl->ehufco[i] = huffcode[p]; + dtbl->ehufsi[i] = huffsize[p]; + } +} + + +/* Outputting bytes to the file */ + +/* Emit a byte, taking 'action' if must suspend. */ +#define emit_byte(state, val, action) { \ + *(state)->next_output_byte++ = (JOCTET)(val); \ + if (--(state)->free_in_buffer == 0) \ + if (!dump_buffer(state)) \ + { action; } \ +} + + +LOCAL(boolean) +dump_buffer(working_state *state) +/* Empty the output buffer; return TRUE if successful, FALSE if must suspend */ +{ + struct jpeg_destination_mgr *dest = state->cinfo->dest; + + if (!(*dest->empty_output_buffer) (state->cinfo)) + return FALSE; + /* After a successful buffer dump, must reset buffer pointers */ + state->next_output_byte = dest->next_output_byte; + state->free_in_buffer = dest->free_in_buffer; + return TRUE; +} + + +/* Outputting bits to the file */ + +/* Output byte b and, speculatively, an additional 0 byte. 0xFF must be + * encoded as 0xFF 0x00, so the output buffer pointer is advanced by 2 if the + * byte is 0xFF. Otherwise, the output buffer pointer is advanced by 1, and + * the speculative 0 byte will be overwritten by the next byte. + */ +#define EMIT_BYTE(b) { \ + buffer[0] = (JOCTET)(b); \ + buffer[1] = 0; \ + buffer -= -2 + ((JOCTET)(b) < 0xFF); \ +} + +/* Output the entire bit buffer. If there are no 0xFF bytes in it, then write + * directly to the output buffer. Otherwise, use the EMIT_BYTE() macro to + * encode 0xFF as 0xFF 0x00. + */ +#if BIT_BUF_SIZE == 64 + +#define FLUSH() { \ + if (put_buffer & 0x8080808080808080 & ~(put_buffer + 0x0101010101010101)) { \ + EMIT_BYTE(put_buffer >> 56) \ + EMIT_BYTE(put_buffer >> 48) \ + EMIT_BYTE(put_buffer >> 40) \ + EMIT_BYTE(put_buffer >> 32) \ + EMIT_BYTE(put_buffer >> 24) \ + EMIT_BYTE(put_buffer >> 16) \ + EMIT_BYTE(put_buffer >> 8) \ + EMIT_BYTE(put_buffer ) \ + } else { \ + buffer[0] = (JOCTET)(put_buffer >> 56); \ + buffer[1] = (JOCTET)(put_buffer >> 48); \ + buffer[2] = (JOCTET)(put_buffer >> 40); \ + buffer[3] = (JOCTET)(put_buffer >> 32); \ + buffer[4] = (JOCTET)(put_buffer >> 24); \ + buffer[5] = (JOCTET)(put_buffer >> 16); \ + buffer[6] = (JOCTET)(put_buffer >> 8); \ + buffer[7] = (JOCTET)(put_buffer); \ + buffer += 8; \ + } \ +} + +#else + +#define FLUSH() { \ + if (put_buffer & 0x80808080 & ~(put_buffer + 0x01010101)) { \ + EMIT_BYTE(put_buffer >> 24) \ + EMIT_BYTE(put_buffer >> 16) \ + EMIT_BYTE(put_buffer >> 8) \ + EMIT_BYTE(put_buffer ) \ + } else { \ + buffer[0] = (JOCTET)(put_buffer >> 24); \ + buffer[1] = (JOCTET)(put_buffer >> 16); \ + buffer[2] = (JOCTET)(put_buffer >> 8); \ + buffer[3] = (JOCTET)(put_buffer); \ + buffer += 4; \ + } \ +} + +#endif + +/* Fill the bit buffer to capacity with the leading bits from code, then output + * the bit buffer and put the remaining bits from code into the bit buffer. + */ +#define PUT_AND_FLUSH(code, size) { \ + put_buffer = (put_buffer << (size + free_bits)) | (code >> -free_bits); \ + FLUSH() \ + free_bits += BIT_BUF_SIZE; \ + put_buffer = code; \ +} + +/* Insert code into the bit buffer and output the bit buffer if needed. + * NOTE: We can't flush with free_bits == 0, since the left shift in + * PUT_AND_FLUSH() would have undefined behavior. + */ +#define PUT_BITS(code, size) { \ + free_bits -= size; \ + if (free_bits < 0) \ + PUT_AND_FLUSH(code, size) \ + else \ + put_buffer = (put_buffer << size) | code; \ +} + +#define PUT_CODE(code, size) { \ + temp &= (((JLONG)1) << nbits) - 1; \ + temp |= code << nbits; \ + nbits += size; \ + PUT_BITS(temp, nbits) \ +} + + +/* Although it is exceedingly rare, it is possible for a Huffman-encoded + * coefficient block to be larger than the 128-byte unencoded block. For each + * of the 64 coefficients, PUT_BITS is invoked twice, and each invocation can + * theoretically store 16 bits (for a maximum of 2048 bits or 256 bytes per + * encoded block.) If, for instance, one artificially sets the AC + * coefficients to alternating values of 32767 and -32768 (using the JPEG + * scanning order-- 1, 8, 16, etc.), then this will produce an encoded block + * larger than 200 bytes. + */ +#define BUFSIZE (DCTSIZE2 * 8) + +#define LOAD_BUFFER() { \ + if (state->free_in_buffer < BUFSIZE) { \ + localbuf = 1; \ + buffer = _buffer; \ + } else \ + buffer = state->next_output_byte; \ +} + +#define STORE_BUFFER() { \ + if (localbuf) { \ + size_t bytes, bytestocopy; \ + bytes = buffer - _buffer; \ + buffer = _buffer; \ + while (bytes > 0) { \ + bytestocopy = MIN(bytes, state->free_in_buffer); \ + memcpy(state->next_output_byte, buffer, bytestocopy); \ + state->next_output_byte += bytestocopy; \ + buffer += bytestocopy; \ + state->free_in_buffer -= bytestocopy; \ + if (state->free_in_buffer == 0) \ + if (!dump_buffer(state)) return FALSE; \ + bytes -= bytestocopy; \ + } \ + } else { \ + state->free_in_buffer -= (buffer - state->next_output_byte); \ + state->next_output_byte = buffer; \ + } \ +} + + +LOCAL(boolean) +flush_bits(working_state *state) +{ + JOCTET _buffer[BUFSIZE], *buffer, temp; + simd_bit_buf_type put_buffer; int put_bits; + int localbuf = 0; + +#ifdef WITH_SIMD + if (state->simd) { +#if defined(__aarch64__) && !defined(NEON_INTRINSICS) + put_bits = state->cur.free_bits; +#else + put_bits = SIMD_BIT_BUF_SIZE - state->cur.free_bits; +#endif + put_buffer = state->cur.put_buffer.simd; + } else +#endif + { + put_bits = BIT_BUF_SIZE - state->cur.free_bits; + put_buffer = state->cur.put_buffer.c; + } + + LOAD_BUFFER() + + while (put_bits >= 8) { + put_bits -= 8; + temp = (JOCTET)(put_buffer >> put_bits); + EMIT_BYTE(temp) + } + if (put_bits) { + /* fill partial byte with ones */ + temp = (JOCTET)((put_buffer << (8 - put_bits)) | (0xFF >> put_bits)); + EMIT_BYTE(temp) + } + +#ifdef WITH_SIMD + if (state->simd) { /* and reset bit buffer to empty */ + state->cur.put_buffer.simd = 0; +#if defined(__aarch64__) && !defined(NEON_INTRINSICS) + state->cur.free_bits = 0; +#else + state->cur.free_bits = SIMD_BIT_BUF_SIZE; +#endif + } else +#endif + { + state->cur.put_buffer.c = 0; + state->cur.free_bits = BIT_BUF_SIZE; + } + STORE_BUFFER() + + return TRUE; +} + + +#ifdef WITH_SIMD + +/* Encode a single block's worth of coefficients */ + +LOCAL(boolean) +encode_one_block_simd(working_state *state, JCOEFPTR block, int last_dc_val, + c_derived_tbl *dctbl, c_derived_tbl *actbl) +{ + JOCTET _buffer[BUFSIZE], *buffer; + int localbuf = 0; + +#ifdef ZERO_BUFFERS + memset(_buffer, 0, sizeof(_buffer)); +#endif + + LOAD_BUFFER() + + buffer = jsimd_huff_encode_one_block(state, buffer, block, last_dc_val, + dctbl, actbl); + + STORE_BUFFER() + + return TRUE; +} + +#endif + +LOCAL(boolean) +encode_one_block(working_state *state, JCOEFPTR block, int last_dc_val, + c_derived_tbl *dctbl, c_derived_tbl *actbl) +{ + int temp, nbits, free_bits; + bit_buf_type put_buffer; + JOCTET _buffer[BUFSIZE], *buffer; + int localbuf = 0; + int max_coef_bits = state->cinfo->data_precision + 2; + + free_bits = state->cur.free_bits; + put_buffer = state->cur.put_buffer.c; + LOAD_BUFFER() + + /* Encode the DC coefficient difference per section F.1.2.1 */ + + temp = block[0] - last_dc_val; + + /* This is a well-known technique for obtaining the absolute value without a + * branch. It is derived from an assembly language technique presented in + * "How to Optimize for the Pentium Processors", Copyright (c) 1996, 1997 by + * Agner Fog. This code assumes we are on a two's complement machine. + */ + nbits = temp >> (CHAR_BIT * sizeof(int) - 1); + temp += nbits; + nbits ^= temp; + + /* Find the number of bits needed for the magnitude of the coefficient */ + nbits = JPEG_NBITS(nbits); + /* Check for out-of-range coefficient values. + * Since we're encoding a difference, the range limit is twice as much. + */ + if (nbits > max_coef_bits + 1) + ERREXIT(state->cinfo, JERR_BAD_DCT_COEF); + + /* Emit the Huffman-coded symbol for the number of bits. + * Emit that number of bits of the value, if positive, + * or the complement of its magnitude, if negative. + */ + PUT_CODE(dctbl->ehufco[nbits], dctbl->ehufsi[nbits]) + + /* Encode the AC coefficients per section F.1.2.2 */ + + { + int r = 0; /* r = run length of zeros */ + +/* Manually unroll the k loop to eliminate the counter variable. This + * improves performance greatly on systems with a limited number of + * registers (such as x86.) + */ +#define kloop(jpeg_natural_order_of_k) { \ + if ((temp = block[jpeg_natural_order_of_k]) == 0) { \ + r += 16; \ + } else { \ + /* Branch-less absolute value, bitwise complement, etc., same as above */ \ + nbits = temp >> (CHAR_BIT * sizeof(int) - 1); \ + temp += nbits; \ + nbits ^= temp; \ + nbits = JPEG_NBITS_NONZERO(nbits); \ + /* Check for out-of-range coefficient values */ \ + if (nbits > max_coef_bits) \ + ERREXIT(state->cinfo, JERR_BAD_DCT_COEF); \ + /* if run length > 15, must emit special run-length-16 codes (0xF0) */ \ + while (r >= 16 * 16) { \ + r -= 16 * 16; \ + PUT_BITS(actbl->ehufco[0xf0], actbl->ehufsi[0xf0]) \ + } \ + /* Emit Huffman symbol for run length / number of bits */ \ + r += nbits; \ + PUT_CODE(actbl->ehufco[r], actbl->ehufsi[r]) \ + r = 0; \ + } \ +} + + /* One iteration for each value in jpeg_natural_order[] */ + kloop(1); kloop(8); kloop(16); kloop(9); kloop(2); kloop(3); + kloop(10); kloop(17); kloop(24); kloop(32); kloop(25); kloop(18); + kloop(11); kloop(4); kloop(5); kloop(12); kloop(19); kloop(26); + kloop(33); kloop(40); kloop(48); kloop(41); kloop(34); kloop(27); + kloop(20); kloop(13); kloop(6); kloop(7); kloop(14); kloop(21); + kloop(28); kloop(35); kloop(42); kloop(49); kloop(56); kloop(57); + kloop(50); kloop(43); kloop(36); kloop(29); kloop(22); kloop(15); + kloop(23); kloop(30); kloop(37); kloop(44); kloop(51); kloop(58); + kloop(59); kloop(52); kloop(45); kloop(38); kloop(31); kloop(39); + kloop(46); kloop(53); kloop(60); kloop(61); kloop(54); kloop(47); + kloop(55); kloop(62); kloop(63); + + /* If the last coef(s) were zero, emit an end-of-block code */ + if (r > 0) { + PUT_BITS(actbl->ehufco[0], actbl->ehufsi[0]) + } + } + + state->cur.put_buffer.c = put_buffer; + state->cur.free_bits = free_bits; + STORE_BUFFER() + + return TRUE; +} + + +/* + * Emit a restart marker & resynchronize predictions. + */ + +LOCAL(boolean) +emit_restart(working_state *state, int restart_num) +{ + int ci; + + if (!flush_bits(state)) + return FALSE; + + emit_byte(state, 0xFF, return FALSE); + emit_byte(state, JPEG_RST0 + restart_num, return FALSE); + + /* Re-initialize DC predictions to 0 */ + for (ci = 0; ci < state->cinfo->comps_in_scan; ci++) + state->cur.last_dc_val[ci] = 0; + + /* The restart counter is not updated until we successfully write the MCU. */ + + return TRUE; +} + + +/* + * Encode and output one MCU's worth of Huffman-compressed coefficients. + */ + +METHODDEF(boolean) +encode_mcu_huff(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + working_state state; + int blkn, ci; + jpeg_component_info *compptr; + + /* Load up working state */ + state.next_output_byte = cinfo->dest->next_output_byte; + state.free_in_buffer = cinfo->dest->free_in_buffer; + state.cur = entropy->saved; + state.cinfo = cinfo; +#ifdef WITH_SIMD + state.simd = entropy->simd; +#endif + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + if (!emit_restart(&state, entropy->next_restart_num)) + return FALSE; + } + + /* Encode the MCU data blocks */ +#ifdef WITH_SIMD + if (entropy->simd) { + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + if (!encode_one_block_simd(&state, + MCU_data[blkn][0], state.cur.last_dc_val[ci], + entropy->dc_derived_tbls[compptr->dc_tbl_no], + entropy->ac_derived_tbls[compptr->ac_tbl_no])) + return FALSE; + /* Update last_dc_val */ + state.cur.last_dc_val[ci] = MCU_data[blkn][0][0]; + } + } else +#endif + { + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + if (!encode_one_block(&state, + MCU_data[blkn][0], state.cur.last_dc_val[ci], + entropy->dc_derived_tbls[compptr->dc_tbl_no], + entropy->ac_derived_tbls[compptr->ac_tbl_no])) + return FALSE; + /* Update last_dc_val */ + state.cur.last_dc_val[ci] = MCU_data[blkn][0][0]; + } + } + + /* Completed MCU, so update state */ + cinfo->dest->next_output_byte = state.next_output_byte; + cinfo->dest->free_in_buffer = state.free_in_buffer; + entropy->saved = state.cur; + + /* Update restart-interval state too */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + return TRUE; +} + + +/* + * Finish up at the end of a Huffman-compressed scan. + */ + +METHODDEF(void) +finish_pass_huff(j_compress_ptr cinfo) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + working_state state; + + /* Load up working state ... flush_bits needs it */ + state.next_output_byte = cinfo->dest->next_output_byte; + state.free_in_buffer = cinfo->dest->free_in_buffer; + state.cur = entropy->saved; + state.cinfo = cinfo; +#ifdef WITH_SIMD + state.simd = entropy->simd; +#endif + + /* Flush out the last data */ + if (!flush_bits(&state)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); + + /* Update state */ + cinfo->dest->next_output_byte = state.next_output_byte; + cinfo->dest->free_in_buffer = state.free_in_buffer; + entropy->saved = state.cur; +} + + +/* + * Huffman coding optimization. + * + * We first scan the supplied data and count the number of uses of each symbol + * that is to be Huffman-coded. (This process MUST agree with the code above.) + * Then we build a Huffman coding tree for the observed counts. + * Symbols which are not needed at all for the particular image are not + * assigned any code, which saves space in the DHT marker as well as in + * the compressed data. + */ + +#ifdef ENTROPY_OPT_SUPPORTED + + +/* Process a single block's worth of coefficients */ + +LOCAL(void) +htest_one_block(j_compress_ptr cinfo, JCOEFPTR block, int last_dc_val, + long dc_counts[], long ac_counts[]) +{ + register int temp; + register int nbits; + register int k, r; + int max_coef_bits = cinfo->data_precision + 2; + + /* Encode the DC coefficient difference per section F.1.2.1 */ + + temp = block[0] - last_dc_val; + if (temp < 0) + temp = -temp; + + /* Find the number of bits needed for the magnitude of the coefficient */ + nbits = 0; + while (temp) { + nbits++; + temp >>= 1; + } + /* Check for out-of-range coefficient values. + * Since we're encoding a difference, the range limit is twice as much. + */ + if (nbits > max_coef_bits + 1) + ERREXIT(cinfo, JERR_BAD_DCT_COEF); + + /* Count the Huffman symbol for the number of bits */ + dc_counts[nbits]++; + + /* Encode the AC coefficients per section F.1.2.2 */ + + r = 0; /* r = run length of zeros */ + + for (k = 1; k < DCTSIZE2; k++) { + if ((temp = block[jpeg_natural_order[k]]) == 0) { + r++; + } else { + /* if run length > 15, must emit special run-length-16 codes (0xF0) */ + while (r > 15) { + ac_counts[0xF0]++; + r -= 16; + } + + /* Find the number of bits needed for the magnitude of the coefficient */ + if (temp < 0) + temp = -temp; + + /* Find the number of bits needed for the magnitude of the coefficient */ + nbits = 1; /* there must be at least one 1 bit */ + while ((temp >>= 1)) + nbits++; + /* Check for out-of-range coefficient values */ + if (nbits > max_coef_bits) + ERREXIT(cinfo, JERR_BAD_DCT_COEF); + + /* Count Huffman symbol for run length / number of bits */ + ac_counts[(r << 4) + nbits]++; + + r = 0; + } + } + + /* If the last coef(s) were zero, emit an end-of-block code */ + if (r > 0) + ac_counts[0]++; +} + + +/* + * Trial-encode one MCU's worth of Huffman-compressed coefficients. + * No data is actually output, so no suspension return is possible. + */ + +METHODDEF(boolean) +encode_mcu_gather(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + int blkn, ci; + jpeg_component_info *compptr; + + /* Take care of restart intervals if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + /* Re-initialize DC predictions to 0 */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) + entropy->saved.last_dc_val[ci] = 0; + /* Update restart state */ + entropy->restarts_to_go = cinfo->restart_interval; + } + entropy->restarts_to_go--; + } + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + htest_one_block(cinfo, MCU_data[blkn][0], entropy->saved.last_dc_val[ci], + entropy->dc_count_ptrs[compptr->dc_tbl_no], + entropy->ac_count_ptrs[compptr->ac_tbl_no]); + entropy->saved.last_dc_val[ci] = MCU_data[blkn][0][0]; + } + + return TRUE; +} + + +/* + * Generate the best Huffman code table for the given counts, fill htbl. + * Note this is also used by jcphuff.c and jclhuff.c. + * + * The JPEG standard requires that no symbol be assigned a codeword of all + * one bits (so that padding bits added at the end of a compressed segment + * can't look like a valid code). Because of the canonical ordering of + * codewords, this just means that there must be an unused slot in the + * longest codeword length category. Annex K (Clause K.2) of + * Rec. ITU-T T.81 (1992) | ISO/IEC 10918-1:1994 suggests reserving such a slot + * by pretending that symbol 256 is a valid symbol with count 1. In theory + * that's not optimal; giving it count zero but including it in the symbol set + * anyway should give a better Huffman code. But the theoretically better code + * actually seems to come out worse in practice, because it produces more + * all-ones bytes (which incur stuffed zero bytes in the final file). In any + * case the difference is tiny. + * + * The JPEG standard requires Huffman codes to be no more than 16 bits long. + * If some symbols have a very small but nonzero probability, the Huffman tree + * must be adjusted to meet the code length restriction. We currently use + * the adjustment method suggested in JPEG section K.2. This method is *not* + * optimal; it may not choose the best possible limited-length code. But + * typically only very-low-frequency symbols will be given less-than-optimal + * lengths, so the code is almost optimal. Experimental comparisons against + * an optimal limited-length-code algorithm indicate that the difference is + * microscopic --- usually less than a hundredth of a percent of total size. + * So the extra complexity of an optimal algorithm doesn't seem worthwhile. + */ + +GLOBAL(void) +jpeg_gen_optimal_table(j_compress_ptr cinfo, JHUFF_TBL *htbl, long freq[]) +{ +#define MAX_CLEN 32 /* assumed maximum initial code length */ + UINT8 bits[MAX_CLEN + 1]; /* bits[k] = # of symbols with code length k */ + int bit_pos[MAX_CLEN + 1]; /* # of symbols with smaller code length */ + int codesize[257]; /* codesize[k] = code length of symbol k */ + int nz_index[257]; /* index of nonzero symbol in the original freq + array */ + int others[257]; /* next symbol in current branch of tree */ + int c1, c2; + int p, i, j; + int num_nz_symbols; + long v, v2; + + /* This algorithm is explained in section K.2 of the JPEG standard */ + + memset(bits, 0, sizeof(bits)); + memset(codesize, 0, sizeof(codesize)); + for (i = 0; i < 257; i++) + others[i] = -1; /* init links to empty */ + + freq[256] = 1; /* make sure 256 has a nonzero count */ + /* Including the pseudo-symbol 256 in the Huffman procedure guarantees + * that no real symbol is given code-value of all ones, because 256 + * will be placed last in the largest codeword category. + */ + + /* Group nonzero frequencies together so we can more easily find the + * smallest. + */ + num_nz_symbols = 0; + for (i = 0; i < 257; i++) { + if (freq[i]) { + nz_index[num_nz_symbols] = i; + freq[num_nz_symbols] = freq[i]; + num_nz_symbols++; + } + } + + /* Huffman's basic algorithm to assign optimal code lengths to symbols */ + + for (;;) { + /* Find the two smallest nonzero frequencies; set c1, c2 = their symbols */ + /* In case of ties, take the larger symbol number. Since we have grouped + * the nonzero symbols together, checking for zero symbols is not + * necessary. + */ + c1 = -1; + c2 = -1; + v = 1000000000L; + v2 = 1000000000L; + for (i = 0; i < num_nz_symbols; i++) { + if (freq[i] <= v2) { + if (freq[i] <= v) { + c2 = c1; + v2 = v; + v = freq[i]; + c1 = i; + } else { + v2 = freq[i]; + c2 = i; + } + } + } + + /* Done if we've merged everything into one frequency */ + if (c2 < 0) + break; + + /* Else merge the two counts/trees */ + freq[c1] += freq[c2]; + /* Set the frequency to a very high value instead of zero, so we don't have + * to check for zero values. + */ + freq[c2] = 1000000001L; + + /* Increment the codesize of everything in c1's tree branch */ + codesize[c1]++; + while (others[c1] >= 0) { + c1 = others[c1]; + codesize[c1]++; + } + + others[c1] = c2; /* chain c2 onto c1's tree branch */ + + /* Increment the codesize of everything in c2's tree branch */ + codesize[c2]++; + while (others[c2] >= 0) { + c2 = others[c2]; + codesize[c2]++; + } + } + + /* Now count the number of symbols of each code length */ + for (i = 0; i < num_nz_symbols; i++) { + /* The JPEG standard seems to think that this can't happen, */ + /* but I'm paranoid... */ + if (codesize[i] > MAX_CLEN) + ERREXIT(cinfo, JERR_HUFF_CLEN_OVERFLOW); + + bits[codesize[i]]++; + } + + /* Count the number of symbols with a length smaller than i bits, so we can + * construct the symbol table more efficiently. Note that this includes the + * pseudo-symbol 256, but since it is the last symbol, it will not affect the + * table. + */ + p = 0; + for (i = 1; i <= MAX_CLEN; i++) { + bit_pos[i] = p; + p += bits[i]; + } + + /* JPEG doesn't allow symbols with code lengths over 16 bits, so if the pure + * Huffman procedure assigned any such lengths, we must adjust the coding. + * Here is what Rec. ITU-T T.81 | ISO/IEC 10918-1 says about how this next + * bit works: Since symbols are paired for the longest Huffman code, the + * symbols are removed from this length category two at a time. The prefix + * for the pair (which is one bit shorter) is allocated to one of the pair; + * then, skipping the BITS entry for that prefix length, a code word from the + * next shortest nonzero BITS entry is converted into a prefix for two code + * words one bit longer. + */ + + for (i = MAX_CLEN; i > 16; i--) { + while (bits[i] > 0) { + j = i - 2; /* find length of new prefix to be used */ + while (bits[j] == 0) + j--; + + bits[i] -= 2; /* remove two symbols */ + bits[i - 1]++; /* one goes in this length */ + bits[j + 1] += 2; /* two new symbols in this length */ + bits[j]--; /* symbol of this length is now a prefix */ + } + } + + /* Remove the count for the pseudo-symbol 256 from the largest codelength */ + while (bits[i] == 0) /* find largest codelength still in use */ + i--; + bits[i]--; + + /* Return final symbol counts (only for lengths 0..16) */ + memcpy(htbl->bits, bits, sizeof(htbl->bits)); + + /* Return a list of the symbols sorted by code length */ + /* It's not real clear to me why we don't need to consider the codelength + * changes made above, but Rec. ITU-T T.81 | ISO/IEC 10918-1 seems to think + * this works. + */ + for (i = 0; i < num_nz_symbols - 1; i++) { + htbl->huffval[bit_pos[codesize[i]]] = (UINT8)nz_index[i]; + bit_pos[codesize[i]]++; + } + + /* Set sent_table FALSE so updated table will be written to JPEG file. */ + htbl->sent_table = FALSE; +} + + +/* + * Finish up a statistics-gathering pass and create the new Huffman tables. + */ + +METHODDEF(void) +finish_pass_gather(j_compress_ptr cinfo) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + int ci, dctbl, actbl; + jpeg_component_info *compptr; + JHUFF_TBL **htblptr; + boolean did_dc[NUM_HUFF_TBLS]; + boolean did_ac[NUM_HUFF_TBLS]; + + /* It's important not to apply jpeg_gen_optimal_table more than once + * per table, because it clobbers the input frequency counts! + */ + memset(did_dc, 0, sizeof(did_dc)); + memset(did_ac, 0, sizeof(did_ac)); + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + dctbl = compptr->dc_tbl_no; + actbl = compptr->ac_tbl_no; + if (!did_dc[dctbl]) { + htblptr = &cinfo->dc_huff_tbl_ptrs[dctbl]; + if (*htblptr == NULL) + *htblptr = jpeg_alloc_huff_table((j_common_ptr)cinfo); + jpeg_gen_optimal_table(cinfo, *htblptr, entropy->dc_count_ptrs[dctbl]); + did_dc[dctbl] = TRUE; + } + if (!did_ac[actbl]) { + htblptr = &cinfo->ac_huff_tbl_ptrs[actbl]; + if (*htblptr == NULL) + *htblptr = jpeg_alloc_huff_table((j_common_ptr)cinfo); + jpeg_gen_optimal_table(cinfo, *htblptr, entropy->ac_count_ptrs[actbl]); + did_ac[actbl] = TRUE; + } + } +} + + +#endif /* ENTROPY_OPT_SUPPORTED */ + + +/* + * Module initialization routine for Huffman entropy encoding. + */ + +GLOBAL(void) +jinit_huff_encoder(j_compress_ptr cinfo) +{ + huff_entropy_ptr entropy; + int i; + + entropy = (huff_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(huff_entropy_encoder)); + cinfo->entropy = (struct jpeg_entropy_encoder *)entropy; + entropy->pub.start_pass = start_pass_huff; + + /* Mark tables unallocated */ + for (i = 0; i < NUM_HUFF_TBLS; i++) { + entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL; +#ifdef ENTROPY_OPT_SUPPORTED + entropy->dc_count_ptrs[i] = entropy->ac_count_ptrs[i] = NULL; +#endif + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jchuff.h b/src/spice2x/external/libjpeg-turbo/jchuff.h new file mode 100644 index 0000000..21f17b8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jchuff.h @@ -0,0 +1,44 @@ +/* + * jchuff.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains declarations for Huffman entropy encoding routines + * that are shared between the sequential encoder (jchuff.c) and the + * progressive encoder (jcphuff.c). No other modules need to see these. + */ + +/* The legal range of a DCT coefficient is + * -1024 .. +1023 for 8-bit data; + * -16384 .. +16383 for 12-bit data. + * Hence the magnitude should always fit in 10 or 14 bits respectively. + */ + +/* The progressive Huffman encoder uses an unsigned 16-bit data type to store + * absolute values of coefficients, because it is possible to inject a + * coefficient value of -32768 into the encoder by attempting to transform a + * malformed 12-bit JPEG image, and the absolute value of -32768 would overflow + * a signed 16-bit integer. + */ +typedef unsigned short UJCOEF; + +/* Derived data constructed for each Huffman table */ + +typedef struct { + unsigned int ehufco[256]; /* code for each symbol */ + char ehufsi[256]; /* length of code for each symbol */ + /* If no code has been allocated for a symbol S, ehufsi[S] contains 0 */ +} c_derived_tbl; + +/* Expand a Huffman table definition into the derived format */ +EXTERN(void) jpeg_make_c_derived_tbl(j_compress_ptr cinfo, boolean isDC, + int tblno, c_derived_tbl **pdtbl); + +/* Generate an optimal table definition given the specified counts */ +EXTERN(void) jpeg_gen_optimal_table(j_compress_ptr cinfo, JHUFF_TBL *htbl, + long freq[]); diff --git a/src/spice2x/external/libjpeg-turbo/jcicc.c b/src/spice2x/external/libjpeg-turbo/jcicc.c new file mode 100644 index 0000000..11037ff --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcicc.c @@ -0,0 +1,105 @@ +/* + * jcicc.c + * + * Copyright (C) 1997-1998, Thomas G. Lane, Todd Newman. + * Copyright (C) 2017, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file provides code to write International Color Consortium (ICC) device + * profiles embedded in JFIF JPEG image files. The ICC has defined a standard + * for including such data in JPEG "APP2" markers. The code given here does + * not know anything about the internal structure of the ICC profile data; it + * just knows how to embed the profile data in a JPEG file while writing it. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jerror.h" + + +/* + * Since an ICC profile can be larger than the maximum size of a JPEG marker + * (64K), we need provisions to split it into multiple markers. The format + * defined by the ICC specifies one or more APP2 markers containing the + * following data: + * Identifying string ASCII "ICC_PROFILE\0" (12 bytes) + * Marker sequence number 1 for first APP2, 2 for next, etc (1 byte) + * Number of markers Total number of APP2's used (1 byte) + * Profile data (remainder of APP2 data) + * Decoders should use the marker sequence numbers to reassemble the profile, + * rather than assuming that the APP2 markers appear in the correct sequence. + */ + +#define ICC_MARKER (JPEG_APP0 + 2) /* JPEG marker code for ICC */ +#define ICC_OVERHEAD_LEN 14 /* size of non-profile data in APP2 */ +#define MAX_BYTES_IN_MARKER 65533 /* maximum data len of a JPEG marker */ +#define MAX_DATA_BYTES_IN_MARKER (MAX_BYTES_IN_MARKER - ICC_OVERHEAD_LEN) + + +/* + * This routine writes the given ICC profile data into a JPEG file. It *must* + * be called AFTER calling jpeg_start_compress() and BEFORE the first call to + * jpeg_write_scanlines(). (This ordering ensures that the APP2 marker(s) will + * appear after the SOI and JFIF or Adobe markers, but before all else.) + */ + +GLOBAL(void) +jpeg_write_icc_profile(j_compress_ptr cinfo, const JOCTET *icc_data_ptr, + unsigned int icc_data_len) +{ + unsigned int num_markers; /* total number of markers we'll write */ + int cur_marker = 1; /* per spec, counting starts at 1 */ + unsigned int length; /* number of bytes to write in this marker */ + + if (icc_data_ptr == NULL || icc_data_len == 0) + ERREXIT(cinfo, JERR_BUFFER_SIZE); + if (cinfo->global_state < CSTATE_SCANNING) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + /* Calculate the number of markers we'll need, rounding up of course */ + num_markers = icc_data_len / MAX_DATA_BYTES_IN_MARKER; + if (num_markers * MAX_DATA_BYTES_IN_MARKER != icc_data_len) + num_markers++; + + while (icc_data_len > 0) { + /* length of profile to put in this marker */ + length = icc_data_len; + if (length > MAX_DATA_BYTES_IN_MARKER) + length = MAX_DATA_BYTES_IN_MARKER; + icc_data_len -= length; + + /* Write the JPEG marker header (APP2 code and marker length) */ + jpeg_write_m_header(cinfo, ICC_MARKER, + (unsigned int)(length + ICC_OVERHEAD_LEN)); + + /* Write the marker identifying string "ICC_PROFILE" (null-terminated). We + * code it in this less-than-transparent way so that the code works even if + * the local character set is not ASCII. + */ + jpeg_write_m_byte(cinfo, 0x49); + jpeg_write_m_byte(cinfo, 0x43); + jpeg_write_m_byte(cinfo, 0x43); + jpeg_write_m_byte(cinfo, 0x5F); + jpeg_write_m_byte(cinfo, 0x50); + jpeg_write_m_byte(cinfo, 0x52); + jpeg_write_m_byte(cinfo, 0x4F); + jpeg_write_m_byte(cinfo, 0x46); + jpeg_write_m_byte(cinfo, 0x49); + jpeg_write_m_byte(cinfo, 0x4C); + jpeg_write_m_byte(cinfo, 0x45); + jpeg_write_m_byte(cinfo, 0x0); + + /* Add the sequencing info */ + jpeg_write_m_byte(cinfo, cur_marker); + jpeg_write_m_byte(cinfo, (int)num_markers); + + /* Add the profile data */ + while (length--) { + jpeg_write_m_byte(cinfo, *icc_data_ptr); + icc_data_ptr++; + } + cur_marker++; + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jcinit.c b/src/spice2x/external/libjpeg-turbo/jcinit.c new file mode 100644 index 0000000..fe8a13a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcinit.c @@ -0,0 +1,149 @@ +/* + * jcinit.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains initialization logic for the JPEG compressor. + * This routine is in charge of selecting the modules to be executed and + * making an initialization call to each one. + * + * Logically, this code belongs in jcmaster.c. It's split out because + * linking this routine implies linking the entire compression library. + * For a transcoding-only application, we want to be able to use jcmaster.c + * without linking in the whole library. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jpegapicomp.h" + + +/* + * Master selection of compression modules. + * This is done once at the start of processing an image. We determine + * which modules will be used and give them appropriate initialization calls. + */ + +GLOBAL(void) +jinit_compress_master(j_compress_ptr cinfo) +{ + /* Initialize master control (includes parameter checking/processing) */ + jinit_c_master_control(cinfo, FALSE /* full compression */); + + /* Preprocessing */ + if (!cinfo->raw_data_in) { + if (cinfo->data_precision == 16) { +#ifdef C_LOSSLESS_SUPPORTED + j16init_color_converter(cinfo); + j16init_downsampler(cinfo); + j16init_c_prep_controller(cinfo, + FALSE /* never need full buffer here */); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); +#endif + } else if (cinfo->data_precision == 12) { + j12init_color_converter(cinfo); + j12init_downsampler(cinfo); + j12init_c_prep_controller(cinfo, + FALSE /* never need full buffer here */); + } else { + jinit_color_converter(cinfo); + jinit_downsampler(cinfo); + jinit_c_prep_controller(cinfo, FALSE /* never need full buffer here */); + } + } + + if (cinfo->master->lossless) { +#ifdef C_LOSSLESS_SUPPORTED + /* Prediction, sample differencing, and point transform */ + if (cinfo->data_precision == 16) + j16init_lossless_compressor(cinfo); + else if (cinfo->data_precision == 12) + j12init_lossless_compressor(cinfo); + else + jinit_lossless_compressor(cinfo); + /* Entropy encoding: either Huffman or arithmetic coding. */ + if (cinfo->arith_code) { + ERREXIT(cinfo, JERR_ARITH_NOTIMPL); + } else { + jinit_lhuff_encoder(cinfo); + } + + /* Need a full-image difference buffer in any multi-pass mode. */ + if (cinfo->data_precision == 16) + j16init_c_diff_controller(cinfo, (boolean)(cinfo->num_scans > 1 || + cinfo->optimize_coding)); + else if (cinfo->data_precision == 12) + j12init_c_diff_controller(cinfo, (boolean)(cinfo->num_scans > 1 || + cinfo->optimize_coding)); + else + jinit_c_diff_controller(cinfo, (boolean)(cinfo->num_scans > 1 || + cinfo->optimize_coding)); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + if (cinfo->data_precision == 16) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + /* Forward DCT */ + if (cinfo->data_precision == 12) + j12init_forward_dct(cinfo); + else + jinit_forward_dct(cinfo); + /* Entropy encoding: either Huffman or arithmetic coding. */ + if (cinfo->arith_code) { +#ifdef C_ARITH_CODING_SUPPORTED + jinit_arith_encoder(cinfo); +#else + ERREXIT(cinfo, JERR_ARITH_NOTIMPL); +#endif + } else { + if (cinfo->progressive_mode) { +#ifdef C_PROGRESSIVE_SUPPORTED + jinit_phuff_encoder(cinfo); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else + jinit_huff_encoder(cinfo); + } + + /* Need a full-image coefficient buffer in any multi-pass mode. */ + if (cinfo->data_precision == 12) + j12init_c_coef_controller(cinfo, (boolean)(cinfo->num_scans > 1 || + cinfo->optimize_coding)); + else + jinit_c_coef_controller(cinfo, (boolean)(cinfo->num_scans > 1 || + cinfo->optimize_coding)); + } + + if (cinfo->data_precision == 16) +#ifdef C_LOSSLESS_SUPPORTED + j16init_c_main_controller(cinfo, FALSE /* never need full buffer here */); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); +#endif + else if (cinfo->data_precision == 12) + j12init_c_main_controller(cinfo, FALSE /* never need full buffer here */); + else + jinit_c_main_controller(cinfo, FALSE /* never need full buffer here */); + + jinit_marker_writer(cinfo); + + /* We can now tell the memory manager to allocate virtual arrays. */ + (*cinfo->mem->realize_virt_arrays) ((j_common_ptr)cinfo); + + /* Write the datastream header (SOI) immediately. + * Frame and scan headers are postponed till later. + * This lets application insert special markers after the SOI. + */ + (*cinfo->marker->write_file_header) (cinfo); +} diff --git a/src/spice2x/external/libjpeg-turbo/jclhuff.c b/src/spice2x/external/libjpeg-turbo/jclhuff.c new file mode 100644 index 0000000..ae41545 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jclhuff.c @@ -0,0 +1,587 @@ +/* + * jclhuff.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains Huffman entropy encoding routines for lossless JPEG. + * + * Much of the complexity here has to do with supporting output suspension. + * If the data destination module demands suspension, we want to be able to + * back up to the start of the current MCU. To do this, we copy state + * variables into local working storage, and update them back to the + * permanent JPEG objects only upon successful completion of an MCU. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jlossls.h" /* Private declarations for lossless codec */ +#include "jchuff.h" /* Declarations shared with jc*huff.c */ + + +#ifdef C_LOSSLESS_SUPPORTED + +/* The legal range of a spatial difference is + * -32767 .. +32768. + * Hence the magnitude should always fit in 16 bits. + */ + +#define MAX_DIFF_BITS 16 + + +/* Expanded entropy encoder object for Huffman encoding in lossless mode. + * + * The savable_state subrecord contains fields that change within an MCU, + * but must not be updated permanently until we complete the MCU. + */ + +typedef struct { + size_t put_buffer; /* current bit-accumulation buffer */ + int put_bits; /* # of bits now in it */ +} savable_state; + + +typedef struct { + int ci, yoffset, MCU_width; +} lhe_input_ptr_info; + + +typedef struct { + struct jpeg_entropy_encoder pub; /* public fields */ + + savable_state saved; /* Bit buffer at start of MCU */ + + /* These fields are NOT loaded into local working state. */ + unsigned int restarts_to_go; /* MCUs left in this restart interval */ + int next_restart_num; /* next restart number to write (0-7) */ + + /* Pointers to derived tables (these workspaces have image lifespan) */ + c_derived_tbl *derived_tbls[NUM_HUFF_TBLS]; + + /* Pointers to derived tables to be used for each data unit within an MCU */ + c_derived_tbl *cur_tbls[C_MAX_BLOCKS_IN_MCU]; + +#ifdef ENTROPY_OPT_SUPPORTED /* Statistics tables for optimization */ + long *count_ptrs[NUM_HUFF_TBLS]; + + /* Pointers to stats tables to be used for each data unit within an MCU */ + long *cur_counts[C_MAX_BLOCKS_IN_MCU]; +#endif + + /* Pointers to the proper input difference row for each group of data units + * within an MCU. For each component, there are Vi groups of Hi data units. + */ + JDIFFROW input_ptr[C_MAX_BLOCKS_IN_MCU]; + + /* Number of input pointers in use for the current MCU. This is the sum + * of all Vi in the MCU. + */ + int num_input_ptrs; + + /* Information used for positioning the input pointers within the input + * difference rows. + */ + lhe_input_ptr_info input_ptr_info[C_MAX_BLOCKS_IN_MCU]; + + /* Index of the proper input pointer for each data unit within an MCU */ + int input_ptr_index[C_MAX_BLOCKS_IN_MCU]; + +} lhuff_entropy_encoder; + +typedef lhuff_entropy_encoder *lhuff_entropy_ptr; + +/* Working state while writing an MCU. + * This struct contains all the fields that are needed by subroutines. + */ + +typedef struct { + JOCTET *next_output_byte; /* => next byte to write in buffer */ + size_t free_in_buffer; /* # of byte spaces remaining in buffer */ + savable_state cur; /* Current bit buffer & DC state */ + j_compress_ptr cinfo; /* dump_buffer needs access to this */ +} working_state; + + +/* Forward declarations */ +METHODDEF(JDIMENSION) encode_mcus_huff(j_compress_ptr cinfo, + JDIFFIMAGE diff_buf, + JDIMENSION MCU_row_num, + JDIMENSION MCU_col_num, + JDIMENSION nMCU); +METHODDEF(void) finish_pass_huff(j_compress_ptr cinfo); +#ifdef ENTROPY_OPT_SUPPORTED +METHODDEF(JDIMENSION) encode_mcus_gather(j_compress_ptr cinfo, + JDIFFIMAGE diff_buf, + JDIMENSION MCU_row_num, + JDIMENSION MCU_col_num, + JDIMENSION nMCU); +METHODDEF(void) finish_pass_gather(j_compress_ptr cinfo); +#endif + + +/* + * Initialize for a Huffman-compressed scan. + * If gather_statistics is TRUE, we do not output anything during the scan, + * just count the Huffman symbols used and generate Huffman code tables. + */ + +METHODDEF(void) +start_pass_lhuff(j_compress_ptr cinfo, boolean gather_statistics) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + int ci, dctbl, sampn, ptrn, yoffset, xoffset; + jpeg_component_info *compptr; + + if (gather_statistics) { +#ifdef ENTROPY_OPT_SUPPORTED + entropy->pub.encode_mcus = encode_mcus_gather; + entropy->pub.finish_pass = finish_pass_gather; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + entropy->pub.encode_mcus = encode_mcus_huff; + entropy->pub.finish_pass = finish_pass_huff; + } + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + dctbl = compptr->dc_tbl_no; + if (gather_statistics) { +#ifdef ENTROPY_OPT_SUPPORTED + /* Check for invalid table indexes */ + /* (make_c_derived_tbl does this in the other path) */ + if (dctbl < 0 || dctbl >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, dctbl); + /* Allocate and zero the statistics tables */ + /* Note that jpeg_gen_optimal_table expects 257 entries in each table! */ + if (entropy->count_ptrs[dctbl] == NULL) + entropy->count_ptrs[dctbl] = (long *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + 257 * sizeof(long)); + memset(entropy->count_ptrs[dctbl], 0, 257 * sizeof(long)); +#endif + } else { + /* Compute derived values for Huffman tables */ + /* We may do this more than once for a table, but it's not expensive */ + jpeg_make_c_derived_tbl(cinfo, TRUE, dctbl, + &entropy->derived_tbls[dctbl]); + } + } + + /* Precalculate encoding info for each sample in an MCU of this scan */ + for (sampn = 0, ptrn = 0; sampn < cinfo->blocks_in_MCU;) { + compptr = cinfo->cur_comp_info[cinfo->MCU_membership[sampn]]; + ci = compptr->component_index; + for (yoffset = 0; yoffset < compptr->MCU_height; yoffset++, ptrn++) { + /* Precalculate the setup info for each input pointer */ + entropy->input_ptr_info[ptrn].ci = ci; + entropy->input_ptr_info[ptrn].yoffset = yoffset; + entropy->input_ptr_info[ptrn].MCU_width = compptr->MCU_width; + for (xoffset = 0; xoffset < compptr->MCU_width; xoffset++, sampn++) { + /* Precalculate the input pointer index for each sample */ + entropy->input_ptr_index[sampn] = ptrn; + /* Precalculate which tables to use for each sample */ + entropy->cur_tbls[sampn] = entropy->derived_tbls[compptr->dc_tbl_no]; + entropy->cur_counts[sampn] = entropy->count_ptrs[compptr->dc_tbl_no]; + } + } + } + entropy->num_input_ptrs = ptrn; + + /* Initialize bit buffer to empty */ + entropy->saved.put_buffer = 0; + entropy->saved.put_bits = 0; + + /* Initialize restart stuff */ + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num = 0; +} + + +/* Outputting bytes to the file */ + +/* Emit a byte, taking 'action' if must suspend. */ +#define emit_byte(state, val, action) { \ + *(state)->next_output_byte++ = (JOCTET)(val); \ + if (--(state)->free_in_buffer == 0) \ + if (!dump_buffer(state)) \ + { action; } \ +} + + +LOCAL(boolean) +dump_buffer(working_state *state) +/* Empty the output buffer; return TRUE if successful, FALSE if must suspend */ +{ + struct jpeg_destination_mgr *dest = state->cinfo->dest; + + if (!(*dest->empty_output_buffer) (state->cinfo)) + return FALSE; + /* After a successful buffer dump, must reset buffer pointers */ + state->next_output_byte = dest->next_output_byte; + state->free_in_buffer = dest->free_in_buffer; + return TRUE; +} + + +/* Outputting bits to the file */ + +/* Only the right 24 bits of put_buffer are used; the valid bits are + * left-justified in this part. At most 16 bits can be passed to emit_bits + * in one call, and we never retain more than 7 bits in put_buffer + * between calls, so 24 bits are sufficient. + */ + +INLINE +LOCAL(boolean) +emit_bits(working_state *state, unsigned int code, int size) +/* Emit some bits; return TRUE if successful, FALSE if must suspend */ +{ + /* This routine is heavily used, so it's worth coding tightly. */ + register size_t put_buffer = (size_t)code; + register int put_bits = state->cur.put_bits; + + /* if size is 0, caller used an invalid Huffman table entry */ + if (size == 0) + ERREXIT(state->cinfo, JERR_HUFF_MISSING_CODE); + + put_buffer &= (((size_t)1) << size) - 1; /* mask off any extra bits in code */ + + put_bits += size; /* new number of bits in buffer */ + + put_buffer <<= 24 - put_bits; /* align incoming bits */ + + put_buffer |= state->cur.put_buffer; /* and merge with old buffer contents */ + + while (put_bits >= 8) { + int c = (int)((put_buffer >> 16) & 0xFF); + + emit_byte(state, c, return FALSE); + if (c == 0xFF) { /* need to stuff a zero byte? */ + emit_byte(state, 0, return FALSE); + } + put_buffer <<= 8; + put_bits -= 8; + } + + state->cur.put_buffer = put_buffer; /* update state variables */ + state->cur.put_bits = put_bits; + + return TRUE; +} + + +LOCAL(boolean) +flush_bits(working_state *state) +{ + if (!emit_bits(state, 0x7F, 7)) /* fill any partial byte with ones */ + return FALSE; + state->cur.put_buffer = 0; /* and reset bit-buffer to empty */ + state->cur.put_bits = 0; + return TRUE; +} + + +/* + * Emit a restart marker & resynchronize predictions. + */ + +LOCAL(boolean) +emit_restart(working_state *state, int restart_num) +{ + if (!flush_bits(state)) + return FALSE; + + emit_byte(state, 0xFF, return FALSE); + emit_byte(state, JPEG_RST0 + restart_num, return FALSE); + + /* The restart counter is not updated until we successfully write the MCU. */ + + return TRUE; +} + + +/* + * Encode and output nMCU MCUs' worth of Huffman-compressed differences. + */ + +METHODDEF(JDIMENSION) +encode_mcus_huff(j_compress_ptr cinfo, JDIFFIMAGE diff_buf, + JDIMENSION MCU_row_num, JDIMENSION MCU_col_num, + JDIMENSION nMCU) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + working_state state; + int sampn, ci, yoffset, MCU_width, ptrn; + JDIMENSION mcu_num; + + /* Load up working state */ + state.next_output_byte = cinfo->dest->next_output_byte; + state.free_in_buffer = cinfo->dest->free_in_buffer; + state.cur = entropy->saved; + state.cinfo = cinfo; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + if (!emit_restart(&state, entropy->next_restart_num)) + return 0; + } + + /* Set input pointer locations based on MCU_col_num */ + for (ptrn = 0; ptrn < entropy->num_input_ptrs; ptrn++) { + ci = entropy->input_ptr_info[ptrn].ci; + yoffset = entropy->input_ptr_info[ptrn].yoffset; + MCU_width = entropy->input_ptr_info[ptrn].MCU_width; + entropy->input_ptr[ptrn] = + diff_buf[ci][MCU_row_num + yoffset] + (MCU_col_num * MCU_width); + } + + for (mcu_num = 0; mcu_num < nMCU; mcu_num++) { + + /* Inner loop handles the samples in the MCU */ + for (sampn = 0; sampn < cinfo->blocks_in_MCU; sampn++) { + register int temp, temp2; + register int nbits; + c_derived_tbl *dctbl = entropy->cur_tbls[sampn]; + + /* Encode the difference per section H.1.2.2 */ + + /* Input the sample difference */ + temp = *entropy->input_ptr[entropy->input_ptr_index[sampn]]++; + + if (temp & 0x8000) { /* instead of temp < 0 */ + temp = (-temp) & 0x7FFF; /* absolute value, mod 2^16 */ + if (temp == 0) /* special case: magnitude = 32768 */ + temp2 = temp = 0x8000; + temp2 = ~temp; /* one's complement of magnitude */ + } else { + temp &= 0x7FFF; /* abs value mod 2^16 */ + temp2 = temp; /* magnitude */ + } + + /* Find the number of bits needed for the magnitude of the difference */ + nbits = 0; + while (temp) { + nbits++; + temp >>= 1; + } + /* Check for out-of-range difference values. + */ + if (nbits > MAX_DIFF_BITS) + ERREXIT(cinfo, JERR_BAD_DCT_COEF); + + /* Emit the Huffman-coded symbol for the number of bits */ + if (!emit_bits(&state, dctbl->ehufco[nbits], dctbl->ehufsi[nbits])) + return mcu_num; + + /* Emit that number of bits of the value, if positive, */ + /* or the complement of its magnitude, if negative. */ + if (nbits && /* emit_bits rejects calls with size 0 */ + nbits != 16) /* special case: no bits should be emitted */ + if (!emit_bits(&state, (unsigned int)temp2, nbits)) + return mcu_num; + } + + /* Completed MCU, so update state */ + cinfo->dest->next_output_byte = state.next_output_byte; + cinfo->dest->free_in_buffer = state.free_in_buffer; + entropy->saved = state.cur; + + /* Update restart-interval state too */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + } + + return nMCU; +} + + +/* + * Finish up at the end of a Huffman-compressed scan. + */ + +METHODDEF(void) +finish_pass_huff(j_compress_ptr cinfo) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + working_state state; + + /* Load up working state ... flush_bits needs it */ + state.next_output_byte = cinfo->dest->next_output_byte; + state.free_in_buffer = cinfo->dest->free_in_buffer; + state.cur = entropy->saved; + state.cinfo = cinfo; + + /* Flush out the last data */ + if (!flush_bits(&state)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); + + /* Update state */ + cinfo->dest->next_output_byte = state.next_output_byte; + cinfo->dest->free_in_buffer = state.free_in_buffer; + entropy->saved = state.cur; +} + + +/* + * Huffman coding optimization. + * + * We first scan the supplied data and count the number of uses of each symbol + * that is to be Huffman-coded. (This process MUST agree with the code above.) + * Then we build a Huffman coding tree for the observed counts. + * Symbols which are not needed at all for the particular image are not + * assigned any code, which saves space in the DHT marker as well as in + * the compressed data. + */ + +#ifdef ENTROPY_OPT_SUPPORTED + +/* + * Trial-encode nMCU MCUs' worth of Huffman-compressed differences. + * No data is actually output, so no suspension return is possible. + */ + +METHODDEF(JDIMENSION) +encode_mcus_gather(j_compress_ptr cinfo, JDIFFIMAGE diff_buf, + JDIMENSION MCU_row_num, JDIMENSION MCU_col_num, + JDIMENSION nMCU) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + int sampn, ci, yoffset, MCU_width, ptrn; + JDIMENSION mcu_num; + + /* Take care of restart intervals if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + /* Update restart state */ + entropy->restarts_to_go = cinfo->restart_interval; + } + entropy->restarts_to_go--; + } + + /* Set input pointer locations based on MCU_col_num */ + for (ptrn = 0; ptrn < entropy->num_input_ptrs; ptrn++) { + ci = entropy->input_ptr_info[ptrn].ci; + yoffset = entropy->input_ptr_info[ptrn].yoffset; + MCU_width = entropy->input_ptr_info[ptrn].MCU_width; + entropy->input_ptr[ptrn] = + diff_buf[ci][MCU_row_num + yoffset] + (MCU_col_num * MCU_width); + } + + for (mcu_num = 0; mcu_num < nMCU; mcu_num++) { + + /* Inner loop handles the samples in the MCU */ + for (sampn = 0; sampn < cinfo->blocks_in_MCU; sampn++) { + register int temp; + register int nbits; + long *counts = entropy->cur_counts[sampn]; + + /* Encode the difference per section H.1.2.2 */ + + /* Input the sample difference */ + temp = *entropy->input_ptr[entropy->input_ptr_index[sampn]]++; + + if (temp & 0x8000) { /* instead of temp < 0 */ + temp = (-temp) & 0x7FFF; /* absolute value, mod 2^16 */ + if (temp == 0) /* special case: magnitude = 32768 */ + temp = 0x8000; + } else + temp &= 0x7FFF; /* abs value mod 2^16 */ + + /* Find the number of bits needed for the magnitude of the difference */ + nbits = 0; + while (temp) { + nbits++; + temp >>= 1; + } + /* Check for out-of-range difference values. + */ + if (nbits > MAX_DIFF_BITS) + ERREXIT(cinfo, JERR_BAD_DCT_COEF); + + /* Count the Huffman symbol for the number of bits */ + counts[nbits]++; + } + } + + return nMCU; +} + + +/* + * Finish up a statistics-gathering pass and create the new Huffman tables. + */ + +METHODDEF(void) +finish_pass_gather(j_compress_ptr cinfo) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + int ci, dctbl; + jpeg_component_info *compptr; + JHUFF_TBL **htblptr; + boolean did_dc[NUM_HUFF_TBLS]; + + /* It's important not to apply jpeg_gen_optimal_table more than once + * per table, because it clobbers the input frequency counts! + */ + memset(did_dc, 0, sizeof(did_dc)); + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + dctbl = compptr->dc_tbl_no; + if (!did_dc[dctbl]) { + htblptr = &cinfo->dc_huff_tbl_ptrs[dctbl]; + if (*htblptr == NULL) + *htblptr = jpeg_alloc_huff_table((j_common_ptr)cinfo); + jpeg_gen_optimal_table(cinfo, *htblptr, entropy->count_ptrs[dctbl]); + did_dc[dctbl] = TRUE; + } + } +} + + +#endif /* ENTROPY_OPT_SUPPORTED */ + + +/* + * Module initialization routine for Huffman entropy encoding. + */ + +GLOBAL(void) +jinit_lhuff_encoder(j_compress_ptr cinfo) +{ + lhuff_entropy_ptr entropy; + int i; + + entropy = (lhuff_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(lhuff_entropy_encoder)); + cinfo->entropy = (struct jpeg_entropy_encoder *)entropy; + entropy->pub.start_pass = start_pass_lhuff; + + /* Mark tables unallocated */ + for (i = 0; i < NUM_HUFF_TBLS; i++) { + entropy->derived_tbls[i] = NULL; +#ifdef ENTROPY_OPT_SUPPORTED + entropy->count_ptrs[i] = NULL; +#endif + } +} + +#endif /* C_LOSSLESS_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jclossls.c b/src/spice2x/external/libjpeg-turbo/jclossls.c new file mode 100644 index 0000000..e9ba92a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jclossls.c @@ -0,0 +1,319 @@ +/* + * jclossls.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1998, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains prediction, sample differencing, and point transform + * routines for the lossless JPEG compressor. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jlossls.h" + +#ifdef C_LOSSLESS_SUPPORTED + + +/************************** Sample differencing **************************/ + +/* + * In order to avoid a performance penalty for checking which predictor is + * being used and which row is being processed for each call of the + * undifferencer, and to promote optimization, we have separate differencing + * functions for each predictor selection value. + * + * We are able to avoid duplicating source code by implementing the predictors + * and differencers as macros. Each of the differencing functions is simply a + * wrapper around a DIFFERENCE macro with the appropriate PREDICTOR macro + * passed as an argument. + */ + +/* Forward declarations */ +LOCAL(void) reset_predictor(j_compress_ptr cinfo, int ci); + + +/* Predictor for the first column of the first row: 2^(P-Pt-1) */ +#define INITIAL_PREDICTORx (1 << (cinfo->data_precision - cinfo->Al - 1)) + +/* Predictor for the first column of the remaining rows: Rb */ +#define INITIAL_PREDICTOR2 prev_row[0] + + +/* + * 1-Dimensional differencer routine. + * + * This macro implements the 1-D horizontal predictor (1). INITIAL_PREDICTOR + * is used as the special case predictor for the first column, which must be + * either INITIAL_PREDICTOR2 or INITIAL_PREDICTORx. The remaining samples + * use PREDICTOR1. + */ + +#define DIFFERENCE_1D(INITIAL_PREDICTOR) \ + lossless_comp_ptr losslessc = (lossless_comp_ptr)cinfo->fdct; \ + boolean restart = FALSE; \ + int samp, Ra; \ + \ + samp = *input_buf++; \ + *diff_buf++ = samp - INITIAL_PREDICTOR; \ + \ + while (--width) { \ + Ra = samp; \ + samp = *input_buf++; \ + *diff_buf++ = samp - PREDICTOR1; \ + } \ + \ + /* Account for restart interval (no-op if not using restarts) */ \ + if (cinfo->restart_interval) { \ + if (--(losslessc->restart_rows_to_go[ci]) == 0) { \ + reset_predictor(cinfo, ci); \ + restart = TRUE; \ + } \ + } + + +/* + * 2-Dimensional differencer routine. + * + * This macro implements the 2-D horizontal predictors (#2-7). PREDICTOR2 is + * used as the special case predictor for the first column. The remaining + * samples use PREDICTOR, which is a function of Ra, Rb, and Rc. + * + * Because prev_row and output_buf may point to the same storage area (in an + * interleaved image with Vi=1, for example), we must take care to buffer Rb/Rc + * before writing the current reconstructed sample value into output_buf. + */ + +#define DIFFERENCE_2D(PREDICTOR) \ + lossless_comp_ptr losslessc = (lossless_comp_ptr)cinfo->fdct; \ + int samp, Ra, Rb, Rc; \ + \ + Rb = *prev_row++; \ + samp = *input_buf++; \ + *diff_buf++ = samp - PREDICTOR2; \ + \ + while (--width) { \ + Rc = Rb; \ + Rb = *prev_row++; \ + Ra = samp; \ + samp = *input_buf++; \ + *diff_buf++ = samp - PREDICTOR; \ + } \ + \ + /* Account for restart interval (no-op if not using restarts) */ \ + if (cinfo->restart_interval) { \ + if (--losslessc->restart_rows_to_go[ci] == 0) \ + reset_predictor(cinfo, ci); \ + } + + +/* + * Differencers for the second and subsequent rows in a scan or restart + * interval. The first sample in the row is differenced using the vertical + * predictor (2). The rest of the samples are differenced using the predictor + * specified in the scan header. + */ + +METHODDEF(void) +jpeg_difference1(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_1D(INITIAL_PREDICTOR2); + (void)(restart); +} + +METHODDEF(void) +jpeg_difference2(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_2D(PREDICTOR2); + (void)(Ra); + (void)(Rc); +} + +METHODDEF(void) +jpeg_difference3(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_2D(PREDICTOR3); + (void)(Ra); +} + +METHODDEF(void) +jpeg_difference4(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_2D(PREDICTOR4); +} + +METHODDEF(void) +jpeg_difference5(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_2D(PREDICTOR5); +} + +METHODDEF(void) +jpeg_difference6(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_2D(PREDICTOR6); +} + +METHODDEF(void) +jpeg_difference7(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_2D(PREDICTOR7); + (void)(Rc); +} + + +/* + * Differencer for the first row in a scan or restart interval. The first + * sample in the row is differenced using the special predictor constant + * x = 2 ^ (P-Pt-1). The rest of the samples are differenced using the + * 1-D horizontal predictor (1). + */ + +METHODDEF(void) +jpeg_difference_first_row(j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, _JSAMPROW prev_row, + JDIFFROW diff_buf, JDIMENSION width) +{ + DIFFERENCE_1D(INITIAL_PREDICTORx); + + /* + * Now that we have differenced the first row, we want to use the + * differencer that corresponds to the predictor specified in the + * scan header. + * + * Note that we don't do this if we have just reset the predictor + * for a new restart interval. + */ + if (!restart) { + switch (cinfo->Ss) { + case 1: + losslessc->predict_difference[ci] = jpeg_difference1; + break; + case 2: + losslessc->predict_difference[ci] = jpeg_difference2; + break; + case 3: + losslessc->predict_difference[ci] = jpeg_difference3; + break; + case 4: + losslessc->predict_difference[ci] = jpeg_difference4; + break; + case 5: + losslessc->predict_difference[ci] = jpeg_difference5; + break; + case 6: + losslessc->predict_difference[ci] = jpeg_difference6; + break; + case 7: + losslessc->predict_difference[ci] = jpeg_difference7; + break; + } + } +} + +/* + * Reset predictor at the start of a pass or restart interval. + */ + +LOCAL(void) +reset_predictor(j_compress_ptr cinfo, int ci) +{ + lossless_comp_ptr losslessc = (lossless_comp_ptr)cinfo->fdct; + + /* Initialize restart counter */ + losslessc->restart_rows_to_go[ci] = + cinfo->restart_interval / cinfo->MCUs_per_row; + + /* Set difference function to first row function */ + losslessc->predict_difference[ci] = jpeg_difference_first_row; +} + + +/********************** Sample downscaling by 2^Pt ***********************/ + +METHODDEF(void) +simple_downscale(j_compress_ptr cinfo, + _JSAMPROW input_buf, _JSAMPROW output_buf, JDIMENSION width) +{ + do { + *output_buf++ = (_JSAMPLE)RIGHT_SHIFT(*input_buf++, cinfo->Al); + } while (--width); +} + + +METHODDEF(void) +noscale(j_compress_ptr cinfo, + _JSAMPROW input_buf, _JSAMPROW output_buf, JDIMENSION width) +{ + memcpy(output_buf, input_buf, width * sizeof(_JSAMPLE)); +} + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_lossless(j_compress_ptr cinfo) +{ + lossless_comp_ptr losslessc = (lossless_comp_ptr)cinfo->fdct; + int ci; + + /* Set scaler function based on Pt */ + if (cinfo->Al) + losslessc->scaler_scale = simple_downscale; + else + losslessc->scaler_scale = noscale; + + /* Check that the restart interval is an integer multiple of the number + * of MCUs in an MCU row. + */ + if (cinfo->restart_interval % cinfo->MCUs_per_row != 0) + ERREXIT2(cinfo, JERR_BAD_RESTART, + cinfo->restart_interval, cinfo->MCUs_per_row); + + /* Set predictors for start of pass */ + for (ci = 0; ci < cinfo->num_components; ci++) + reset_predictor(cinfo, ci); +} + + +/* + * Initialize the lossless compressor. + */ + +GLOBAL(void) +_jinit_lossless_compressor(j_compress_ptr cinfo) +{ + lossless_comp_ptr losslessc; + + /* Create subobject in permanent pool */ + losslessc = (lossless_comp_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(jpeg_lossless_compressor)); + cinfo->fdct = (struct jpeg_forward_dct *)losslessc; + losslessc->pub.start_pass = start_pass_lossless; +} + +#endif /* C_LOSSLESS_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jcmainct.c b/src/spice2x/external/libjpeg-turbo/jcmainct.c new file mode 100644 index 0000000..fe8fc0b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcmainct.c @@ -0,0 +1,173 @@ +/* + * jcmainct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the main buffer controller for compression. + * The main buffer lies between the pre-processor and the JPEG + * compressor proper; it holds downsampled data in the JPEG colorspace. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_c_main_controller pub; /* public fields */ + + JDIMENSION cur_iMCU_row; /* number of current iMCU row */ + JDIMENSION rowgroup_ctr; /* counts row groups received in iMCU row */ + boolean suspended; /* remember if we suspended output */ + J_BUF_MODE pass_mode; /* current operating mode */ + + /* If using just a strip buffer, this points to the entire set of buffers + * (we allocate one for each component). In the full-image case, this + * points to the currently accessible strips of the virtual arrays. + */ + _JSAMPARRAY buffer[MAX_COMPONENTS]; +} my_main_controller; + +typedef my_main_controller *my_main_ptr; + + +/* Forward declarations */ +METHODDEF(void) process_data_simple_main(j_compress_ptr cinfo, + _JSAMPARRAY input_buf, + JDIMENSION *in_row_ctr, + JDIMENSION in_rows_avail); + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_main(j_compress_ptr cinfo, J_BUF_MODE pass_mode) +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + + /* Do nothing in raw-data mode. */ + if (cinfo->raw_data_in) + return; + + if (pass_mode != JBUF_PASS_THRU) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + + main_ptr->cur_iMCU_row = 0; /* initialize counters */ + main_ptr->rowgroup_ctr = 0; + main_ptr->suspended = FALSE; + main_ptr->pass_mode = pass_mode; /* save mode for use by process_data */ + main_ptr->pub._process_data = process_data_simple_main; +} + + +/* + * Process some data. + * This routine handles the simple pass-through mode, + * where we have only a strip buffer. + */ + +METHODDEF(void) +process_data_simple_main(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail) +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + JDIMENSION data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + while (main_ptr->cur_iMCU_row < cinfo->total_iMCU_rows) { + /* Read input data if we haven't filled the main buffer yet */ + if (main_ptr->rowgroup_ctr < data_unit) + (*cinfo->prep->_pre_process_data) (cinfo, input_buf, in_row_ctr, + in_rows_avail, main_ptr->buffer, + &main_ptr->rowgroup_ctr, data_unit); + + /* If we don't have a full iMCU row buffered, return to application for + * more data. Note that preprocessor will always pad to fill the iMCU row + * at the bottom of the image. + */ + if (main_ptr->rowgroup_ctr != data_unit) + return; + + /* Send the completed row to the compressor */ + if (!(*cinfo->coef->_compress_data) (cinfo, main_ptr->buffer)) { + /* If compressor did not consume the whole row, then we must need to + * suspend processing and return to the application. In this situation + * we pretend we didn't yet consume the last input row; otherwise, if + * it happened to be the last row of the image, the application would + * think we were done. + */ + if (!main_ptr->suspended) { + (*in_row_ctr)--; + main_ptr->suspended = TRUE; + } + return; + } + /* We did finish the row. Undo our little suspension hack if a previous + * call suspended; then mark the main buffer empty. + */ + if (main_ptr->suspended) { + (*in_row_ctr)++; + main_ptr->suspended = FALSE; + } + main_ptr->rowgroup_ctr = 0; + main_ptr->cur_iMCU_row++; + } +} + + +/* + * Initialize main buffer controller. + */ + +GLOBAL(void) +_jinit_c_main_controller(j_compress_ptr cinfo, boolean need_full_buffer) +{ + my_main_ptr main_ptr; + int ci; + jpeg_component_info *compptr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + main_ptr = (my_main_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_main_controller)); + cinfo->main = (struct jpeg_c_main_controller *)main_ptr; + main_ptr->pub.start_pass = start_pass_main; + + /* We don't need to create a buffer in raw-data mode. */ + if (cinfo->raw_data_in) + return; + + /* Create the buffer. It holds downsampled data, so each component + * may be of a different size. + */ + if (need_full_buffer) { + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + } else { + /* Allocate a strip buffer for each component */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + main_ptr->buffer[ci] = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + compptr->width_in_blocks * data_unit, + (JDIMENSION)(compptr->v_samp_factor * data_unit)); + } + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jcmarker.c b/src/spice2x/external/libjpeg-turbo/jcmarker.c new file mode 100644 index 0000000..a064d4d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcmarker.c @@ -0,0 +1,670 @@ +/* + * jcmarker.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1998, Thomas G. Lane. + * Modified 2003-2010 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2010, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains routines to write JPEG datastream markers. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jpegapicomp.h" + + +typedef enum { /* JPEG marker codes */ + M_SOF0 = 0xc0, + M_SOF1 = 0xc1, + M_SOF2 = 0xc2, + M_SOF3 = 0xc3, + + M_SOF5 = 0xc5, + M_SOF6 = 0xc6, + M_SOF7 = 0xc7, + + M_JPG = 0xc8, + M_SOF9 = 0xc9, + M_SOF10 = 0xca, + M_SOF11 = 0xcb, + + M_SOF13 = 0xcd, + M_SOF14 = 0xce, + M_SOF15 = 0xcf, + + M_DHT = 0xc4, + + M_DAC = 0xcc, + + M_RST0 = 0xd0, + M_RST1 = 0xd1, + M_RST2 = 0xd2, + M_RST3 = 0xd3, + M_RST4 = 0xd4, + M_RST5 = 0xd5, + M_RST6 = 0xd6, + M_RST7 = 0xd7, + + M_SOI = 0xd8, + M_EOI = 0xd9, + M_SOS = 0xda, + M_DQT = 0xdb, + M_DNL = 0xdc, + M_DRI = 0xdd, + M_DHP = 0xde, + M_EXP = 0xdf, + + M_APP0 = 0xe0, + M_APP1 = 0xe1, + M_APP2 = 0xe2, + M_APP3 = 0xe3, + M_APP4 = 0xe4, + M_APP5 = 0xe5, + M_APP6 = 0xe6, + M_APP7 = 0xe7, + M_APP8 = 0xe8, + M_APP9 = 0xe9, + M_APP10 = 0xea, + M_APP11 = 0xeb, + M_APP12 = 0xec, + M_APP13 = 0xed, + M_APP14 = 0xee, + M_APP15 = 0xef, + + M_JPG0 = 0xf0, + M_JPG13 = 0xfd, + M_COM = 0xfe, + + M_TEM = 0x01, + + M_ERROR = 0x100 +} JPEG_MARKER; + + +/* Private state */ + +typedef struct { + struct jpeg_marker_writer pub; /* public fields */ + + unsigned int last_restart_interval; /* last DRI value emitted; 0 after SOI */ +} my_marker_writer; + +typedef my_marker_writer *my_marker_ptr; + + +/* + * Basic output routines. + * + * Note that we do not support suspension while writing a marker. + * Therefore, an application using suspension must ensure that there is + * enough buffer space for the initial markers (typ. 600-700 bytes) before + * calling jpeg_start_compress, and enough space to write the trailing EOI + * (a few bytes) before calling jpeg_finish_compress. Multipass compression + * modes are not supported at all with suspension, so those two are the only + * points where markers will be written. + */ + +LOCAL(void) +emit_byte(j_compress_ptr cinfo, int val) +/* Emit a byte */ +{ + struct jpeg_destination_mgr *dest = cinfo->dest; + + *(dest->next_output_byte)++ = (JOCTET)val; + if (--dest->free_in_buffer == 0) { + if (!(*dest->empty_output_buffer) (cinfo)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); + } +} + + +LOCAL(void) +emit_marker(j_compress_ptr cinfo, JPEG_MARKER mark) +/* Emit a marker code */ +{ + emit_byte(cinfo, 0xFF); + emit_byte(cinfo, (int)mark); +} + + +LOCAL(void) +emit_2bytes(j_compress_ptr cinfo, int value) +/* Emit a 2-byte integer; these are always MSB first in JPEG files */ +{ + emit_byte(cinfo, (value >> 8) & 0xFF); + emit_byte(cinfo, value & 0xFF); +} + + +/* + * Routines to write specific marker types. + */ + +LOCAL(int) +emit_dqt(j_compress_ptr cinfo, int index) +/* Emit a DQT marker */ +/* Returns the precision used (0 = 8bits, 1 = 16bits) for baseline checking */ +{ + JQUANT_TBL *qtbl = cinfo->quant_tbl_ptrs[index]; + int prec; + int i; + + if (qtbl == NULL) + ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, index); + + prec = 0; + for (i = 0; i < DCTSIZE2; i++) { + if (qtbl->quantval[i] > 255) + prec = 1; + } + + if (!qtbl->sent_table) { + emit_marker(cinfo, M_DQT); + + emit_2bytes(cinfo, prec ? DCTSIZE2 * 2 + 1 + 2 : DCTSIZE2 + 1 + 2); + + emit_byte(cinfo, index + (prec << 4)); + + for (i = 0; i < DCTSIZE2; i++) { + /* The table entries must be emitted in zigzag order. */ + unsigned int qval = qtbl->quantval[jpeg_natural_order[i]]; + if (prec) + emit_byte(cinfo, (int)(qval >> 8)); + emit_byte(cinfo, (int)(qval & 0xFF)); + } + + qtbl->sent_table = TRUE; + } + + return prec; +} + + +LOCAL(void) +emit_dht(j_compress_ptr cinfo, int index, boolean is_ac) +/* Emit a DHT marker */ +{ + JHUFF_TBL *htbl; + int length, i; + + if (is_ac) { + htbl = cinfo->ac_huff_tbl_ptrs[index]; + index += 0x10; /* output index has AC bit set */ + } else { + htbl = cinfo->dc_huff_tbl_ptrs[index]; + } + + if (htbl == NULL) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, index); + + if (!htbl->sent_table) { + emit_marker(cinfo, M_DHT); + + length = 0; + for (i = 1; i <= 16; i++) + length += htbl->bits[i]; + + emit_2bytes(cinfo, length + 2 + 1 + 16); + emit_byte(cinfo, index); + + for (i = 1; i <= 16; i++) + emit_byte(cinfo, htbl->bits[i]); + + for (i = 0; i < length; i++) + emit_byte(cinfo, htbl->huffval[i]); + + htbl->sent_table = TRUE; + } +} + + +LOCAL(void) +emit_dac(j_compress_ptr cinfo) +/* Emit a DAC marker */ +/* Since the useful info is so small, we want to emit all the tables in */ +/* one DAC marker. Therefore this routine does its own scan of the table. */ +{ +#ifdef C_ARITH_CODING_SUPPORTED + char dc_in_use[NUM_ARITH_TBLS]; + char ac_in_use[NUM_ARITH_TBLS]; + int length, i; + jpeg_component_info *compptr; + + for (i = 0; i < NUM_ARITH_TBLS; i++) + dc_in_use[i] = ac_in_use[i] = 0; + + for (i = 0; i < cinfo->comps_in_scan; i++) { + compptr = cinfo->cur_comp_info[i]; + /* DC needs no table for refinement scan */ + if (cinfo->Ss == 0 && cinfo->Ah == 0) + dc_in_use[compptr->dc_tbl_no] = 1; + /* AC needs no table when not present */ + if (cinfo->Se) + ac_in_use[compptr->ac_tbl_no] = 1; + } + + length = 0; + for (i = 0; i < NUM_ARITH_TBLS; i++) + length += dc_in_use[i] + ac_in_use[i]; + + if (length) { + emit_marker(cinfo, M_DAC); + + emit_2bytes(cinfo, length * 2 + 2); + + for (i = 0; i < NUM_ARITH_TBLS; i++) { + if (dc_in_use[i]) { + emit_byte(cinfo, i); + emit_byte(cinfo, cinfo->arith_dc_L[i] + (cinfo->arith_dc_U[i] << 4)); + } + if (ac_in_use[i]) { + emit_byte(cinfo, i + 0x10); + emit_byte(cinfo, cinfo->arith_ac_K[i]); + } + } + } +#endif /* C_ARITH_CODING_SUPPORTED */ +} + + +LOCAL(void) +emit_dri(j_compress_ptr cinfo) +/* Emit a DRI marker */ +{ + emit_marker(cinfo, M_DRI); + + emit_2bytes(cinfo, 4); /* fixed length */ + + emit_2bytes(cinfo, (int)cinfo->restart_interval); +} + + +LOCAL(void) +emit_sof(j_compress_ptr cinfo, JPEG_MARKER code) +/* Emit a SOF marker */ +{ + int ci; + jpeg_component_info *compptr; + + emit_marker(cinfo, code); + + emit_2bytes(cinfo, 3 * cinfo->num_components + 2 + 5 + 1); /* length */ + + /* Make sure image isn't bigger than SOF field can handle */ + if ((long)cinfo->_jpeg_height > 65535L || (long)cinfo->_jpeg_width > 65535L) + ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int)65535); + + emit_byte(cinfo, cinfo->data_precision); + emit_2bytes(cinfo, (int)cinfo->_jpeg_height); + emit_2bytes(cinfo, (int)cinfo->_jpeg_width); + + emit_byte(cinfo, cinfo->num_components); + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + emit_byte(cinfo, compptr->component_id); + emit_byte(cinfo, (compptr->h_samp_factor << 4) + compptr->v_samp_factor); + emit_byte(cinfo, compptr->quant_tbl_no); + } +} + + +LOCAL(void) +emit_sos(j_compress_ptr cinfo) +/* Emit a SOS marker */ +{ + int i, td, ta; + jpeg_component_info *compptr; + + emit_marker(cinfo, M_SOS); + + emit_2bytes(cinfo, 2 * cinfo->comps_in_scan + 2 + 1 + 3); /* length */ + + emit_byte(cinfo, cinfo->comps_in_scan); + + for (i = 0; i < cinfo->comps_in_scan; i++) { + compptr = cinfo->cur_comp_info[i]; + emit_byte(cinfo, compptr->component_id); + + /* We emit 0 for unused field(s); this is recommended by the P&M text + * but does not seem to be specified in the standard. + */ + + /* DC needs no table for refinement scan */ + td = cinfo->Ss == 0 && cinfo->Ah == 0 ? compptr->dc_tbl_no : 0; + /* AC needs no table when not present */ + ta = cinfo->Se ? compptr->ac_tbl_no : 0; + + emit_byte(cinfo, (td << 4) + ta); + } + + emit_byte(cinfo, cinfo->Ss); + emit_byte(cinfo, cinfo->Se); + emit_byte(cinfo, (cinfo->Ah << 4) + cinfo->Al); +} + + +LOCAL(void) +emit_jfif_app0(j_compress_ptr cinfo) +/* Emit a JFIF-compliant APP0 marker */ +{ + /* + * Length of APP0 block (2 bytes) + * Block ID (4 bytes - ASCII "JFIF") + * Zero byte (1 byte to terminate the ID string) + * Version Major, Minor (2 bytes - major first) + * Units (1 byte - 0x00 = none, 0x01 = inch, 0x02 = cm) + * Xdpu (2 bytes - dots per unit horizontal) + * Ydpu (2 bytes - dots per unit vertical) + * Thumbnail X size (1 byte) + * Thumbnail Y size (1 byte) + */ + + emit_marker(cinfo, M_APP0); + + emit_2bytes(cinfo, 2 + 4 + 1 + 2 + 1 + 2 + 2 + 1 + 1); /* length */ + + emit_byte(cinfo, 0x4A); /* Identifier: ASCII "JFIF" */ + emit_byte(cinfo, 0x46); + emit_byte(cinfo, 0x49); + emit_byte(cinfo, 0x46); + emit_byte(cinfo, 0); + emit_byte(cinfo, cinfo->JFIF_major_version); /* Version fields */ + emit_byte(cinfo, cinfo->JFIF_minor_version); + emit_byte(cinfo, cinfo->density_unit); /* Pixel size information */ + emit_2bytes(cinfo, (int)cinfo->X_density); + emit_2bytes(cinfo, (int)cinfo->Y_density); + emit_byte(cinfo, 0); /* No thumbnail image */ + emit_byte(cinfo, 0); +} + + +LOCAL(void) +emit_adobe_app14(j_compress_ptr cinfo) +/* Emit an Adobe APP14 marker */ +{ + /* + * Length of APP14 block (2 bytes) + * Block ID (5 bytes - ASCII "Adobe") + * Version Number (2 bytes - currently 100) + * Flags0 (2 bytes - currently 0) + * Flags1 (2 bytes - currently 0) + * Color transform (1 byte) + * + * Although Adobe TN 5116 mentions Version = 101, all the Adobe files + * now in circulation seem to use Version = 100, so that's what we write. + * + * We write the color transform byte as 1 if the JPEG color space is + * YCbCr, 2 if it's YCCK, 0 otherwise. Adobe's definition has to do with + * whether the encoder performed a transformation, which is pretty useless. + */ + + emit_marker(cinfo, M_APP14); + + emit_2bytes(cinfo, 2 + 5 + 2 + 2 + 2 + 1); /* length */ + + emit_byte(cinfo, 0x41); /* Identifier: ASCII "Adobe" */ + emit_byte(cinfo, 0x64); + emit_byte(cinfo, 0x6F); + emit_byte(cinfo, 0x62); + emit_byte(cinfo, 0x65); + emit_2bytes(cinfo, 100); /* Version */ + emit_2bytes(cinfo, 0); /* Flags0 */ + emit_2bytes(cinfo, 0); /* Flags1 */ + switch (cinfo->jpeg_color_space) { + case JCS_YCbCr: + emit_byte(cinfo, 1); /* Color transform = 1 */ + break; + case JCS_YCCK: + emit_byte(cinfo, 2); /* Color transform = 2 */ + break; + default: + emit_byte(cinfo, 0); /* Color transform = 0 */ + break; + } +} + + +/* + * These routines allow writing an arbitrary marker with parameters. + * The only intended use is to emit COM or APPn markers after calling + * write_file_header and before calling write_frame_header. + * Other uses are not guaranteed to produce desirable results. + * Counting the parameter bytes properly is the caller's responsibility. + */ + +METHODDEF(void) +write_marker_header(j_compress_ptr cinfo, int marker, unsigned int datalen) +/* Emit an arbitrary marker header */ +{ + if (datalen > (unsigned int)65533) /* safety check */ + ERREXIT(cinfo, JERR_BAD_LENGTH); + + emit_marker(cinfo, (JPEG_MARKER)marker); + + emit_2bytes(cinfo, (int)(datalen + 2)); /* total length */ +} + +METHODDEF(void) +write_marker_byte(j_compress_ptr cinfo, int val) +/* Emit one byte of marker parameters following write_marker_header */ +{ + emit_byte(cinfo, val); +} + + +/* + * Write datastream header. + * This consists of an SOI and optional APPn markers. + * We recommend use of the JFIF marker, but not the Adobe marker, + * when using YCbCr or grayscale data. The JFIF marker should NOT + * be used for any other JPEG colorspace. The Adobe marker is helpful + * to distinguish RGB, CMYK, and YCCK colorspaces. + * Note that an application can write additional header markers after + * jpeg_start_compress returns. + */ + +METHODDEF(void) +write_file_header(j_compress_ptr cinfo) +{ + my_marker_ptr marker = (my_marker_ptr)cinfo->marker; + + emit_marker(cinfo, M_SOI); /* first the SOI */ + + /* SOI is defined to reset restart interval to 0 */ + marker->last_restart_interval = 0; + + if (cinfo->write_JFIF_header) /* next an optional JFIF APP0 */ + emit_jfif_app0(cinfo); + if (cinfo->write_Adobe_marker) /* next an optional Adobe APP14 */ + emit_adobe_app14(cinfo); +} + + +/* + * Write frame header. + * This consists of DQT and SOFn markers. + * Note that we do not emit the SOF until we have emitted the DQT(s). + * This avoids compatibility problems with incorrect implementations that + * try to error-check the quant table numbers as soon as they see the SOF. + */ + +METHODDEF(void) +write_frame_header(j_compress_ptr cinfo) +{ + int ci, prec = 0; + boolean is_baseline; + jpeg_component_info *compptr; + + if (!cinfo->master->lossless) { + /* Emit DQT for each quantization table. + * Note that emit_dqt() suppresses any duplicate tables. + */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + prec += emit_dqt(cinfo, compptr->quant_tbl_no); + } + /* now prec is nonzero iff there are any 16-bit quant tables. */ + } + + /* Check for a non-baseline specification. + * Note we assume that Huffman table numbers won't be changed later. + */ + if (cinfo->arith_code || cinfo->progressive_mode || + cinfo->master->lossless || cinfo->data_precision != 8) { + is_baseline = FALSE; + } else { + is_baseline = TRUE; + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + if (compptr->dc_tbl_no > 1 || compptr->ac_tbl_no > 1) + is_baseline = FALSE; + } + if (prec && is_baseline) { + is_baseline = FALSE; + /* If it's baseline except for quantizer size, warn the user */ + TRACEMS(cinfo, 0, JTRC_16BIT_TABLES); + } + } + + /* Emit the proper SOF marker */ + if (cinfo->arith_code) { + if (cinfo->progressive_mode) + emit_sof(cinfo, M_SOF10); /* SOF code for progressive arithmetic */ + else + emit_sof(cinfo, M_SOF9); /* SOF code for sequential arithmetic */ + } else { + if (cinfo->progressive_mode) + emit_sof(cinfo, M_SOF2); /* SOF code for progressive Huffman */ + else if (cinfo->master->lossless) + emit_sof(cinfo, M_SOF3); /* SOF code for lossless Huffman */ + else if (is_baseline) + emit_sof(cinfo, M_SOF0); /* SOF code for baseline implementation */ + else + emit_sof(cinfo, M_SOF1); /* SOF code for non-baseline Huffman file */ + } +} + + +/* + * Write scan header. + * This consists of DHT or DAC markers, optional DRI, and SOS. + * Compressed data will be written following the SOS. + */ + +METHODDEF(void) +write_scan_header(j_compress_ptr cinfo) +{ + my_marker_ptr marker = (my_marker_ptr)cinfo->marker; + int i; + jpeg_component_info *compptr; + + if (cinfo->arith_code) { + /* Emit arith conditioning info. We may have some duplication + * if the file has multiple scans, but it's so small it's hardly + * worth worrying about. + */ + emit_dac(cinfo); + } else { + /* Emit Huffman tables. + * Note that emit_dht() suppresses any duplicate tables. + */ + for (i = 0; i < cinfo->comps_in_scan; i++) { + compptr = cinfo->cur_comp_info[i]; + /* DC needs no table for refinement scan */ + if ((cinfo->Ss == 0 && cinfo->Ah == 0) || cinfo->master->lossless) + emit_dht(cinfo, compptr->dc_tbl_no, FALSE); + /* AC needs no table when not present, and lossless mode uses only DC + tables. */ + if (cinfo->Se && !cinfo->master->lossless) + emit_dht(cinfo, compptr->ac_tbl_no, TRUE); + } + } + + /* Emit DRI if required --- note that DRI value could change for each scan. + * We avoid wasting space with unnecessary DRIs, however. + */ + if (cinfo->restart_interval != marker->last_restart_interval) { + emit_dri(cinfo); + marker->last_restart_interval = cinfo->restart_interval; + } + + emit_sos(cinfo); +} + + +/* + * Write datastream trailer. + */ + +METHODDEF(void) +write_file_trailer(j_compress_ptr cinfo) +{ + emit_marker(cinfo, M_EOI); +} + + +/* + * Write an abbreviated table-specification datastream. + * This consists of SOI, DQT and DHT tables, and EOI. + * Any table that is defined and not marked sent_table = TRUE will be + * emitted. Note that all tables will be marked sent_table = TRUE at exit. + */ + +METHODDEF(void) +write_tables_only(j_compress_ptr cinfo) +{ + int i; + + emit_marker(cinfo, M_SOI); + + for (i = 0; i < NUM_QUANT_TBLS; i++) { + if (cinfo->quant_tbl_ptrs[i] != NULL) + (void)emit_dqt(cinfo, i); + } + + if (!cinfo->arith_code) { + for (i = 0; i < NUM_HUFF_TBLS; i++) { + if (cinfo->dc_huff_tbl_ptrs[i] != NULL) + emit_dht(cinfo, i, FALSE); + if (cinfo->ac_huff_tbl_ptrs[i] != NULL) + emit_dht(cinfo, i, TRUE); + } + } + + emit_marker(cinfo, M_EOI); +} + + +/* + * Initialize the marker writer module. + */ + +GLOBAL(void) +jinit_marker_writer(j_compress_ptr cinfo) +{ + my_marker_ptr marker; + + /* Create the subobject */ + marker = (my_marker_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_marker_writer)); + cinfo->marker = (struct jpeg_marker_writer *)marker; + /* Initialize method pointers */ + marker->pub.write_file_header = write_file_header; + marker->pub.write_frame_header = write_frame_header; + marker->pub.write_scan_header = write_scan_header; + marker->pub.write_file_trailer = write_file_trailer; + marker->pub.write_tables_only = write_tables_only; + marker->pub.write_marker_header = write_marker_header; + marker->pub.write_marker_byte = write_marker_byte; + /* Initialize private state */ + marker->last_restart_interval = 0; +} diff --git a/src/spice2x/external/libjpeg-turbo/jcmaster.c b/src/spice2x/external/libjpeg-turbo/jcmaster.c new file mode 100644 index 0000000..5d89178 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcmaster.c @@ -0,0 +1,793 @@ +/* + * jcmaster.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Modified 2003-2010 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2010, 2016, 2018, 2022-2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains master control logic for the JPEG compressor. + * These routines are concerned with parameter validation, initial setup, + * and inter-pass control (determining the number of passes and the work + * to be done in each pass). + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jpegapicomp.h" +#include "jcmaster.h" + + +/* + * Support routines that do various essential calculations. + */ + +#if JPEG_LIB_VERSION >= 70 +/* + * Compute JPEG image dimensions and related values. + * NOTE: this is exported for possible use by application. + * Hence it mustn't do anything that can't be done twice. + */ + +GLOBAL(void) +jpeg_calc_jpeg_dimensions(j_compress_ptr cinfo) +/* Do computations that are needed before master selection phase */ +{ + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + /* Hardwire it to "no scaling" */ + cinfo->jpeg_width = cinfo->image_width; + cinfo->jpeg_height = cinfo->image_height; + cinfo->min_DCT_h_scaled_size = data_unit; + cinfo->min_DCT_v_scaled_size = data_unit; +} +#endif + + +LOCAL(boolean) +using_std_huff_tables(j_compress_ptr cinfo) +{ + int i; + + static const UINT8 bits_dc_luminance[17] = { + /* 0-base */ 0, 0, 1, 5, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0 + }; + static const UINT8 val_dc_luminance[] = { + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 + }; + + static const UINT8 bits_dc_chrominance[17] = { + /* 0-base */ 0, 0, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 + }; + static const UINT8 val_dc_chrominance[] = { + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 + }; + + static const UINT8 bits_ac_luminance[17] = { + /* 0-base */ 0, 0, 2, 1, 3, 3, 2, 4, 3, 5, 5, 4, 4, 0, 0, 1, 0x7d + }; + static const UINT8 val_ac_luminance[] = { + 0x01, 0x02, 0x03, 0x00, 0x04, 0x11, 0x05, 0x12, + 0x21, 0x31, 0x41, 0x06, 0x13, 0x51, 0x61, 0x07, + 0x22, 0x71, 0x14, 0x32, 0x81, 0x91, 0xa1, 0x08, + 0x23, 0x42, 0xb1, 0xc1, 0x15, 0x52, 0xd1, 0xf0, + 0x24, 0x33, 0x62, 0x72, 0x82, 0x09, 0x0a, 0x16, + 0x17, 0x18, 0x19, 0x1a, 0x25, 0x26, 0x27, 0x28, + 0x29, 0x2a, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, + 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, + 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, + 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, + 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, + 0x7a, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, + 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, + 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, + 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, + 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, 0xc4, 0xc5, + 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, 0xd3, 0xd4, + 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xe1, 0xe2, + 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, + 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, + 0xf9, 0xfa + }; + + static const UINT8 bits_ac_chrominance[17] = { + /* 0-base */ 0, 0, 2, 1, 2, 4, 4, 3, 4, 7, 5, 4, 4, 0, 1, 2, 0x77 + }; + static const UINT8 val_ac_chrominance[] = { + 0x00, 0x01, 0x02, 0x03, 0x11, 0x04, 0x05, 0x21, + 0x31, 0x06, 0x12, 0x41, 0x51, 0x07, 0x61, 0x71, + 0x13, 0x22, 0x32, 0x81, 0x08, 0x14, 0x42, 0x91, + 0xa1, 0xb1, 0xc1, 0x09, 0x23, 0x33, 0x52, 0xf0, + 0x15, 0x62, 0x72, 0xd1, 0x0a, 0x16, 0x24, 0x34, + 0xe1, 0x25, 0xf1, 0x17, 0x18, 0x19, 0x1a, 0x26, + 0x27, 0x28, 0x29, 0x2a, 0x35, 0x36, 0x37, 0x38, + 0x39, 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, + 0x49, 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, + 0x59, 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, + 0x69, 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, + 0x79, 0x7a, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, + 0x88, 0x89, 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, + 0x97, 0x98, 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, + 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, + 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, + 0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, + 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, + 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, + 0xea, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, + 0xf9, 0xfa + }; + + if (cinfo->dc_huff_tbl_ptrs[0] == NULL || + cinfo->ac_huff_tbl_ptrs[0] == NULL || + cinfo->dc_huff_tbl_ptrs[1] == NULL || + cinfo->ac_huff_tbl_ptrs[1] == NULL) + return FALSE; + + for (i = 2; i < NUM_HUFF_TBLS; i++) { + if (cinfo->dc_huff_tbl_ptrs[i] != NULL || + cinfo->ac_huff_tbl_ptrs[i] != NULL) + return FALSE; + } + + if (memcmp(cinfo->dc_huff_tbl_ptrs[0]->bits, bits_dc_luminance, + sizeof(bits_dc_luminance)) || + memcmp(cinfo->dc_huff_tbl_ptrs[0]->huffval, val_dc_luminance, + sizeof(val_dc_luminance)) || + memcmp(cinfo->ac_huff_tbl_ptrs[0]->bits, bits_ac_luminance, + sizeof(bits_ac_luminance)) || + memcmp(cinfo->ac_huff_tbl_ptrs[0]->huffval, val_ac_luminance, + sizeof(val_ac_luminance)) || + memcmp(cinfo->dc_huff_tbl_ptrs[1]->bits, bits_dc_chrominance, + sizeof(bits_dc_chrominance)) || + memcmp(cinfo->dc_huff_tbl_ptrs[1]->huffval, val_dc_chrominance, + sizeof(val_dc_chrominance)) || + memcmp(cinfo->ac_huff_tbl_ptrs[1]->bits, bits_ac_chrominance, + sizeof(bits_ac_chrominance)) || + memcmp(cinfo->ac_huff_tbl_ptrs[1]->huffval, val_ac_chrominance, + sizeof(val_ac_chrominance))) + return FALSE; + + return TRUE; +} + + +LOCAL(void) +initial_setup(j_compress_ptr cinfo, boolean transcode_only) +/* Do computations that are needed before master selection phase */ +{ + int ci; + jpeg_component_info *compptr; + long samplesperrow; + JDIMENSION jd_samplesperrow; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + +#if JPEG_LIB_VERSION >= 70 +#if JPEG_LIB_VERSION >= 80 + if (!transcode_only) +#endif + jpeg_calc_jpeg_dimensions(cinfo); +#endif + + /* Sanity check on image dimensions */ + if (cinfo->_jpeg_height <= 0 || cinfo->_jpeg_width <= 0 || + cinfo->num_components <= 0 || cinfo->input_components <= 0) + ERREXIT(cinfo, JERR_EMPTY_IMAGE); + + /* Make sure image isn't bigger than I can handle */ + if ((long)cinfo->_jpeg_height > (long)JPEG_MAX_DIMENSION || + (long)cinfo->_jpeg_width > (long)JPEG_MAX_DIMENSION) + ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int)JPEG_MAX_DIMENSION); + + /* Width of an input scanline must be representable as JDIMENSION. */ + samplesperrow = (long)cinfo->image_width * (long)cinfo->input_components; + jd_samplesperrow = (JDIMENSION)samplesperrow; + if ((long)jd_samplesperrow != samplesperrow) + ERREXIT(cinfo, JERR_WIDTH_OVERFLOW); + +#ifdef C_LOSSLESS_SUPPORTED + if (cinfo->data_precision != 8 && cinfo->data_precision != 12 && + cinfo->data_precision != 16) +#else + if (cinfo->data_precision != 8 && cinfo->data_precision != 12) +#endif + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + /* Check that number of components won't exceed internal array sizes */ + if (cinfo->num_components > MAX_COMPONENTS) + ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components, + MAX_COMPONENTS); + + /* Compute maximum sampling factors; check factor validity */ + cinfo->max_h_samp_factor = 1; + cinfo->max_v_samp_factor = 1; + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + if (compptr->h_samp_factor <= 0 || + compptr->h_samp_factor > MAX_SAMP_FACTOR || + compptr->v_samp_factor <= 0 || + compptr->v_samp_factor > MAX_SAMP_FACTOR) + ERREXIT(cinfo, JERR_BAD_SAMPLING); + cinfo->max_h_samp_factor = MAX(cinfo->max_h_samp_factor, + compptr->h_samp_factor); + cinfo->max_v_samp_factor = MAX(cinfo->max_v_samp_factor, + compptr->v_samp_factor); + } + + /* Compute dimensions of components */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Fill in the correct component_index value; don't rely on application */ + compptr->component_index = ci; + /* For compression, we never do DCT scaling. */ +#if JPEG_LIB_VERSION >= 70 + compptr->DCT_h_scaled_size = compptr->DCT_v_scaled_size = data_unit; +#else + compptr->DCT_scaled_size = data_unit; +#endif + /* Size in data units */ + compptr->width_in_blocks = (JDIMENSION) + jdiv_round_up((long)cinfo->_jpeg_width * (long)compptr->h_samp_factor, + (long)(cinfo->max_h_samp_factor * data_unit)); + compptr->height_in_blocks = (JDIMENSION) + jdiv_round_up((long)cinfo->_jpeg_height * (long)compptr->v_samp_factor, + (long)(cinfo->max_v_samp_factor * data_unit)); + /* Size in samples */ + compptr->downsampled_width = (JDIMENSION) + jdiv_round_up((long)cinfo->_jpeg_width * (long)compptr->h_samp_factor, + (long)cinfo->max_h_samp_factor); + compptr->downsampled_height = (JDIMENSION) + jdiv_round_up((long)cinfo->_jpeg_height * (long)compptr->v_samp_factor, + (long)cinfo->max_v_samp_factor); + /* Mark component needed (this flag isn't actually used for compression) */ + compptr->component_needed = TRUE; + } + + /* Compute number of fully interleaved MCU rows (number of times that + * main controller will call coefficient or difference controller). + */ + cinfo->total_iMCU_rows = (JDIMENSION) + jdiv_round_up((long)cinfo->_jpeg_height, + (long)(cinfo->max_v_samp_factor * data_unit)); +} + + +#if defined(C_MULTISCAN_FILES_SUPPORTED) || defined(C_LOSSLESS_SUPPORTED) +#define NEED_SCAN_SCRIPT +#endif + +#ifdef NEED_SCAN_SCRIPT + +LOCAL(void) +validate_script(j_compress_ptr cinfo) +/* Verify that the scan script in cinfo->scan_info[] is valid; also + * determine whether it uses progressive JPEG, and set cinfo->progressive_mode. + */ +{ + const jpeg_scan_info *scanptr; + int scanno, ncomps, ci, coefi, thisi; + int Ss, Se, Ah, Al; + boolean component_sent[MAX_COMPONENTS]; +#ifdef C_PROGRESSIVE_SUPPORTED + int *last_bitpos_ptr; + int last_bitpos[MAX_COMPONENTS][DCTSIZE2]; + /* -1 until that coefficient has been seen; then last Al for it */ +#endif + + if (cinfo->num_scans <= 0) + ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, 0); + +#ifndef C_MULTISCAN_FILES_SUPPORTED + if (cinfo->num_scans > 1) + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + + scanptr = cinfo->scan_info; + if (scanptr->Ss != 0 && scanptr->Se == 0) { +#ifdef C_LOSSLESS_SUPPORTED + cinfo->master->lossless = TRUE; + cinfo->progressive_mode = FALSE; + for (ci = 0; ci < cinfo->num_components; ci++) + component_sent[ci] = FALSE; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } + /* For sequential JPEG, all scans must have Ss=0, Se=DCTSIZE2-1; + * for progressive JPEG, no scan can have this. + */ + else if (scanptr->Ss != 0 || scanptr->Se != DCTSIZE2 - 1) { +#ifdef C_PROGRESSIVE_SUPPORTED + cinfo->progressive_mode = TRUE; + cinfo->master->lossless = FALSE; + last_bitpos_ptr = &last_bitpos[0][0]; + for (ci = 0; ci < cinfo->num_components; ci++) + for (coefi = 0; coefi < DCTSIZE2; coefi++) + *last_bitpos_ptr++ = -1; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + cinfo->progressive_mode = cinfo->master->lossless = FALSE; + for (ci = 0; ci < cinfo->num_components; ci++) + component_sent[ci] = FALSE; + } + + for (scanno = 1; scanno <= cinfo->num_scans; scanptr++, scanno++) { + /* Validate component indexes */ + ncomps = scanptr->comps_in_scan; + if (ncomps <= 0 || ncomps > MAX_COMPS_IN_SCAN) + ERREXIT2(cinfo, JERR_COMPONENT_COUNT, ncomps, MAX_COMPS_IN_SCAN); + for (ci = 0; ci < ncomps; ci++) { + thisi = scanptr->component_index[ci]; + if (thisi < 0 || thisi >= cinfo->num_components) + ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, scanno); + /* Components must appear in SOF order within each scan */ + if (ci > 0 && thisi <= scanptr->component_index[ci - 1]) + ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, scanno); + } + /* Validate progression parameters */ + Ss = scanptr->Ss; + Se = scanptr->Se; + Ah = scanptr->Ah; + Al = scanptr->Al; + if (cinfo->progressive_mode) { +#ifdef C_PROGRESSIVE_SUPPORTED + /* Rec. ITU-T T.81 | ISO/IEC 10918-1 simply gives the ranges 0..13 for Ah + * and Al, but that seems wrong: the upper bound ought to depend on data + * precision. Perhaps they really meant 0..N+1 for N-bit precision. + * Here we allow 0..10 for 8-bit data; Al larger than 10 results in + * out-of-range reconstructed DC values during the first DC scan, + * which might cause problems for some decoders. + */ + int max_Ah_Al = cinfo->data_precision == 12 ? 13 : 10; + + if (Ss < 0 || Ss >= DCTSIZE2 || Se < Ss || Se >= DCTSIZE2 || + Ah < 0 || Ah > max_Ah_Al || Al < 0 || Al > max_Ah_Al) + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + if (Ss == 0) { + if (Se != 0) /* DC and AC together not OK */ + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + } else { + if (ncomps != 1) /* AC scans must be for only one component */ + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + } + for (ci = 0; ci < ncomps; ci++) { + last_bitpos_ptr = &last_bitpos[scanptr->component_index[ci]][0]; + if (Ss != 0 && last_bitpos_ptr[0] < 0) /* AC without prior DC scan */ + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + for (coefi = Ss; coefi <= Se; coefi++) { + if (last_bitpos_ptr[coefi] < 0) { + /* first scan of this coefficient */ + if (Ah != 0) + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + } else { + /* not first scan */ + if (Ah != last_bitpos_ptr[coefi] || Al != Ah - 1) + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + } + last_bitpos_ptr[coefi] = Al; + } + } +#endif + } else { +#ifdef C_LOSSLESS_SUPPORTED + if (cinfo->master->lossless) { + /* The JPEG spec simply gives the range 0..15 for Al (Pt), but that + * seems wrong: the upper bound ought to depend on data precision. + * Perhaps they really meant 0..N-1 for N-bit precision, which is what + * we allow here. Values greater than or equal to the data precision + * will result in a blank image. + */ + if (Ss < 1 || Ss > 7 || /* predictor selection value */ + Se != 0 || Ah != 0 || + Al < 0 || Al >= cinfo->data_precision) /* point transform */ + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + } else +#endif + { + /* For sequential JPEG, all progression parameters must be these: */ + if (Ss != 0 || Se != DCTSIZE2 - 1 || Ah != 0 || Al != 0) + ERREXIT1(cinfo, JERR_BAD_PROG_SCRIPT, scanno); + } + /* Make sure components are not sent twice */ + for (ci = 0; ci < ncomps; ci++) { + thisi = scanptr->component_index[ci]; + if (component_sent[thisi]) + ERREXIT1(cinfo, JERR_BAD_SCAN_SCRIPT, scanno); + component_sent[thisi] = TRUE; + } + } + } + + /* Now verify that everything got sent. */ + if (cinfo->progressive_mode) { +#ifdef C_PROGRESSIVE_SUPPORTED + /* For progressive mode, we only check that at least some DC data + * got sent for each component; the spec does not require that all bits + * of all coefficients be transmitted. Would it be wiser to enforce + * transmission of all coefficient bits?? + */ + for (ci = 0; ci < cinfo->num_components; ci++) { + if (last_bitpos[ci][0] < 0) + ERREXIT(cinfo, JERR_MISSING_DATA); + } +#endif + } else { + for (ci = 0; ci < cinfo->num_components; ci++) { + if (!component_sent[ci]) + ERREXIT(cinfo, JERR_MISSING_DATA); + } + } +} + +#endif /* NEED_SCAN_SCRIPT */ + + +LOCAL(void) +select_scan_parameters(j_compress_ptr cinfo) +/* Set up the scan parameters for the current scan */ +{ + int ci; + +#ifdef NEED_SCAN_SCRIPT + if (cinfo->scan_info != NULL) { + /* Prepare for current scan --- the script is already validated */ + my_master_ptr master = (my_master_ptr)cinfo->master; + const jpeg_scan_info *scanptr = cinfo->scan_info + master->scan_number; + + cinfo->comps_in_scan = scanptr->comps_in_scan; + for (ci = 0; ci < scanptr->comps_in_scan; ci++) { + cinfo->cur_comp_info[ci] = + &cinfo->comp_info[scanptr->component_index[ci]]; + } + cinfo->Ss = scanptr->Ss; + cinfo->Se = scanptr->Se; + cinfo->Ah = scanptr->Ah; + cinfo->Al = scanptr->Al; + } else +#endif + { + /* Prepare for single sequential-JPEG scan containing all components */ + if (cinfo->num_components > MAX_COMPS_IN_SCAN) + ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components, + MAX_COMPS_IN_SCAN); + cinfo->comps_in_scan = cinfo->num_components; + for (ci = 0; ci < cinfo->num_components; ci++) { + cinfo->cur_comp_info[ci] = &cinfo->comp_info[ci]; + } + if (!cinfo->master->lossless) { + cinfo->Ss = 0; + cinfo->Se = DCTSIZE2 - 1; + cinfo->Ah = 0; + cinfo->Al = 0; + } + } +} + + +LOCAL(void) +per_scan_setup(j_compress_ptr cinfo) +/* Do computations that are needed before processing a JPEG scan */ +/* cinfo->comps_in_scan and cinfo->cur_comp_info[] are already set */ +{ + int ci, mcublks, tmp; + jpeg_component_info *compptr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + if (cinfo->comps_in_scan == 1) { + + /* Noninterleaved (single-component) scan */ + compptr = cinfo->cur_comp_info[0]; + + /* Overall image size in MCUs */ + cinfo->MCUs_per_row = compptr->width_in_blocks; + cinfo->MCU_rows_in_scan = compptr->height_in_blocks; + + /* For noninterleaved scan, always one block per MCU */ + compptr->MCU_width = 1; + compptr->MCU_height = 1; + compptr->MCU_blocks = 1; + compptr->MCU_sample_width = data_unit; + compptr->last_col_width = 1; + /* For noninterleaved scans, it is convenient to define last_row_height + * as the number of block rows present in the last iMCU row. + */ + tmp = (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (tmp == 0) tmp = compptr->v_samp_factor; + compptr->last_row_height = tmp; + + /* Prepare array describing MCU composition */ + cinfo->blocks_in_MCU = 1; + cinfo->MCU_membership[0] = 0; + + } else { + + /* Interleaved (multi-component) scan */ + if (cinfo->comps_in_scan <= 0 || cinfo->comps_in_scan > MAX_COMPS_IN_SCAN) + ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->comps_in_scan, + MAX_COMPS_IN_SCAN); + + /* Overall image size in MCUs */ + cinfo->MCUs_per_row = (JDIMENSION) + jdiv_round_up((long)cinfo->_jpeg_width, + (long)(cinfo->max_h_samp_factor * data_unit)); + cinfo->MCU_rows_in_scan = (JDIMENSION) + jdiv_round_up((long)cinfo->_jpeg_height, + (long)(cinfo->max_v_samp_factor * data_unit)); + + cinfo->blocks_in_MCU = 0; + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* Sampling factors give # of blocks of component in each MCU */ + compptr->MCU_width = compptr->h_samp_factor; + compptr->MCU_height = compptr->v_samp_factor; + compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height; + compptr->MCU_sample_width = compptr->MCU_width * data_unit; + /* Figure number of non-dummy blocks in last MCU column & row */ + tmp = (int)(compptr->width_in_blocks % compptr->MCU_width); + if (tmp == 0) tmp = compptr->MCU_width; + compptr->last_col_width = tmp; + tmp = (int)(compptr->height_in_blocks % compptr->MCU_height); + if (tmp == 0) tmp = compptr->MCU_height; + compptr->last_row_height = tmp; + /* Prepare array describing MCU composition */ + mcublks = compptr->MCU_blocks; + if (cinfo->blocks_in_MCU + mcublks > C_MAX_BLOCKS_IN_MCU) + ERREXIT(cinfo, JERR_BAD_MCU_SIZE); + while (mcublks-- > 0) { + cinfo->MCU_membership[cinfo->blocks_in_MCU++] = ci; + } + } + + } + + /* Convert restart specified in rows to actual MCU count. */ + /* Note that count must fit in 16 bits, so we provide limiting. */ + if (cinfo->restart_in_rows > 0) { + long nominal = (long)cinfo->restart_in_rows * (long)cinfo->MCUs_per_row; + cinfo->restart_interval = (unsigned int)MIN(nominal, 65535L); + } +} + + +/* + * Per-pass setup. + * This is called at the beginning of each pass. We determine which modules + * will be active during this pass and give them appropriate start_pass calls. + * We also set is_last_pass to indicate whether any more passes will be + * required. + */ + +METHODDEF(void) +prepare_for_pass(j_compress_ptr cinfo) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + + switch (master->pass_type) { + case main_pass: + /* Initial pass: will collect input data, and do either Huffman + * optimization or data output for the first scan. + */ + select_scan_parameters(cinfo); + per_scan_setup(cinfo); + if (!cinfo->raw_data_in) { + (*cinfo->cconvert->start_pass) (cinfo); + (*cinfo->downsample->start_pass) (cinfo); + (*cinfo->prep->start_pass) (cinfo, JBUF_PASS_THRU); + } + (*cinfo->fdct->start_pass) (cinfo); + (*cinfo->entropy->start_pass) (cinfo, cinfo->optimize_coding); + (*cinfo->coef->start_pass) (cinfo, + (master->total_passes > 1 ? + JBUF_SAVE_AND_PASS : JBUF_PASS_THRU)); + (*cinfo->main->start_pass) (cinfo, JBUF_PASS_THRU); + if (cinfo->optimize_coding) { + /* No immediate data output; postpone writing frame/scan headers */ + master->pub.call_pass_startup = FALSE; + } else { + /* Will write frame/scan headers at first jpeg_write_scanlines call */ + master->pub.call_pass_startup = TRUE; + } + break; +#ifdef ENTROPY_OPT_SUPPORTED + case huff_opt_pass: + /* Do Huffman optimization for a scan after the first one. */ + select_scan_parameters(cinfo); + per_scan_setup(cinfo); + if (cinfo->Ss != 0 || cinfo->Ah == 0 || cinfo->arith_code || + cinfo->master->lossless) { + (*cinfo->entropy->start_pass) (cinfo, TRUE); + (*cinfo->coef->start_pass) (cinfo, JBUF_CRANK_DEST); + master->pub.call_pass_startup = FALSE; + break; + } + /* Special case: Huffman DC refinement scans need no Huffman table + * and therefore we can skip the optimization pass for them. + */ + master->pass_type = output_pass; + master->pass_number++; +#endif + FALLTHROUGH /*FALLTHROUGH*/ + case output_pass: + /* Do a data-output pass. */ + /* We need not repeat per-scan setup if prior optimization pass did it. */ + if (!cinfo->optimize_coding) { + select_scan_parameters(cinfo); + per_scan_setup(cinfo); + } + (*cinfo->entropy->start_pass) (cinfo, FALSE); + (*cinfo->coef->start_pass) (cinfo, JBUF_CRANK_DEST); + /* We emit frame/scan headers now */ + if (master->scan_number == 0) + (*cinfo->marker->write_frame_header) (cinfo); + (*cinfo->marker->write_scan_header) (cinfo); + master->pub.call_pass_startup = FALSE; + break; + default: + ERREXIT(cinfo, JERR_NOT_COMPILED); + } + + master->pub.is_last_pass = (master->pass_number == master->total_passes - 1); + + /* Set up progress monitor's pass info if present */ + if (cinfo->progress != NULL) { + cinfo->progress->completed_passes = master->pass_number; + cinfo->progress->total_passes = master->total_passes; + } +} + + +/* + * Special start-of-pass hook. + * This is called by jpeg_write_scanlines if call_pass_startup is TRUE. + * In single-pass processing, we need this hook because we don't want to + * write frame/scan headers during jpeg_start_compress; we want to let the + * application write COM markers etc. between jpeg_start_compress and the + * jpeg_write_scanlines loop. + * In multi-pass processing, this routine is not used. + */ + +METHODDEF(void) +pass_startup(j_compress_ptr cinfo) +{ + cinfo->master->call_pass_startup = FALSE; /* reset flag so call only once */ + + (*cinfo->marker->write_frame_header) (cinfo); + (*cinfo->marker->write_scan_header) (cinfo); +} + + +/* + * Finish up at end of pass. + */ + +METHODDEF(void) +finish_pass_master(j_compress_ptr cinfo) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + + /* The entropy coder always needs an end-of-pass call, + * either to analyze statistics or to flush its output buffer. + */ + (*cinfo->entropy->finish_pass) (cinfo); + + /* Update state for next pass */ + switch (master->pass_type) { + case main_pass: + /* next pass is either output of scan 0 (after optimization) + * or output of scan 1 (if no optimization). + */ + master->pass_type = output_pass; + if (!cinfo->optimize_coding) + master->scan_number++; + break; + case huff_opt_pass: + /* next pass is always output of current scan */ + master->pass_type = output_pass; + break; + case output_pass: + /* next pass is either optimization or output of next scan */ + if (cinfo->optimize_coding) + master->pass_type = huff_opt_pass; + master->scan_number++; + break; + } + + master->pass_number++; +} + + +/* + * Initialize master compression control. + */ + +GLOBAL(void) +jinit_c_master_control(j_compress_ptr cinfo, boolean transcode_only) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + boolean empty_huff_tables = TRUE; + int i; + + master->pub.prepare_for_pass = prepare_for_pass; + master->pub.pass_startup = pass_startup; + master->pub.finish_pass = finish_pass_master; + master->pub.is_last_pass = FALSE; + + if (cinfo->scan_info != NULL) { +#ifdef NEED_SCAN_SCRIPT + validate_script(cinfo); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + cinfo->progressive_mode = FALSE; + cinfo->num_scans = 1; + } + + /* Disable smoothing and subsampling in lossless mode, since those are lossy + * algorithms. Set the JPEG colorspace to the input colorspace. Disable raw + * (downsampled) data input, because it isn't particularly useful without + * subsampling and has not been tested in lossless mode. + */ + if (cinfo->master->lossless) { + int ci; + jpeg_component_info *compptr; + + cinfo->raw_data_in = FALSE; + cinfo->smoothing_factor = 0; + jpeg_default_colorspace(cinfo); + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) + compptr->h_samp_factor = compptr->v_samp_factor = 1; + } + + /* Validate parameters, determine derived values */ + initial_setup(cinfo, transcode_only); + + if (cinfo->arith_code) + cinfo->optimize_coding = FALSE; + else { + if (cinfo->master->lossless || /* TEMPORARY HACK ??? */ + cinfo->progressive_mode) + cinfo->optimize_coding = TRUE; /* assume default tables no good for + progressive mode or lossless mode */ + for (i = 0; i < NUM_HUFF_TBLS; i++) { + if (cinfo->dc_huff_tbl_ptrs[i] != NULL || + cinfo->ac_huff_tbl_ptrs[i] != NULL) { + empty_huff_tables = FALSE; + break; + } + } + if (cinfo->data_precision == 12 && !cinfo->optimize_coding && + (empty_huff_tables || using_std_huff_tables(cinfo))) + cinfo->optimize_coding = TRUE; /* assume default tables no good for + 12-bit data precision */ + } + + /* Initialize my private state */ + if (transcode_only) { + /* no main pass in transcoding */ + if (cinfo->optimize_coding) + master->pass_type = huff_opt_pass; + else + master->pass_type = output_pass; + } else { + /* for normal compression, first pass is always this type: */ + master->pass_type = main_pass; + } + master->scan_number = 0; + master->pass_number = 0; + if (cinfo->optimize_coding) + master->total_passes = cinfo->num_scans * 2; + else + master->total_passes = cinfo->num_scans; + + master->jpeg_version = PACKAGE_NAME " version " VERSION " (build " BUILD ")"; +} diff --git a/src/spice2x/external/libjpeg-turbo/jcmaster.h b/src/spice2x/external/libjpeg-turbo/jcmaster.h new file mode 100644 index 0000000..3b13289 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcmaster.h @@ -0,0 +1,43 @@ +/* + * jcmaster.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1995, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2016, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains master control structure for the JPEG compressor. + */ + +/* Private state */ + +typedef enum { + main_pass, /* input data, also do first output step */ + huff_opt_pass, /* Huffman code optimization pass */ + output_pass /* data output pass */ +} c_pass_type; + +typedef struct { + struct jpeg_comp_master pub; /* public fields */ + + c_pass_type pass_type; /* the type of the current pass */ + + int pass_number; /* # of passes completed */ + int total_passes; /* total # of passes needed */ + + int scan_number; /* current index in scan_info[] */ + + /* + * This is here so we can add libjpeg-turbo version/build information to the + * global string table without introducing a new global symbol. Adding this + * information to the global string table allows one to examine a binary + * object and determine which version of libjpeg-turbo it was built from or + * linked against. + */ + const char *jpeg_version; + +} my_comp_master; + +typedef my_comp_master *my_master_ptr; diff --git a/src/spice2x/external/libjpeg-turbo/jcomapi.c b/src/spice2x/external/libjpeg-turbo/jcomapi.c new file mode 100644 index 0000000..84f37e1 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcomapi.c @@ -0,0 +1,110 @@ +/* + * jcomapi.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains application interface routines that are used for both + * compression and decompression. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" + + +/* + * Abort processing of a JPEG compression or decompression operation, + * but don't destroy the object itself. + * + * For this, we merely clean up all the nonpermanent memory pools. + * Note that temp files (virtual arrays) are not allowed to belong to + * the permanent pool, so we will be able to close all temp files here. + * Closing a data source or destination, if necessary, is the application's + * responsibility. + */ + +GLOBAL(void) +jpeg_abort(j_common_ptr cinfo) +{ + int pool; + + /* Do nothing if called on a not-initialized or destroyed JPEG object. */ + if (cinfo->mem == NULL) + return; + + /* Releasing pools in reverse order might help avoid fragmentation + * with some (brain-damaged) malloc libraries. + */ + for (pool = JPOOL_NUMPOOLS - 1; pool > JPOOL_PERMANENT; pool--) { + (*cinfo->mem->free_pool) (cinfo, pool); + } + + /* Reset overall state for possible reuse of object */ + if (cinfo->is_decompressor) { + cinfo->global_state = DSTATE_START; + /* Try to keep application from accessing now-deleted marker list. + * A bit kludgy to do it here, but this is the most central place. + */ + ((j_decompress_ptr)cinfo)->marker_list = NULL; + ((j_decompress_ptr)cinfo)->master->marker_list_end = NULL; + } else { + cinfo->global_state = CSTATE_START; + } +} + + +/* + * Destruction of a JPEG object. + * + * Everything gets deallocated except the master jpeg_compress_struct itself + * and the error manager struct. Both of these are supplied by the application + * and must be freed, if necessary, by the application. (Often they are on + * the stack and so don't need to be freed anyway.) + * Closing a data source or destination, if necessary, is the application's + * responsibility. + */ + +GLOBAL(void) +jpeg_destroy(j_common_ptr cinfo) +{ + /* We need only tell the memory manager to release everything. */ + /* NB: mem pointer is NULL if memory mgr failed to initialize. */ + if (cinfo->mem != NULL) + (*cinfo->mem->self_destruct) (cinfo); + cinfo->mem = NULL; /* be safe if jpeg_destroy is called twice */ + cinfo->global_state = 0; /* mark it destroyed */ +} + + +/* + * Convenience routines for allocating quantization and Huffman tables. + * (Would jutils.c be a more reasonable place to put these?) + */ + +GLOBAL(JQUANT_TBL *) +jpeg_alloc_quant_table(j_common_ptr cinfo) +{ + JQUANT_TBL *tbl; + + tbl = (JQUANT_TBL *) + (*cinfo->mem->alloc_small) (cinfo, JPOOL_PERMANENT, sizeof(JQUANT_TBL)); + tbl->sent_table = FALSE; /* make sure this is false in any new table */ + return tbl; +} + + +GLOBAL(JHUFF_TBL *) +jpeg_alloc_huff_table(j_common_ptr cinfo) +{ + JHUFF_TBL *tbl; + + tbl = (JHUFF_TBL *) + (*cinfo->mem->alloc_small) (cinfo, JPOOL_PERMANENT, sizeof(JHUFF_TBL)); + tbl->sent_table = FALSE; /* make sure this is false in any new table */ + return tbl; +} diff --git a/src/spice2x/external/libjpeg-turbo/jconfig.h.in b/src/spice2x/external/libjpeg-turbo/jconfig.h.in new file mode 100644 index 0000000..6cb8296 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jconfig.h.in @@ -0,0 +1,60 @@ +/* Version ID for the JPEG library. + * Might be useful for tests like "#if JPEG_LIB_VERSION >= 60". + */ +#define JPEG_LIB_VERSION @JPEG_LIB_VERSION@ + +/* libjpeg-turbo version */ +#define LIBJPEG_TURBO_VERSION @VERSION@ + +/* libjpeg-turbo version in integer form */ +#define LIBJPEG_TURBO_VERSION_NUMBER @LIBJPEG_TURBO_VERSION_NUMBER@ + +/* Support arithmetic encoding when using 8-bit samples */ +#cmakedefine C_ARITH_CODING_SUPPORTED 1 + +/* Support arithmetic decoding when using 8-bit samples */ +#cmakedefine D_ARITH_CODING_SUPPORTED 1 + +/* Support in-memory source/destination managers */ +#define MEM_SRCDST_SUPPORTED 1 + +/* Use accelerated SIMD routines when using 8-bit samples */ +#cmakedefine WITH_SIMD 1 + +/* This version of libjpeg-turbo supports run-time selection of data precision, + * so BITS_IN_JSAMPLE is no longer used to specify the data precision at build + * time. However, some downstream software expects the macro to be defined. + * Since 12-bit data precision is an opt-in feature that requires explicitly + * calling 12-bit-specific libjpeg API functions and using 12-bit-specific data + * types, the unmodified portion of the libjpeg API still behaves as if it were + * built for 8-bit precision, and JSAMPLE is still literally an 8-bit data + * type. Thus, it is correct to define BITS_IN_JSAMPLE to 8 here. + */ +#ifndef BITS_IN_JSAMPLE +#define BITS_IN_JSAMPLE 8 +#endif + +#ifdef _WIN32 + +#undef RIGHT_SHIFT_IS_UNSIGNED + +/* Define "boolean" as unsigned char, not int, per Windows custom */ +#ifndef __RPCNDR_H__ /* don't conflict if rpcndr.h already read */ +typedef unsigned char boolean; +#endif +#define HAVE_BOOLEAN /* prevent jmorecfg.h from redefining it */ + +/* Define "INT32" as int, not long, per Windows custom */ +#if !(defined(_BASETSD_H_) || defined(_BASETSD_H)) /* don't conflict if basetsd.h already read */ +typedef short INT16; +typedef signed int INT32; +#endif +#define XMD_H /* prevent jmorecfg.h from redefining it */ + +#else + +/* Define if your (broken) compiler shifts signed values as if they were + unsigned. */ +#cmakedefine RIGHT_SHIFT_IS_UNSIGNED 1 + +#endif diff --git a/src/spice2x/external/libjpeg-turbo/jconfigint.h.in b/src/spice2x/external/libjpeg-turbo/jconfigint.h.in new file mode 100644 index 0000000..5c14e32 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jconfigint.h.in @@ -0,0 +1,76 @@ +/* libjpeg-turbo build number */ +#define BUILD "@BUILD@" + +/* How to hide global symbols. */ +#define HIDDEN @HIDDEN@ + +/* Compiler's inline keyword */ +#undef inline + +/* How to obtain function inlining. */ +#define INLINE @INLINE@ + +/* How to obtain thread-local storage */ +#define THREAD_LOCAL @THREAD_LOCAL@ + +/* Define to the full name of this package. */ +#define PACKAGE_NAME "@CMAKE_PROJECT_NAME@" + +/* Version number of package */ +#define VERSION "@VERSION@" + +/* The size of `size_t', as computed by sizeof. */ +#define SIZEOF_SIZE_T @SIZE_T@ + +/* Define if your compiler has __builtin_ctzl() and sizeof(unsigned long) == sizeof(size_t). */ +#cmakedefine HAVE_BUILTIN_CTZL + +/* Define to 1 if you have the header file. */ +#cmakedefine HAVE_INTRIN_H + +#if defined(_MSC_VER) && defined(HAVE_INTRIN_H) +#if (SIZEOF_SIZE_T == 8) +#define HAVE_BITSCANFORWARD64 +#elif (SIZEOF_SIZE_T == 4) +#define HAVE_BITSCANFORWARD +#endif +#endif + +#if defined(__has_attribute) +#if __has_attribute(fallthrough) +#define FALLTHROUGH __attribute__((fallthrough)); +#else +#define FALLTHROUGH +#endif +#else +#define FALLTHROUGH +#endif + +/* + * Define BITS_IN_JSAMPLE as either + * 8 for 8-bit sample values (the usual setting) + * 12 for 12-bit sample values + * Only 8 and 12 are legal data precisions for lossy JPEG according to the + * JPEG standard, and the IJG code does not support anything else! + */ + +#ifndef BITS_IN_JSAMPLE +#define BITS_IN_JSAMPLE 8 /* use 8 or 12 */ +#endif + +#undef C_ARITH_CODING_SUPPORTED +#undef D_ARITH_CODING_SUPPORTED +#undef WITH_SIMD + +#if BITS_IN_JSAMPLE == 8 + +/* Support arithmetic encoding */ +#cmakedefine C_ARITH_CODING_SUPPORTED 1 + +/* Support arithmetic decoding */ +#cmakedefine D_ARITH_CODING_SUPPORTED 1 + +/* Use accelerated SIMD routines. */ +#cmakedefine WITH_SIMD 1 + +#endif diff --git a/src/spice2x/external/libjpeg-turbo/jcparam.c b/src/spice2x/external/libjpeg-turbo/jcparam.c new file mode 100644 index 0000000..1d6e225 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcparam.c @@ -0,0 +1,585 @@ +/* + * jcparam.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1998, Thomas G. Lane. + * Modified 2003-2008 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2009-2011, 2018, 2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains optional default-setting code for the JPEG compressor. + * Applications do not have to use this file, but those that don't use it + * must know a lot more about the innards of the JPEG code. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jstdhuff.c" + + +/* + * Quantization table setup routines + */ + +GLOBAL(void) +jpeg_add_quant_table(j_compress_ptr cinfo, int which_tbl, + const unsigned int *basic_table, int scale_factor, + boolean force_baseline) +/* Define a quantization table equal to the basic_table times + * a scale factor (given as a percentage). + * If force_baseline is TRUE, the computed quantization table entries + * are limited to 1..255 for JPEG baseline compatibility. + */ +{ + JQUANT_TBL **qtblptr; + int i; + long temp; + + /* Safety check to ensure start_compress not called yet. */ + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + if (which_tbl < 0 || which_tbl >= NUM_QUANT_TBLS) + ERREXIT1(cinfo, JERR_DQT_INDEX, which_tbl); + + qtblptr = &cinfo->quant_tbl_ptrs[which_tbl]; + + if (*qtblptr == NULL) + *qtblptr = jpeg_alloc_quant_table((j_common_ptr)cinfo); + + for (i = 0; i < DCTSIZE2; i++) { + temp = ((long)basic_table[i] * scale_factor + 50L) / 100L; + /* limit the values to the valid range */ + if (temp <= 0L) temp = 1L; + if (temp > 32767L) temp = 32767L; /* max quantizer needed for 12 bits */ + if (force_baseline && temp > 255L) + temp = 255L; /* limit to baseline range if requested */ + (*qtblptr)->quantval[i] = (UINT16)temp; + } + + /* Initialize sent_table FALSE so table will be written to JPEG file. */ + (*qtblptr)->sent_table = FALSE; +} + + +/* These are the sample quantization tables given in Annex K (Clause K.1) of + * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994. + * The spec says that the values given produce "good" quality, and + * when divided by 2, "very good" quality. + */ +static const unsigned int std_luminance_quant_tbl[DCTSIZE2] = { + 16, 11, 10, 16, 24, 40, 51, 61, + 12, 12, 14, 19, 26, 58, 60, 55, + 14, 13, 16, 24, 40, 57, 69, 56, + 14, 17, 22, 29, 51, 87, 80, 62, + 18, 22, 37, 56, 68, 109, 103, 77, + 24, 35, 55, 64, 81, 104, 113, 92, + 49, 64, 78, 87, 103, 121, 120, 101, + 72, 92, 95, 98, 112, 100, 103, 99 +}; +static const unsigned int std_chrominance_quant_tbl[DCTSIZE2] = { + 17, 18, 24, 47, 99, 99, 99, 99, + 18, 21, 26, 66, 99, 99, 99, 99, + 24, 26, 56, 99, 99, 99, 99, 99, + 47, 66, 99, 99, 99, 99, 99, 99, + 99, 99, 99, 99, 99, 99, 99, 99, + 99, 99, 99, 99, 99, 99, 99, 99, + 99, 99, 99, 99, 99, 99, 99, 99, + 99, 99, 99, 99, 99, 99, 99, 99 +}; + + +#if JPEG_LIB_VERSION >= 70 +GLOBAL(void) +jpeg_default_qtables(j_compress_ptr cinfo, boolean force_baseline) +/* Set or change the 'quality' (quantization) setting, using default tables + * and straight percentage-scaling quality scales. + * This entry point allows different scalings for luminance and chrominance. + */ +{ + /* Set up two quantization tables using the specified scaling */ + jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl, + cinfo->q_scale_factor[0], force_baseline); + jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl, + cinfo->q_scale_factor[1], force_baseline); +} +#endif + + +GLOBAL(void) +jpeg_set_linear_quality(j_compress_ptr cinfo, int scale_factor, + boolean force_baseline) +/* Set or change the 'quality' (quantization) setting, using default tables + * and a straight percentage-scaling quality scale. In most cases it's better + * to use jpeg_set_quality (below); this entry point is provided for + * applications that insist on a linear percentage scaling. + */ +{ + /* Set up two quantization tables using the specified scaling */ + jpeg_add_quant_table(cinfo, 0, std_luminance_quant_tbl, + scale_factor, force_baseline); + jpeg_add_quant_table(cinfo, 1, std_chrominance_quant_tbl, + scale_factor, force_baseline); +} + + +GLOBAL(int) +jpeg_quality_scaling(int quality) +/* Convert a user-specified quality rating to a percentage scaling factor + * for an underlying quantization table, using our recommended scaling curve. + * The input 'quality' factor should be 0 (terrible) to 100 (very good). + */ +{ + /* Safety limit on quality factor. Convert 0 to 1 to avoid zero divide. */ + if (quality <= 0) quality = 1; + if (quality > 100) quality = 100; + + /* The basic table is used as-is (scaling 100) for a quality of 50. + * Qualities 50..100 are converted to scaling percentage 200 - 2*Q; + * note that at Q=100 the scaling is 0, which will cause jpeg_add_quant_table + * to make all the table entries 1 (hence, minimum quantization loss). + * Qualities 1..50 are converted to scaling percentage 5000/Q. + */ + if (quality < 50) + quality = 5000 / quality; + else + quality = 200 - quality * 2; + + return quality; +} + + +GLOBAL(void) +jpeg_set_quality(j_compress_ptr cinfo, int quality, boolean force_baseline) +/* Set or change the 'quality' (quantization) setting, using default tables. + * This is the standard quality-adjusting entry point for typical user + * interfaces; only those who want detailed control over quantization tables + * would use the preceding three routines directly. + */ +{ + /* Convert user 0-100 rating to percentage scaling */ + quality = jpeg_quality_scaling(quality); + + /* Set up standard quality tables */ + jpeg_set_linear_quality(cinfo, quality, force_baseline); +} + + +/* + * Default parameter setup for compression. + * + * Applications that don't choose to use this routine must do their + * own setup of all these parameters. Alternately, you can call this + * to establish defaults and then alter parameters selectively. This + * is the recommended approach since, if we add any new parameters, + * your code will still work (they'll be set to reasonable defaults). + */ + +GLOBAL(void) +jpeg_set_defaults(j_compress_ptr cinfo) +{ + int i; + + /* Safety check to ensure start_compress not called yet. */ + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + /* Allocate comp_info array large enough for maximum component count. + * Array is made permanent in case application wants to compress + * multiple images at same param settings. + */ + if (cinfo->comp_info == NULL) + cinfo->comp_info = (jpeg_component_info *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + MAX_COMPONENTS * sizeof(jpeg_component_info)); + + /* Initialize everything not dependent on the color space */ + +#if JPEG_LIB_VERSION >= 70 + cinfo->scale_num = 1; /* 1:1 scaling */ + cinfo->scale_denom = 1; +#endif + /* Set up two quantization tables using default quality of 75 */ + jpeg_set_quality(cinfo, 75, TRUE); + /* Set up two Huffman tables */ + std_huff_tables((j_common_ptr)cinfo); + + /* Initialize default arithmetic coding conditioning */ + for (i = 0; i < NUM_ARITH_TBLS; i++) { + cinfo->arith_dc_L[i] = 0; + cinfo->arith_dc_U[i] = 1; + cinfo->arith_ac_K[i] = 5; + } + + /* Default is no multiple-scan output */ + cinfo->scan_info = NULL; + cinfo->num_scans = 0; + + /* Expect normal source image, not raw downsampled data */ + cinfo->raw_data_in = FALSE; + + /* Use Huffman coding, not arithmetic coding, by default */ + cinfo->arith_code = FALSE; + + /* By default, don't do extra passes to optimize entropy coding */ + cinfo->optimize_coding = FALSE; + /* The standard Huffman tables are only valid for 8-bit data precision. + * If the precision is higher, force optimization on so that usable + * tables will be computed. This test can be removed if default tables + * are supplied that are valid for the desired precision. + */ + if (cinfo->data_precision == 12) + cinfo->optimize_coding = TRUE; + + /* By default, use the simpler non-cosited sampling alignment */ + cinfo->CCIR601_sampling = FALSE; + +#if JPEG_LIB_VERSION >= 70 + /* By default, apply fancy downsampling */ + cinfo->do_fancy_downsampling = TRUE; +#endif + + /* No input smoothing */ + cinfo->smoothing_factor = 0; + + /* DCT algorithm preference */ + cinfo->dct_method = JDCT_DEFAULT; + + /* No restart markers */ + cinfo->restart_interval = 0; + cinfo->restart_in_rows = 0; + + /* Fill in default JFIF marker parameters. Note that whether the marker + * will actually be written is determined by jpeg_set_colorspace. + * + * By default, the library emits JFIF version code 1.01. + * An application that wants to emit JFIF 1.02 extension markers should set + * JFIF_minor_version to 2. We could probably get away with just defaulting + * to 1.02, but there may still be some decoders in use that will complain + * about that; saying 1.01 should minimize compatibility problems. + */ + cinfo->JFIF_major_version = 1; /* Default JFIF version = 1.01 */ + cinfo->JFIF_minor_version = 1; + cinfo->density_unit = 0; /* Pixel size is unknown by default */ + cinfo->X_density = 1; /* Pixel aspect ratio is square by default */ + cinfo->Y_density = 1; + + /* Choose JPEG colorspace based on input space, set defaults accordingly */ + + jpeg_default_colorspace(cinfo); +} + + +/* + * Select an appropriate JPEG colorspace for in_color_space. + */ + +GLOBAL(void) +jpeg_default_colorspace(j_compress_ptr cinfo) +{ + switch (cinfo->in_color_space) { + case JCS_GRAYSCALE: + jpeg_set_colorspace(cinfo, JCS_GRAYSCALE); + break; + case JCS_RGB: + case JCS_EXT_RGB: + case JCS_EXT_RGBX: + case JCS_EXT_BGR: + case JCS_EXT_BGRX: + case JCS_EXT_XBGR: + case JCS_EXT_XRGB: + case JCS_EXT_RGBA: + case JCS_EXT_BGRA: + case JCS_EXT_ABGR: + case JCS_EXT_ARGB: + if (cinfo->master->lossless) + jpeg_set_colorspace(cinfo, JCS_RGB); + else + jpeg_set_colorspace(cinfo, JCS_YCbCr); + break; + case JCS_YCbCr: + jpeg_set_colorspace(cinfo, JCS_YCbCr); + break; + case JCS_CMYK: + jpeg_set_colorspace(cinfo, JCS_CMYK); /* By default, no translation */ + break; + case JCS_YCCK: + jpeg_set_colorspace(cinfo, JCS_YCCK); + break; + case JCS_UNKNOWN: + jpeg_set_colorspace(cinfo, JCS_UNKNOWN); + break; + default: + ERREXIT(cinfo, JERR_BAD_IN_COLORSPACE); + } +} + + +/* + * Set the JPEG colorspace, and choose colorspace-dependent default values. + */ + +GLOBAL(void) +jpeg_set_colorspace(j_compress_ptr cinfo, J_COLOR_SPACE colorspace) +{ + jpeg_component_info *compptr; + int ci; + +#define SET_COMP(index, id, hsamp, vsamp, quant, dctbl, actbl) \ + (compptr = &cinfo->comp_info[index], \ + compptr->component_id = (id), \ + compptr->h_samp_factor = (hsamp), \ + compptr->v_samp_factor = (vsamp), \ + compptr->quant_tbl_no = (quant), \ + compptr->dc_tbl_no = (dctbl), \ + compptr->ac_tbl_no = (actbl) ) + + /* Safety check to ensure start_compress not called yet. */ + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + /* For all colorspaces, we use Q and Huff tables 0 for luminance components, + * tables 1 for chrominance components. + */ + + cinfo->jpeg_color_space = colorspace; + + cinfo->write_JFIF_header = FALSE; /* No marker for non-JFIF colorspaces */ + cinfo->write_Adobe_marker = FALSE; /* write no Adobe marker by default */ + + switch (colorspace) { + case JCS_GRAYSCALE: + cinfo->write_JFIF_header = TRUE; /* Write a JFIF marker */ + cinfo->num_components = 1; + /* JFIF specifies component ID 1 */ + SET_COMP(0, 1, 1, 1, 0, 0, 0); + break; + case JCS_RGB: + cinfo->write_Adobe_marker = TRUE; /* write Adobe marker to flag RGB */ + cinfo->num_components = 3; + SET_COMP(0, 0x52 /* 'R' */, 1, 1, 0, 0, 0); + SET_COMP(1, 0x47 /* 'G' */, 1, 1, 0, 0, 0); + SET_COMP(2, 0x42 /* 'B' */, 1, 1, 0, 0, 0); + break; + case JCS_YCbCr: + cinfo->write_JFIF_header = TRUE; /* Write a JFIF marker */ + cinfo->num_components = 3; + /* JFIF specifies component IDs 1,2,3 */ + /* We default to 2x2 subsamples of chrominance */ + SET_COMP(0, 1, 2, 2, 0, 0, 0); + SET_COMP(1, 2, 1, 1, 1, 1, 1); + SET_COMP(2, 3, 1, 1, 1, 1, 1); + break; + case JCS_CMYK: + cinfo->write_Adobe_marker = TRUE; /* write Adobe marker to flag CMYK */ + cinfo->num_components = 4; + SET_COMP(0, 0x43 /* 'C' */, 1, 1, 0, 0, 0); + SET_COMP(1, 0x4D /* 'M' */, 1, 1, 0, 0, 0); + SET_COMP(2, 0x59 /* 'Y' */, 1, 1, 0, 0, 0); + SET_COMP(3, 0x4B /* 'K' */, 1, 1, 0, 0, 0); + break; + case JCS_YCCK: + cinfo->write_Adobe_marker = TRUE; /* write Adobe marker to flag YCCK */ + cinfo->num_components = 4; + SET_COMP(0, 1, 2, 2, 0, 0, 0); + SET_COMP(1, 2, 1, 1, 1, 1, 1); + SET_COMP(2, 3, 1, 1, 1, 1, 1); + SET_COMP(3, 4, 2, 2, 0, 0, 0); + break; + case JCS_UNKNOWN: + cinfo->num_components = cinfo->input_components; + if (cinfo->num_components < 1 || cinfo->num_components > MAX_COMPONENTS) + ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components, + MAX_COMPONENTS); + for (ci = 0; ci < cinfo->num_components; ci++) { + SET_COMP(ci, ci, 1, 1, 0, 0, 0); + } + break; + default: + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + } +} + + +#ifdef C_PROGRESSIVE_SUPPORTED + +LOCAL(jpeg_scan_info *) +fill_a_scan(jpeg_scan_info *scanptr, int ci, int Ss, int Se, int Ah, int Al) +/* Support routine: generate one scan for specified component */ +{ + scanptr->comps_in_scan = 1; + scanptr->component_index[0] = ci; + scanptr->Ss = Ss; + scanptr->Se = Se; + scanptr->Ah = Ah; + scanptr->Al = Al; + scanptr++; + return scanptr; +} + +LOCAL(jpeg_scan_info *) +fill_scans(jpeg_scan_info *scanptr, int ncomps, int Ss, int Se, int Ah, int Al) +/* Support routine: generate one scan for each component */ +{ + int ci; + + for (ci = 0; ci < ncomps; ci++) { + scanptr->comps_in_scan = 1; + scanptr->component_index[0] = ci; + scanptr->Ss = Ss; + scanptr->Se = Se; + scanptr->Ah = Ah; + scanptr->Al = Al; + scanptr++; + } + return scanptr; +} + +LOCAL(jpeg_scan_info *) +fill_dc_scans(jpeg_scan_info *scanptr, int ncomps, int Ah, int Al) +/* Support routine: generate interleaved DC scan if possible, else N scans */ +{ + int ci; + + if (ncomps <= MAX_COMPS_IN_SCAN) { + /* Single interleaved DC scan */ + scanptr->comps_in_scan = ncomps; + for (ci = 0; ci < ncomps; ci++) + scanptr->component_index[ci] = ci; + scanptr->Ss = scanptr->Se = 0; + scanptr->Ah = Ah; + scanptr->Al = Al; + scanptr++; + } else { + /* Noninterleaved DC scan for each component */ + scanptr = fill_scans(scanptr, ncomps, 0, 0, Ah, Al); + } + return scanptr; +} + + +/* + * Create a recommended progressive-JPEG script. + * cinfo->num_components and cinfo->jpeg_color_space must be correct. + */ + +GLOBAL(void) +jpeg_simple_progression(j_compress_ptr cinfo) +{ + int ncomps = cinfo->num_components; + int nscans; + jpeg_scan_info *scanptr; + + /* Safety check to ensure start_compress not called yet. */ + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + if (cinfo->master->lossless) { + cinfo->master->lossless = FALSE; + jpeg_default_colorspace(cinfo); + } + + /* Figure space needed for script. Calculation must match code below! */ + if (ncomps == 3 && cinfo->jpeg_color_space == JCS_YCbCr) { + /* Custom script for YCbCr color images. */ + nscans = 10; + } else { + /* All-purpose script for other color spaces. */ + if (ncomps > MAX_COMPS_IN_SCAN) + nscans = 6 * ncomps; /* 2 DC + 4 AC scans per component */ + else + nscans = 2 + 4 * ncomps; /* 2 DC scans; 4 AC scans per component */ + } + + /* Allocate space for script. + * We need to put it in the permanent pool in case the application performs + * multiple compressions without changing the settings. To avoid a memory + * leak if jpeg_simple_progression is called repeatedly for the same JPEG + * object, we try to re-use previously allocated space, and we allocate + * enough space to handle YCbCr even if initially asked for grayscale. + */ + if (cinfo->script_space == NULL || cinfo->script_space_size < nscans) { + cinfo->script_space_size = MAX(nscans, 10); + cinfo->script_space = (jpeg_scan_info *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + cinfo->script_space_size * sizeof(jpeg_scan_info)); + } + scanptr = cinfo->script_space; + cinfo->scan_info = scanptr; + cinfo->num_scans = nscans; + + if (ncomps == 3 && cinfo->jpeg_color_space == JCS_YCbCr) { + /* Custom script for YCbCr color images. */ + /* Initial DC scan */ + scanptr = fill_dc_scans(scanptr, ncomps, 0, 1); + /* Initial AC scan: get some luma data out in a hurry */ + scanptr = fill_a_scan(scanptr, 0, 1, 5, 0, 2); + /* Chroma data is too small to be worth expending many scans on */ + scanptr = fill_a_scan(scanptr, 2, 1, 63, 0, 1); + scanptr = fill_a_scan(scanptr, 1, 1, 63, 0, 1); + /* Complete spectral selection for luma AC */ + scanptr = fill_a_scan(scanptr, 0, 6, 63, 0, 2); + /* Refine next bit of luma AC */ + scanptr = fill_a_scan(scanptr, 0, 1, 63, 2, 1); + /* Finish DC successive approximation */ + scanptr = fill_dc_scans(scanptr, ncomps, 1, 0); + /* Finish AC successive approximation */ + scanptr = fill_a_scan(scanptr, 2, 1, 63, 1, 0); + scanptr = fill_a_scan(scanptr, 1, 1, 63, 1, 0); + /* Luma bottom bit comes last since it's usually largest scan */ + scanptr = fill_a_scan(scanptr, 0, 1, 63, 1, 0); + } else { + /* All-purpose script for other color spaces. */ + /* Successive approximation first pass */ + scanptr = fill_dc_scans(scanptr, ncomps, 0, 1); + scanptr = fill_scans(scanptr, ncomps, 1, 5, 0, 2); + scanptr = fill_scans(scanptr, ncomps, 6, 63, 0, 2); + /* Successive approximation second pass */ + scanptr = fill_scans(scanptr, ncomps, 1, 63, 2, 1); + /* Successive approximation final pass */ + scanptr = fill_dc_scans(scanptr, ncomps, 1, 0); + scanptr = fill_scans(scanptr, ncomps, 1, 63, 1, 0); + } +} + +#endif /* C_PROGRESSIVE_SUPPORTED */ + + +#ifdef C_LOSSLESS_SUPPORTED + +/* + * Enable lossless mode. + */ + +GLOBAL(void) +jpeg_enable_lossless(j_compress_ptr cinfo, int predictor_selection_value, + int point_transform) +{ + /* Safety check to ensure start_compress not called yet. */ + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + cinfo->master->lossless = TRUE; + cinfo->Ss = predictor_selection_value; + cinfo->Se = 0; + cinfo->Ah = 0; + cinfo->Al = point_transform; + + /* The JPEG spec simply gives the range 0..15 for Al (Pt), but that seems + * wrong: the upper bound ought to depend on data precision. Perhaps they + * really meant 0..N-1 for N-bit precision, which is what we allow here. + * Values greater than or equal to the data precision will result in a blank + * image. + */ + if (cinfo->Ss < 1 || cinfo->Ss > 7 || + cinfo->Al < 0 || cinfo->Al >= cinfo->data_precision) + ERREXIT4(cinfo, JERR_BAD_PROGRESSION, + cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al); +} + +#endif /* C_LOSSLESS_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jcphuff.c b/src/spice2x/external/libjpeg-turbo/jcphuff.c new file mode 100644 index 0000000..5828732 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcphuff.c @@ -0,0 +1,1102 @@ +/* + * jcphuff.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1995-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2011, 2015, 2018, 2021-2022, 2024, D. R. Commander. + * Copyright (C) 2016, 2018, 2022, Matthieu Darbois. + * Copyright (C) 2020, Arm Limited. + * Copyright (C) 2021, Alex Richardson. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains Huffman entropy encoding routines for progressive JPEG. + * + * We do not support output suspension in this module, since the library + * currently does not allow multiple-scan files to be written with output + * suspension. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#ifdef WITH_SIMD +#include "jsimd.h" +#else +#include "jchuff.h" /* Declarations shared with jc*huff.c */ +#endif +#include + +#ifdef HAVE_INTRIN_H +#include +#ifdef _MSC_VER +#ifdef HAVE_BITSCANFORWARD64 +#pragma intrinsic(_BitScanForward64) +#endif +#ifdef HAVE_BITSCANFORWARD +#pragma intrinsic(_BitScanForward) +#endif +#endif +#endif + +#ifdef C_PROGRESSIVE_SUPPORTED + +#include "jpeg_nbits.h" + + +/* Expanded entropy encoder object for progressive Huffman encoding. */ + +typedef struct { + struct jpeg_entropy_encoder pub; /* public fields */ + + /* Pointer to routine to prepare data for encode_mcu_AC_first() */ + void (*AC_first_prepare) (const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *values, size_t *zerobits); + /* Pointer to routine to prepare data for encode_mcu_AC_refine() */ + int (*AC_refine_prepare) (const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *absvalues, size_t *bits); + + /* Mode flag: TRUE for optimization, FALSE for actual data output */ + boolean gather_statistics; + + /* Bit-level coding status. + * next_output_byte/free_in_buffer are local copies of cinfo->dest fields. + */ + JOCTET *next_output_byte; /* => next byte to write in buffer */ + size_t free_in_buffer; /* # of byte spaces remaining in buffer */ + size_t put_buffer; /* current bit-accumulation buffer */ + int put_bits; /* # of bits now in it */ + j_compress_ptr cinfo; /* link to cinfo (needed for dump_buffer) */ + + /* Coding status for DC components */ + int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */ + + /* Coding status for AC components */ + int ac_tbl_no; /* the table number of the single component */ + unsigned int EOBRUN; /* run length of EOBs */ + unsigned int BE; /* # of buffered correction bits before MCU */ + char *bit_buffer; /* buffer for correction bits (1 per char) */ + /* packing correction bits tightly would save some space but cost time... */ + + unsigned int restarts_to_go; /* MCUs left in this restart interval */ + int next_restart_num; /* next restart number to write (0-7) */ + + /* Pointers to derived tables (these workspaces have image lifespan). + * Since any one scan codes only DC or only AC, we only need one set + * of tables, not one for DC and one for AC. + */ + c_derived_tbl *derived_tbls[NUM_HUFF_TBLS]; + + /* Statistics tables for optimization; again, one set is enough */ + long *count_ptrs[NUM_HUFF_TBLS]; +} phuff_entropy_encoder; + +typedef phuff_entropy_encoder *phuff_entropy_ptr; + +/* MAX_CORR_BITS is the number of bits the AC refinement correction-bit + * buffer can hold. Larger sizes may slightly improve compression, but + * 1000 is already well into the realm of overkill. + * The minimum safe size is 64 bits. + */ + +#define MAX_CORR_BITS 1000 /* Max # of correction bits I can buffer */ + +/* IRIGHT_SHIFT is like RIGHT_SHIFT, but works on int rather than JLONG. + * We assume that int right shift is unsigned if JLONG right shift is, + * which should be safe. + */ + +#ifdef RIGHT_SHIFT_IS_UNSIGNED +#define ISHIFT_TEMPS int ishift_temp; +#define IRIGHT_SHIFT(x, shft) \ + ((ishift_temp = (x)) < 0 ? \ + (ishift_temp >> (shft)) | ((~0) << (16 - (shft))) : \ + (ishift_temp >> (shft))) +#else +#define ISHIFT_TEMPS +#define IRIGHT_SHIFT(x, shft) ((x) >> (shft)) +#endif + +#define PAD(v, p) ((v + (p) - 1) & (~((p) - 1))) + +/* Forward declarations */ +METHODDEF(boolean) encode_mcu_DC_first(j_compress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(void) encode_mcu_AC_first_prepare + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *values, size_t *zerobits); +METHODDEF(boolean) encode_mcu_AC_first(j_compress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(boolean) encode_mcu_DC_refine(j_compress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(int) encode_mcu_AC_refine_prepare + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *absvalues, size_t *bits); +METHODDEF(boolean) encode_mcu_AC_refine(j_compress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(void) finish_pass_phuff(j_compress_ptr cinfo); +METHODDEF(void) finish_pass_gather_phuff(j_compress_ptr cinfo); + + +/* Count bit loop zeroes */ +INLINE +METHODDEF(int) +count_zeroes(size_t *x) +{ +#if defined(HAVE_BUILTIN_CTZL) + int result; + result = __builtin_ctzl(*x); + *x >>= result; +#elif defined(HAVE_BITSCANFORWARD64) + unsigned long result; + _BitScanForward64(&result, *x); + *x >>= result; +#elif defined(HAVE_BITSCANFORWARD) + unsigned long result; + _BitScanForward(&result, *x); + *x >>= result; +#else + int result = 0; + while ((*x & 1) == 0) { + ++result; + *x >>= 1; + } +#endif + return (int)result; +} + + +/* + * Initialize for a Huffman-compressed scan using progressive JPEG. + */ + +METHODDEF(void) +start_pass_phuff(j_compress_ptr cinfo, boolean gather_statistics) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + boolean is_DC_band; + int ci, tbl; + jpeg_component_info *compptr; + + entropy->cinfo = cinfo; + entropy->gather_statistics = gather_statistics; + + is_DC_band = (cinfo->Ss == 0); + + /* We assume jcmaster.c already validated the scan parameters. */ + + /* Select execution routines */ + if (cinfo->Ah == 0) { + if (is_DC_band) + entropy->pub.encode_mcu = encode_mcu_DC_first; + else + entropy->pub.encode_mcu = encode_mcu_AC_first; +#ifdef WITH_SIMD + if (jsimd_can_encode_mcu_AC_first_prepare()) + entropy->AC_first_prepare = jsimd_encode_mcu_AC_first_prepare; + else +#endif + entropy->AC_first_prepare = encode_mcu_AC_first_prepare; + } else { + if (is_DC_band) + entropy->pub.encode_mcu = encode_mcu_DC_refine; + else { + entropy->pub.encode_mcu = encode_mcu_AC_refine; +#ifdef WITH_SIMD + if (jsimd_can_encode_mcu_AC_refine_prepare()) + entropy->AC_refine_prepare = jsimd_encode_mcu_AC_refine_prepare; + else +#endif + entropy->AC_refine_prepare = encode_mcu_AC_refine_prepare; + /* AC refinement needs a correction bit buffer */ + if (entropy->bit_buffer == NULL) + entropy->bit_buffer = (char *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + MAX_CORR_BITS * sizeof(char)); + } + } + if (gather_statistics) + entropy->pub.finish_pass = finish_pass_gather_phuff; + else + entropy->pub.finish_pass = finish_pass_phuff; + + /* Only DC coefficients may be interleaved, so cinfo->comps_in_scan = 1 + * for AC coefficients. + */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* Initialize DC predictions to 0 */ + entropy->last_dc_val[ci] = 0; + /* Get table index */ + if (is_DC_band) { + if (cinfo->Ah != 0) /* DC refinement needs no table */ + continue; + tbl = compptr->dc_tbl_no; + } else { + entropy->ac_tbl_no = tbl = compptr->ac_tbl_no; + } + if (gather_statistics) { + /* Check for invalid table index */ + /* (make_c_derived_tbl does this in the other path) */ + if (tbl < 0 || tbl >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tbl); + /* Allocate and zero the statistics tables */ + /* Note that jpeg_gen_optimal_table expects 257 entries in each table! */ + if (entropy->count_ptrs[tbl] == NULL) + entropy->count_ptrs[tbl] = (long *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + 257 * sizeof(long)); + memset(entropy->count_ptrs[tbl], 0, 257 * sizeof(long)); + } else { + /* Compute derived values for Huffman table */ + /* We may do this more than once for a table, but it's not expensive */ + jpeg_make_c_derived_tbl(cinfo, is_DC_band, tbl, + &entropy->derived_tbls[tbl]); + } + } + + /* Initialize AC stuff */ + entropy->EOBRUN = 0; + entropy->BE = 0; + + /* Initialize bit buffer to empty */ + entropy->put_buffer = 0; + entropy->put_bits = 0; + + /* Initialize restart stuff */ + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num = 0; +} + + +/* Outputting bytes to the file. + * NB: these must be called only when actually outputting, + * that is, entropy->gather_statistics == FALSE. + */ + +/* Emit a byte */ +#define emit_byte(entropy, val) { \ + *(entropy)->next_output_byte++ = (JOCTET)(val); \ + if (--(entropy)->free_in_buffer == 0) \ + dump_buffer(entropy); \ +} + + +LOCAL(void) +dump_buffer(phuff_entropy_ptr entropy) +/* Empty the output buffer; we do not support suspension in this module. */ +{ + struct jpeg_destination_mgr *dest = entropy->cinfo->dest; + + if (!(*dest->empty_output_buffer) (entropy->cinfo)) + ERREXIT(entropy->cinfo, JERR_CANT_SUSPEND); + /* After a successful buffer dump, must reset buffer pointers */ + entropy->next_output_byte = dest->next_output_byte; + entropy->free_in_buffer = dest->free_in_buffer; +} + + +/* Outputting bits to the file */ + +/* Only the right 24 bits of put_buffer are used; the valid bits are + * left-justified in this part. At most 16 bits can be passed to emit_bits + * in one call, and we never retain more than 7 bits in put_buffer + * between calls, so 24 bits are sufficient. + */ + +LOCAL(void) +emit_bits(phuff_entropy_ptr entropy, unsigned int code, int size) +/* Emit some bits, unless we are in gather mode */ +{ + /* This routine is heavily used, so it's worth coding tightly. */ + register size_t put_buffer = (size_t)code; + register int put_bits = entropy->put_bits; + + /* if size is 0, caller used an invalid Huffman table entry */ + if (size == 0) + ERREXIT(entropy->cinfo, JERR_HUFF_MISSING_CODE); + + if (entropy->gather_statistics) + return; /* do nothing if we're only getting stats */ + + put_buffer &= (((size_t)1) << size) - 1; /* mask off any extra bits in code */ + + put_bits += size; /* new number of bits in buffer */ + + put_buffer <<= 24 - put_bits; /* align incoming bits */ + + put_buffer |= entropy->put_buffer; /* and merge with old buffer contents */ + + while (put_bits >= 8) { + int c = (int)((put_buffer >> 16) & 0xFF); + + emit_byte(entropy, c); + if (c == 0xFF) { /* need to stuff a zero byte? */ + emit_byte(entropy, 0); + } + put_buffer <<= 8; + put_bits -= 8; + } + + entropy->put_buffer = put_buffer; /* update variables */ + entropy->put_bits = put_bits; +} + + +LOCAL(void) +flush_bits(phuff_entropy_ptr entropy) +{ + emit_bits(entropy, 0x7F, 7); /* fill any partial byte with ones */ + entropy->put_buffer = 0; /* and reset bit-buffer to empty */ + entropy->put_bits = 0; +} + + +/* + * Emit (or just count) a Huffman symbol. + */ + +LOCAL(void) +emit_symbol(phuff_entropy_ptr entropy, int tbl_no, int symbol) +{ + if (entropy->gather_statistics) + entropy->count_ptrs[tbl_no][symbol]++; + else { + c_derived_tbl *tbl = entropy->derived_tbls[tbl_no]; + emit_bits(entropy, tbl->ehufco[symbol], tbl->ehufsi[symbol]); + } +} + + +/* + * Emit bits from a correction bit buffer. + */ + +LOCAL(void) +emit_buffered_bits(phuff_entropy_ptr entropy, char *bufstart, + unsigned int nbits) +{ + if (entropy->gather_statistics) + return; /* no real work */ + + while (nbits > 0) { + emit_bits(entropy, (unsigned int)(*bufstart), 1); + bufstart++; + nbits--; + } +} + + +/* + * Emit any pending EOBRUN symbol. + */ + +LOCAL(void) +emit_eobrun(phuff_entropy_ptr entropy) +{ + register int temp, nbits; + + if (entropy->EOBRUN > 0) { /* if there is any pending EOBRUN */ + temp = entropy->EOBRUN; + nbits = JPEG_NBITS_NONZERO(temp) - 1; + /* safety check: shouldn't happen given limited correction-bit buffer */ + if (nbits > 14) + ERREXIT(entropy->cinfo, JERR_HUFF_MISSING_CODE); + + emit_symbol(entropy, entropy->ac_tbl_no, nbits << 4); + if (nbits) + emit_bits(entropy, entropy->EOBRUN, nbits); + + entropy->EOBRUN = 0; + + /* Emit any buffered correction bits */ + emit_buffered_bits(entropy, entropy->bit_buffer, entropy->BE); + entropy->BE = 0; + } +} + + +/* + * Emit a restart marker & resynchronize predictions. + */ + +LOCAL(void) +emit_restart(phuff_entropy_ptr entropy, int restart_num) +{ + int ci; + + emit_eobrun(entropy); + + if (!entropy->gather_statistics) { + flush_bits(entropy); + emit_byte(entropy, 0xFF); + emit_byte(entropy, JPEG_RST0 + restart_num); + } + + if (entropy->cinfo->Ss == 0) { + /* Re-initialize DC predictions to 0 */ + for (ci = 0; ci < entropy->cinfo->comps_in_scan; ci++) + entropy->last_dc_val[ci] = 0; + } else { + /* Re-initialize all AC-related fields to 0 */ + entropy->EOBRUN = 0; + entropy->BE = 0; + } +} + + +/* + * MCU encoding for DC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +METHODDEF(boolean) +encode_mcu_DC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + register int temp, temp2, temp3; + register int nbits; + int blkn, ci; + int Al = cinfo->Al; + JBLOCKROW block; + jpeg_component_info *compptr; + ISHIFT_TEMPS + int max_coef_bits = cinfo->data_precision + 2; + + entropy->next_output_byte = cinfo->dest->next_output_byte; + entropy->free_in_buffer = cinfo->dest->free_in_buffer; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) + if (entropy->restarts_to_go == 0) + emit_restart(entropy, entropy->next_restart_num); + + /* Encode the MCU data blocks */ + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data[blkn]; + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + + /* Compute the DC value after the required point transform by Al. + * This is simply an arithmetic right shift. + */ + temp2 = IRIGHT_SHIFT((int)((*block)[0]), Al); + + /* DC differences are figured on the point-transformed values. */ + temp = temp2 - entropy->last_dc_val[ci]; + entropy->last_dc_val[ci] = temp2; + + /* Encode the DC coefficient difference per section G.1.2.1 */ + + /* This is a well-known technique for obtaining the absolute value without + * a branch. It is derived from an assembly language technique presented + * in "How to Optimize for the Pentium Processors", Copyright (c) 1996, + * 1997 by Agner Fog. + */ + temp3 = temp >> (CHAR_BIT * sizeof(int) - 1); + temp ^= temp3; + temp -= temp3; /* temp is abs value of input */ + /* For a negative input, want temp2 = bitwise complement of abs(input) */ + temp2 = temp ^ temp3; + + /* Find the number of bits needed for the magnitude of the coefficient */ + nbits = JPEG_NBITS(temp); + /* Check for out-of-range coefficient values. + * Since we're encoding a difference, the range limit is twice as much. + */ + if (nbits > max_coef_bits + 1) + ERREXIT(cinfo, JERR_BAD_DCT_COEF); + + /* Count/emit the Huffman-coded symbol for the number of bits */ + emit_symbol(entropy, compptr->dc_tbl_no, nbits); + + /* Emit that number of bits of the value, if positive, */ + /* or the complement of its magnitude, if negative. */ + if (nbits) /* emit_bits rejects calls with size 0 */ + emit_bits(entropy, (unsigned int)temp2, nbits); + } + + cinfo->dest->next_output_byte = entropy->next_output_byte; + cinfo->dest->free_in_buffer = entropy->free_in_buffer; + + /* Update restart-interval state too */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + return TRUE; +} + + +/* + * Data preparation for encode_mcu_AC_first(). + */ + +#define COMPUTE_ABSVALUES_AC_FIRST(Sl) { \ + for (k = 0; k < Sl; k++) { \ + temp = block[jpeg_natural_order_start[k]]; \ + if (temp == 0) \ + continue; \ + /* We must apply the point transform by Al. For AC coefficients this \ + * is an integer division with rounding towards 0. To do this portably \ + * in C, we shift after obtaining the absolute value; so the code is \ + * interwoven with finding the abs value (temp) and output bits (temp2). \ + */ \ + temp2 = temp >> (CHAR_BIT * sizeof(int) - 1); \ + temp ^= temp2; \ + temp -= temp2; /* temp is abs value of input */ \ + temp >>= Al; /* apply the point transform */ \ + /* Watch out for case that nonzero coef is zero after point transform */ \ + if (temp == 0) \ + continue; \ + /* For a negative coef, want temp2 = bitwise complement of abs(coef) */ \ + temp2 ^= temp; \ + values[k] = (UJCOEF)temp; \ + values[k + DCTSIZE2] = (UJCOEF)temp2; \ + zerobits |= ((size_t)1U) << k; \ + } \ +} + +METHODDEF(void) +encode_mcu_AC_first_prepare(const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *values, size_t *bits) +{ + register int k, temp, temp2; + size_t zerobits = 0U; + int Sl0 = Sl; + +#if SIZEOF_SIZE_T == 4 + if (Sl0 > 32) + Sl0 = 32; +#endif + + COMPUTE_ABSVALUES_AC_FIRST(Sl0); + + bits[0] = zerobits; +#if SIZEOF_SIZE_T == 4 + zerobits = 0U; + + if (Sl > 32) { + Sl -= 32; + jpeg_natural_order_start += 32; + values += 32; + + COMPUTE_ABSVALUES_AC_FIRST(Sl); + } + bits[1] = zerobits; +#endif +} + +/* + * MCU encoding for AC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +#define ENCODE_COEFS_AC_FIRST(label) { \ + while (zerobits) { \ + r = count_zeroes(&zerobits); \ + cvalue += r; \ +label \ + temp = cvalue[0]; \ + temp2 = cvalue[DCTSIZE2]; \ + \ + /* if run length > 15, must emit special run-length-16 codes (0xF0) */ \ + while (r > 15) { \ + emit_symbol(entropy, entropy->ac_tbl_no, 0xF0); \ + r -= 16; \ + } \ + \ + /* Find the number of bits needed for the magnitude of the coefficient */ \ + nbits = JPEG_NBITS_NONZERO(temp); /* there must be at least one 1 bit */ \ + /* Check for out-of-range coefficient values */ \ + if (nbits > max_coef_bits) \ + ERREXIT(cinfo, JERR_BAD_DCT_COEF); \ + \ + /* Count/emit Huffman symbol for run length / number of bits */ \ + emit_symbol(entropy, entropy->ac_tbl_no, (r << 4) + nbits); \ + \ + /* Emit that number of bits of the value, if positive, */ \ + /* or the complement of its magnitude, if negative. */ \ + emit_bits(entropy, (unsigned int)temp2, nbits); \ + \ + cvalue++; \ + zerobits >>= 1; \ + } \ +} + +METHODDEF(boolean) +encode_mcu_AC_first(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + register int temp, temp2; + register int nbits, r; + int Sl = cinfo->Se - cinfo->Ss + 1; + int Al = cinfo->Al; + UJCOEF values_unaligned[2 * DCTSIZE2 + 15]; + UJCOEF *values; + const UJCOEF *cvalue; + size_t zerobits; + size_t bits[8 / SIZEOF_SIZE_T]; + int max_coef_bits = cinfo->data_precision + 2; + +#ifdef ZERO_BUFFERS + memset(values_unaligned, 0, sizeof(values_unaligned)); + memset(bits, 0, sizeof(bits)); +#endif + + entropy->next_output_byte = cinfo->dest->next_output_byte; + entropy->free_in_buffer = cinfo->dest->free_in_buffer; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) + if (entropy->restarts_to_go == 0) + emit_restart(entropy, entropy->next_restart_num); + +#ifdef WITH_SIMD + cvalue = values = (UJCOEF *)PAD((JUINTPTR)values_unaligned, 16); +#else + /* Not using SIMD, so alignment is not needed */ + cvalue = values = values_unaligned; +#endif + + /* Prepare data */ + entropy->AC_first_prepare(MCU_data[0][0], jpeg_natural_order + cinfo->Ss, + Sl, Al, values, bits); + + zerobits = bits[0]; +#if SIZEOF_SIZE_T == 4 + zerobits |= bits[1]; +#endif + + /* Emit any pending EOBRUN */ + if (zerobits && (entropy->EOBRUN > 0)) + emit_eobrun(entropy); + +#if SIZEOF_SIZE_T == 4 + zerobits = bits[0]; +#endif + + /* Encode the AC coefficients per section G.1.2.2, fig. G.3 */ + + ENCODE_COEFS_AC_FIRST((void)0;); + +#if SIZEOF_SIZE_T == 4 + zerobits = bits[1]; + if (zerobits) { + int diff = ((values + DCTSIZE2 / 2) - cvalue); + r = count_zeroes(&zerobits); + r += diff; + cvalue += r; + goto first_iter_ac_first; + } + + ENCODE_COEFS_AC_FIRST(first_iter_ac_first:); +#endif + + if (cvalue < (values + Sl)) { /* If there are trailing zeroes, */ + entropy->EOBRUN++; /* count an EOB */ + if (entropy->EOBRUN == 0x7FFF) + emit_eobrun(entropy); /* force it out to avoid overflow */ + } + + cinfo->dest->next_output_byte = entropy->next_output_byte; + cinfo->dest->free_in_buffer = entropy->free_in_buffer; + + /* Update restart-interval state too */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + return TRUE; +} + + +/* + * MCU encoding for DC successive approximation refinement scan. + * Note: we assume such scans can be multi-component, although the spec + * is not very clear on the point. + */ + +METHODDEF(boolean) +encode_mcu_DC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + register int temp; + int blkn; + int Al = cinfo->Al; + JBLOCKROW block; + + entropy->next_output_byte = cinfo->dest->next_output_byte; + entropy->free_in_buffer = cinfo->dest->free_in_buffer; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) + if (entropy->restarts_to_go == 0) + emit_restart(entropy, entropy->next_restart_num); + + /* Encode the MCU data blocks */ + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data[blkn]; + + /* We simply emit the Al'th bit of the DC coefficient value. */ + temp = (*block)[0]; + emit_bits(entropy, (unsigned int)(temp >> Al), 1); + } + + cinfo->dest->next_output_byte = entropy->next_output_byte; + cinfo->dest->free_in_buffer = entropy->free_in_buffer; + + /* Update restart-interval state too */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + return TRUE; +} + + +/* + * Data preparation for encode_mcu_AC_refine(). + */ + +#define COMPUTE_ABSVALUES_AC_REFINE(Sl, koffset) { \ + /* It is convenient to make a pre-pass to determine the transformed \ + * coefficients' absolute values and the EOB position. \ + */ \ + for (k = 0; k < Sl; k++) { \ + temp = block[jpeg_natural_order_start[k]]; \ + /* We must apply the point transform by Al. For AC coefficients this \ + * is an integer division with rounding towards 0. To do this portably \ + * in C, we shift after obtaining the absolute value. \ + */ \ + temp2 = temp >> (CHAR_BIT * sizeof(int) - 1); \ + temp ^= temp2; \ + temp -= temp2; /* temp is abs value of input */ \ + temp >>= Al; /* apply the point transform */ \ + if (temp != 0) { \ + zerobits |= ((size_t)1U) << k; \ + signbits |= ((size_t)(temp2 + 1)) << k; \ + } \ + absvalues[k] = (UJCOEF)temp; /* save abs value for main pass */ \ + if (temp == 1) \ + EOB = k + koffset; /* EOB = index of last newly-nonzero coef */ \ + } \ +} + +METHODDEF(int) +encode_mcu_AC_refine_prepare(const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *absvalues, size_t *bits) +{ + register int k, temp, temp2; + int EOB = 0; + size_t zerobits = 0U, signbits = 0U; + int Sl0 = Sl; + +#if SIZEOF_SIZE_T == 4 + if (Sl0 > 32) + Sl0 = 32; +#endif + + COMPUTE_ABSVALUES_AC_REFINE(Sl0, 0); + + bits[0] = zerobits; +#if SIZEOF_SIZE_T == 8 + bits[1] = signbits; +#else + bits[2] = signbits; + + zerobits = 0U; + signbits = 0U; + + if (Sl > 32) { + Sl -= 32; + jpeg_natural_order_start += 32; + absvalues += 32; + + COMPUTE_ABSVALUES_AC_REFINE(Sl, 32); + } + + bits[1] = zerobits; + bits[3] = signbits; +#endif + + return EOB; +} + + +/* + * MCU encoding for AC successive approximation refinement scan. + */ + +#define ENCODE_COEFS_AC_REFINE(label) { \ + while (zerobits) { \ + idx = count_zeroes(&zerobits); \ + r += idx; \ + cabsvalue += idx; \ + signbits >>= idx; \ +label \ + /* Emit any required ZRLs, but not if they can be folded into EOB */ \ + while (r > 15 && (cabsvalue <= EOBPTR)) { \ + /* emit any pending EOBRUN and the BE correction bits */ \ + emit_eobrun(entropy); \ + /* Emit ZRL */ \ + emit_symbol(entropy, entropy->ac_tbl_no, 0xF0); \ + r -= 16; \ + /* Emit buffered correction bits that must be associated with ZRL */ \ + emit_buffered_bits(entropy, BR_buffer, BR); \ + BR_buffer = entropy->bit_buffer; /* BE bits are gone now */ \ + BR = 0; \ + } \ + \ + temp = *cabsvalue++; \ + \ + /* If the coef was previously nonzero, it only needs a correction bit. \ + * NOTE: a straight translation of the spec's figure G.7 would suggest \ + * that we also need to test r > 15. But if r > 15, we can only get here \ + * if k > EOB, which implies that this coefficient is not 1. \ + */ \ + if (temp > 1) { \ + /* The correction bit is the next bit of the absolute value. */ \ + BR_buffer[BR++] = (char)(temp & 1); \ + signbits >>= 1; \ + zerobits >>= 1; \ + continue; \ + } \ + \ + /* Emit any pending EOBRUN and the BE correction bits */ \ + emit_eobrun(entropy); \ + \ + /* Count/emit Huffman symbol for run length / number of bits */ \ + emit_symbol(entropy, entropy->ac_tbl_no, (r << 4) + 1); \ + \ + /* Emit output bit for newly-nonzero coef */ \ + temp = signbits & 1; /* ((*block)[jpeg_natural_order_start[k]] < 0) ? 0 : 1 */ \ + emit_bits(entropy, (unsigned int)temp, 1); \ + \ + /* Emit buffered correction bits that must be associated with this code */ \ + emit_buffered_bits(entropy, BR_buffer, BR); \ + BR_buffer = entropy->bit_buffer; /* BE bits are gone now */ \ + BR = 0; \ + r = 0; /* reset zero run length */ \ + signbits >>= 1; \ + zerobits >>= 1; \ + } \ +} + +METHODDEF(boolean) +encode_mcu_AC_refine(j_compress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + register int temp, r, idx; + char *BR_buffer; + unsigned int BR; + int Sl = cinfo->Se - cinfo->Ss + 1; + int Al = cinfo->Al; + UJCOEF absvalues_unaligned[DCTSIZE2 + 15]; + UJCOEF *absvalues; + const UJCOEF *cabsvalue, *EOBPTR; + size_t zerobits, signbits; + size_t bits[16 / SIZEOF_SIZE_T]; + +#ifdef ZERO_BUFFERS + memset(absvalues_unaligned, 0, sizeof(absvalues_unaligned)); + memset(bits, 0, sizeof(bits)); +#endif + + entropy->next_output_byte = cinfo->dest->next_output_byte; + entropy->free_in_buffer = cinfo->dest->free_in_buffer; + + /* Emit restart marker if needed */ + if (cinfo->restart_interval) + if (entropy->restarts_to_go == 0) + emit_restart(entropy, entropy->next_restart_num); + +#ifdef WITH_SIMD + cabsvalue = absvalues = (UJCOEF *)PAD((JUINTPTR)absvalues_unaligned, 16); +#else + /* Not using SIMD, so alignment is not needed */ + cabsvalue = absvalues = absvalues_unaligned; +#endif + + /* Prepare data */ + EOBPTR = absvalues + + entropy->AC_refine_prepare(MCU_data[0][0], jpeg_natural_order + cinfo->Ss, + Sl, Al, absvalues, bits); + + /* Encode the AC coefficients per section G.1.2.3, fig. G.7 */ + + r = 0; /* r = run length of zeros */ + BR = 0; /* BR = count of buffered bits added now */ + BR_buffer = entropy->bit_buffer + entropy->BE; /* Append bits to buffer */ + + zerobits = bits[0]; +#if SIZEOF_SIZE_T == 8 + signbits = bits[1]; +#else + signbits = bits[2]; +#endif + ENCODE_COEFS_AC_REFINE((void)0;); + +#if SIZEOF_SIZE_T == 4 + zerobits = bits[1]; + signbits = bits[3]; + + if (zerobits) { + int diff = ((absvalues + DCTSIZE2 / 2) - cabsvalue); + idx = count_zeroes(&zerobits); + signbits >>= idx; + idx += diff; + r += idx; + cabsvalue += idx; + goto first_iter_ac_refine; + } + + ENCODE_COEFS_AC_REFINE(first_iter_ac_refine:); +#endif + + r |= (int)((absvalues + Sl) - cabsvalue); + + if (r > 0 || BR > 0) { /* If there are trailing zeroes, */ + entropy->EOBRUN++; /* count an EOB */ + entropy->BE += BR; /* concat my correction bits to older ones */ + /* We force out the EOB if we risk either: + * 1. overflow of the EOB counter; + * 2. overflow of the correction bit buffer during the next MCU. + */ + if (entropy->EOBRUN == 0x7FFF || + entropy->BE > (MAX_CORR_BITS - DCTSIZE2 + 1)) + emit_eobrun(entropy); + } + + cinfo->dest->next_output_byte = entropy->next_output_byte; + cinfo->dest->free_in_buffer = entropy->free_in_buffer; + + /* Update restart-interval state too */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) { + entropy->restarts_to_go = cinfo->restart_interval; + entropy->next_restart_num++; + entropy->next_restart_num &= 7; + } + entropy->restarts_to_go--; + } + + return TRUE; +} + + +/* + * Finish up at the end of a Huffman-compressed progressive scan. + */ + +METHODDEF(void) +finish_pass_phuff(j_compress_ptr cinfo) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + + entropy->next_output_byte = cinfo->dest->next_output_byte; + entropy->free_in_buffer = cinfo->dest->free_in_buffer; + + /* Flush out any buffered data */ + emit_eobrun(entropy); + flush_bits(entropy); + + cinfo->dest->next_output_byte = entropy->next_output_byte; + cinfo->dest->free_in_buffer = entropy->free_in_buffer; +} + + +/* + * Finish up a statistics-gathering pass and create the new Huffman tables. + */ + +METHODDEF(void) +finish_pass_gather_phuff(j_compress_ptr cinfo) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + boolean is_DC_band; + int ci, tbl; + jpeg_component_info *compptr; + JHUFF_TBL **htblptr; + boolean did[NUM_HUFF_TBLS]; + + /* Flush out buffered data (all we care about is counting the EOB symbol) */ + emit_eobrun(entropy); + + is_DC_band = (cinfo->Ss == 0); + + /* It's important not to apply jpeg_gen_optimal_table more than once + * per table, because it clobbers the input frequency counts! + */ + memset(did, 0, sizeof(did)); + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + if (is_DC_band) { + if (cinfo->Ah != 0) /* DC refinement needs no table */ + continue; + tbl = compptr->dc_tbl_no; + } else { + tbl = compptr->ac_tbl_no; + } + if (!did[tbl]) { + if (is_DC_band) + htblptr = &cinfo->dc_huff_tbl_ptrs[tbl]; + else + htblptr = &cinfo->ac_huff_tbl_ptrs[tbl]; + if (*htblptr == NULL) + *htblptr = jpeg_alloc_huff_table((j_common_ptr)cinfo); + jpeg_gen_optimal_table(cinfo, *htblptr, entropy->count_ptrs[tbl]); + did[tbl] = TRUE; + } + } +} + + +/* + * Module initialization routine for progressive Huffman entropy encoding. + */ + +GLOBAL(void) +jinit_phuff_encoder(j_compress_ptr cinfo) +{ + phuff_entropy_ptr entropy; + int i; + + entropy = (phuff_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(phuff_entropy_encoder)); + cinfo->entropy = (struct jpeg_entropy_encoder *)entropy; + entropy->pub.start_pass = start_pass_phuff; + + /* Mark tables unallocated */ + for (i = 0; i < NUM_HUFF_TBLS; i++) { + entropy->derived_tbls[i] = NULL; + entropy->count_ptrs[i] = NULL; + } + entropy->bit_buffer = NULL; /* needed only in AC refinement scan */ +} + +#endif /* C_PROGRESSIVE_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jcprepct.c b/src/spice2x/external/libjpeg-turbo/jcprepct.c new file mode 100644 index 0000000..ac2311c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcprepct.c @@ -0,0 +1,365 @@ +/* + * jcprepct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the compression preprocessing controller. + * This controller manages the color conversion, downsampling, + * and edge expansion steps. + * + * Most of the complexity here is associated with buffering input rows + * as required by the downsampler. See the comments at the head of + * jcsample.c for the downsampler's needs. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) + +/* At present, jcsample.c can request context rows only for smoothing. + * In the future, we might also need context rows for CCIR601 sampling + * or other more-complex downsampling procedures. The code to support + * context rows should be compiled only if needed. + */ +#ifdef INPUT_SMOOTHING_SUPPORTED +#define CONTEXT_ROWS_SUPPORTED +#endif + + +/* + * For the simple (no-context-row) case, we just need to buffer one + * row group's worth of pixels for the downsampling step. At the bottom of + * the image, we pad to a full row group by replicating the last pixel row. + * The downsampler's last output row is then replicated if needed to pad + * out to a full iMCU row. + * + * When providing context rows, we must buffer three row groups' worth of + * pixels. Three row groups are physically allocated, but the row pointer + * arrays are made five row groups high, with the extra pointers above and + * below "wrapping around" to point to the last and first real row groups. + * This allows the downsampler to access the proper context rows. + * At the top and bottom of the image, we create dummy context rows by + * copying the first or last real pixel row. This copying could be avoided + * by pointer hacking as is done in jdmainct.c, but it doesn't seem worth the + * trouble on the compression side. + */ + + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_c_prep_controller pub; /* public fields */ + + /* Downsampling input buffer. This buffer holds color-converted data + * until we have enough to do a downsample step. + */ + _JSAMPARRAY color_buf[MAX_COMPONENTS]; + + JDIMENSION rows_to_go; /* counts rows remaining in source image */ + int next_buf_row; /* index of next row to store in color_buf */ + +#ifdef CONTEXT_ROWS_SUPPORTED /* only needed for context case */ + int this_row_group; /* starting row index of group to process */ + int next_buf_stop; /* downsample when we reach this index */ +#endif +} my_prep_controller; + +typedef my_prep_controller *my_prep_ptr; + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_prep(j_compress_ptr cinfo, J_BUF_MODE pass_mode) +{ + my_prep_ptr prep = (my_prep_ptr)cinfo->prep; + + if (pass_mode != JBUF_PASS_THRU) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + + /* Initialize total-height counter for detecting bottom of image */ + prep->rows_to_go = cinfo->image_height; + /* Mark the conversion buffer empty */ + prep->next_buf_row = 0; +#ifdef CONTEXT_ROWS_SUPPORTED + /* Preset additional state variables for context mode. + * These aren't used in non-context mode, so we needn't test which mode. + */ + prep->this_row_group = 0; + /* Set next_buf_stop to stop after two row groups have been read in. */ + prep->next_buf_stop = 2 * cinfo->max_v_samp_factor; +#endif +} + + +/* + * Expand an image vertically from height input_rows to height output_rows, + * by duplicating the bottom row. + */ + +LOCAL(void) +expand_bottom_edge(_JSAMPARRAY image_data, JDIMENSION num_cols, int input_rows, + int output_rows) +{ + register int row; + + for (row = input_rows; row < output_rows; row++) { + _jcopy_sample_rows(image_data, input_rows - 1, image_data, row, 1, + num_cols); + } +} + + +/* + * Process some data in the simple no-context case. + * + * Preprocessor output data is counted in "row groups". A row group + * is defined to be v_samp_factor sample rows of each component. + * Downsampling will produce this much data from each max_v_samp_factor + * input rows. + */ + +METHODDEF(void) +pre_process_data(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail, + _JSAMPIMAGE output_buf, JDIMENSION *out_row_group_ctr, + JDIMENSION out_row_groups_avail) +{ + my_prep_ptr prep = (my_prep_ptr)cinfo->prep; + int numrows, ci; + JDIMENSION inrows; + jpeg_component_info *compptr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + while (*in_row_ctr < in_rows_avail && + *out_row_group_ctr < out_row_groups_avail) { + /* Do color conversion to fill the conversion buffer. */ + inrows = in_rows_avail - *in_row_ctr; + numrows = cinfo->max_v_samp_factor - prep->next_buf_row; + numrows = (int)MIN((JDIMENSION)numrows, inrows); + (*cinfo->cconvert->_color_convert) (cinfo, input_buf + *in_row_ctr, + prep->color_buf, + (JDIMENSION)prep->next_buf_row, + numrows); + *in_row_ctr += numrows; + prep->next_buf_row += numrows; + prep->rows_to_go -= numrows; + /* If at bottom of image, pad to fill the conversion buffer. */ + if (prep->rows_to_go == 0 && + prep->next_buf_row < cinfo->max_v_samp_factor) { + for (ci = 0; ci < cinfo->num_components; ci++) { + expand_bottom_edge(prep->color_buf[ci], cinfo->image_width, + prep->next_buf_row, cinfo->max_v_samp_factor); + } + prep->next_buf_row = cinfo->max_v_samp_factor; + } + /* If we've filled the conversion buffer, empty it. */ + if (prep->next_buf_row == cinfo->max_v_samp_factor) { + (*cinfo->downsample->_downsample) (cinfo, + prep->color_buf, (JDIMENSION)0, + output_buf, *out_row_group_ctr); + prep->next_buf_row = 0; + (*out_row_group_ctr)++; + } + /* If at bottom of image, pad the output to a full iMCU height. + * Note we assume the caller is providing a one-iMCU-height output buffer! + */ + if (prep->rows_to_go == 0 && *out_row_group_ctr < out_row_groups_avail) { + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + expand_bottom_edge(output_buf[ci], + compptr->width_in_blocks * data_unit, + (int)(*out_row_group_ctr * compptr->v_samp_factor), + (int)(out_row_groups_avail * compptr->v_samp_factor)); + } + *out_row_group_ctr = out_row_groups_avail; + break; /* can exit outer loop without test */ + } + } +} + + +#ifdef CONTEXT_ROWS_SUPPORTED + +/* + * Process some data in the context case. + */ + +METHODDEF(void) +pre_process_context(j_compress_ptr cinfo, _JSAMPARRAY input_buf, + JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail, + _JSAMPIMAGE output_buf, JDIMENSION *out_row_group_ctr, + JDIMENSION out_row_groups_avail) +{ + my_prep_ptr prep = (my_prep_ptr)cinfo->prep; + int numrows, ci; + int buf_height = cinfo->max_v_samp_factor * 3; + JDIMENSION inrows; + + while (*out_row_group_ctr < out_row_groups_avail) { + if (*in_row_ctr < in_rows_avail) { + /* Do color conversion to fill the conversion buffer. */ + inrows = in_rows_avail - *in_row_ctr; + numrows = prep->next_buf_stop - prep->next_buf_row; + numrows = (int)MIN((JDIMENSION)numrows, inrows); + (*cinfo->cconvert->_color_convert) (cinfo, input_buf + *in_row_ctr, + prep->color_buf, + (JDIMENSION)prep->next_buf_row, + numrows); + /* Pad at top of image, if first time through */ + if (prep->rows_to_go == cinfo->image_height) { + for (ci = 0; ci < cinfo->num_components; ci++) { + int row; + for (row = 1; row <= cinfo->max_v_samp_factor; row++) { + _jcopy_sample_rows(prep->color_buf[ci], 0, prep->color_buf[ci], + -row, 1, cinfo->image_width); + } + } + } + *in_row_ctr += numrows; + prep->next_buf_row += numrows; + prep->rows_to_go -= numrows; + } else { + /* Return for more data, unless we are at the bottom of the image. */ + if (prep->rows_to_go != 0) + break; + /* When at bottom of image, pad to fill the conversion buffer. */ + if (prep->next_buf_row < prep->next_buf_stop) { + for (ci = 0; ci < cinfo->num_components; ci++) { + expand_bottom_edge(prep->color_buf[ci], cinfo->image_width, + prep->next_buf_row, prep->next_buf_stop); + } + prep->next_buf_row = prep->next_buf_stop; + } + } + /* If we've gotten enough data, downsample a row group. */ + if (prep->next_buf_row == prep->next_buf_stop) { + (*cinfo->downsample->_downsample) (cinfo, prep->color_buf, + (JDIMENSION)prep->this_row_group, + output_buf, *out_row_group_ctr); + (*out_row_group_ctr)++; + /* Advance pointers with wraparound as necessary. */ + prep->this_row_group += cinfo->max_v_samp_factor; + if (prep->this_row_group >= buf_height) + prep->this_row_group = 0; + if (prep->next_buf_row >= buf_height) + prep->next_buf_row = 0; + prep->next_buf_stop = prep->next_buf_row + cinfo->max_v_samp_factor; + } + } +} + + +/* + * Create the wrapped-around downsampling input buffer needed for context mode. + */ + +LOCAL(void) +create_context_buffer(j_compress_ptr cinfo) +{ + my_prep_ptr prep = (my_prep_ptr)cinfo->prep; + int rgroup_height = cinfo->max_v_samp_factor; + int ci, i; + jpeg_component_info *compptr; + _JSAMPARRAY true_buffer, fake_buffer; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + /* Grab enough space for fake row pointers for all the components; + * we need five row groups' worth of pointers for each component. + */ + fake_buffer = (_JSAMPARRAY) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (cinfo->num_components * 5 * rgroup_height) * + sizeof(_JSAMPROW)); + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Allocate the actual buffer space (3 row groups) for this component. + * We make the buffer wide enough to allow the downsampler to edge-expand + * horizontally within the buffer, if it so chooses. + */ + true_buffer = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (JDIMENSION)(((long)compptr->width_in_blocks * data_unit * + cinfo->max_h_samp_factor) / compptr->h_samp_factor), + (JDIMENSION)(3 * rgroup_height)); + /* Copy true buffer row pointers into the middle of the fake row array */ + memcpy(fake_buffer + rgroup_height, true_buffer, + 3 * rgroup_height * sizeof(_JSAMPROW)); + /* Fill in the above and below wraparound pointers */ + for (i = 0; i < rgroup_height; i++) { + fake_buffer[i] = true_buffer[2 * rgroup_height + i]; + fake_buffer[4 * rgroup_height + i] = true_buffer[i]; + } + prep->color_buf[ci] = fake_buffer + rgroup_height; + fake_buffer += 5 * rgroup_height; /* point to space for next component */ + } +} + +#endif /* CONTEXT_ROWS_SUPPORTED */ + + +/* + * Initialize preprocessing controller. + */ + +GLOBAL(void) +_jinit_c_prep_controller(j_compress_ptr cinfo, boolean need_full_buffer) +{ + my_prep_ptr prep; + int ci; + jpeg_component_info *compptr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (need_full_buffer) /* safety check */ + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + + prep = (my_prep_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_prep_controller)); + cinfo->prep = (struct jpeg_c_prep_controller *)prep; + prep->pub.start_pass = start_pass_prep; + + /* Allocate the color conversion buffer. + * We make the buffer wide enough to allow the downsampler to edge-expand + * horizontally within the buffer, if it so chooses. + */ + if (cinfo->downsample->need_context_rows) { + /* Set up to provide context rows */ +#ifdef CONTEXT_ROWS_SUPPORTED + prep->pub._pre_process_data = pre_process_context; + create_context_buffer(cinfo); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + /* No context, just make it tall enough for one row group */ + prep->pub._pre_process_data = pre_process_data; + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + prep->color_buf[ci] = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (JDIMENSION)(((long)compptr->width_in_blocks * data_unit * + cinfo->max_h_samp_factor) / compptr->h_samp_factor), + (JDIMENSION)cinfo->max_v_samp_factor); + } + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jcsample.c b/src/spice2x/external/libjpeg-turbo/jcsample.c new file mode 100644 index 0000000..30e6e54 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jcsample.c @@ -0,0 +1,543 @@ +/* + * jcsample.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2014, MIPS Technologies, Inc., California. + * Copyright (C) 2015, 2019, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains downsampling routines. + * + * Downsampling input data is counted in "row groups". A row group + * is defined to be max_v_samp_factor pixel rows of each component, + * from which the downsampler produces v_samp_factor sample rows. + * A single row group is processed in each call to the downsampler module. + * + * The downsampler is responsible for edge-expansion of its output data + * to fill an integral number of DCT blocks horizontally. The source buffer + * may be modified if it is helpful for this purpose (the source buffer is + * allocated wide enough to correspond to the desired output width). + * The caller (the prep controller) is responsible for vertical padding. + * + * The downsampler may request "context rows" by setting need_context_rows + * during startup. In this case, the input arrays will contain at least + * one row group's worth of pixels above and below the passed-in data; + * the caller will create dummy rows at image top and bottom by replicating + * the first or last real pixel row. + * + * An excellent reference for image resampling is + * Digital Image Warping, George Wolberg, 1990. + * Pub. by IEEE Computer Society Press, Los Alamitos, CA. ISBN 0-8186-8944-7. + * + * The downsampling algorithm used here is a simple average of the source + * pixels covered by the output pixel. The hi-falutin sampling literature + * refers to this as a "box filter". In general the characteristics of a box + * filter are not very good, but for the specific cases we normally use (1:1 + * and 2:1 ratios) the box is equivalent to a "triangle filter" which is not + * nearly so bad. If you intend to use other sampling ratios, you'd be well + * advised to improve this code. + * + * A simple input-smoothing capability is provided. This is mainly intended + * for cleaning up color-dithered GIF input files (if you find it inadequate, + * we suggest using an external filtering program such as pnmconvol). When + * enabled, each input pixel P is replaced by a weighted sum of itself and its + * eight neighbors. P's weight is 1-8*SF and each neighbor's weight is SF, + * where SF = (smoothing_factor / 1024). + * Currently, smoothing is only supported for 2h2v sampling factors. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsimd.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) + +/* Pointer to routine to downsample a single component */ +typedef void (*downsample1_ptr) (j_compress_ptr cinfo, + jpeg_component_info *compptr, + _JSAMPARRAY input_data, + _JSAMPARRAY output_data); + +/* Private subobject */ + +typedef struct { + struct jpeg_downsampler pub; /* public fields */ + + /* Downsampling method pointers, one per component */ + downsample1_ptr methods[MAX_COMPONENTS]; +} my_downsampler; + +typedef my_downsampler *my_downsample_ptr; + + +/* + * Initialize for a downsampling pass. + */ + +METHODDEF(void) +start_pass_downsample(j_compress_ptr cinfo) +{ + /* no work for now */ +} + + +/* + * Expand a component horizontally from width input_cols to width output_cols, + * by duplicating the rightmost samples. + */ + +LOCAL(void) +expand_right_edge(_JSAMPARRAY image_data, int num_rows, JDIMENSION input_cols, + JDIMENSION output_cols) +{ + register _JSAMPROW ptr; + register _JSAMPLE pixval; + register int count; + int row; + int numcols = (int)(output_cols - input_cols); + + if (numcols > 0) { + for (row = 0; row < num_rows; row++) { + ptr = image_data[row] + input_cols; + pixval = ptr[-1]; + for (count = numcols; count > 0; count--) + *ptr++ = pixval; + } + } +} + + +/* + * Do downsampling for a whole row group (all components). + * + * In this version we simply downsample each component independently. + */ + +METHODDEF(void) +sep_downsample(j_compress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_index, _JSAMPIMAGE output_buf, + JDIMENSION out_row_group_index) +{ + my_downsample_ptr downsample = (my_downsample_ptr)cinfo->downsample; + int ci; + jpeg_component_info *compptr; + _JSAMPARRAY in_ptr, out_ptr; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + in_ptr = input_buf[ci] + in_row_index; + out_ptr = output_buf[ci] + (out_row_group_index * compptr->v_samp_factor); + (*downsample->methods[ci]) (cinfo, compptr, in_ptr, out_ptr); + } +} + + +/* + * Downsample pixel values of a single component. + * One row group is processed per call. + * This version handles arbitrary integral sampling ratios, without smoothing. + * Note that this version is not actually used for customary sampling ratios. + */ + +METHODDEF(void) +int_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY output_data) +{ + int inrow, outrow, h_expand, v_expand, numpix, numpix2, h, v; + JDIMENSION outcol, outcol_h; /* outcol_h == outcol*h_expand */ + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + JDIMENSION output_cols = compptr->width_in_blocks * data_unit; + _JSAMPROW inptr, outptr; + JLONG outvalue; + + h_expand = cinfo->max_h_samp_factor / compptr->h_samp_factor; + v_expand = cinfo->max_v_samp_factor / compptr->v_samp_factor; + numpix = h_expand * v_expand; + numpix2 = numpix / 2; + + /* Expand input data enough to let all the output samples be generated + * by the standard loop. Special-casing padded output would be more + * efficient. + */ + expand_right_edge(input_data, cinfo->max_v_samp_factor, cinfo->image_width, + output_cols * h_expand); + + inrow = 0; + for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) { + outptr = output_data[outrow]; + for (outcol = 0, outcol_h = 0; outcol < output_cols; + outcol++, outcol_h += h_expand) { + outvalue = 0; + for (v = 0; v < v_expand; v++) { + inptr = input_data[inrow + v] + outcol_h; + for (h = 0; h < h_expand; h++) { + outvalue += (JLONG)(*inptr++); + } + } + *outptr++ = (_JSAMPLE)((outvalue + numpix2) / numpix); + } + inrow += v_expand; + } +} + + +/* + * Downsample pixel values of a single component. + * This version handles the special case of a full-size component, + * without smoothing. + */ + +METHODDEF(void) +fullsize_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY output_data) +{ + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + /* Copy the data */ + _jcopy_sample_rows(input_data, 0, output_data, 0, cinfo->max_v_samp_factor, + cinfo->image_width); + /* Edge-expand */ + expand_right_edge(output_data, cinfo->max_v_samp_factor, cinfo->image_width, + compptr->width_in_blocks * data_unit); +} + + +/* + * Downsample pixel values of a single component. + * This version handles the common case of 2:1 horizontal and 1:1 vertical, + * without smoothing. + * + * A note about the "bias" calculations: when rounding fractional values to + * integer, we do not want to always round 0.5 up to the next integer. + * If we did that, we'd introduce a noticeable bias towards larger values. + * Instead, this code is arranged so that 0.5 will be rounded up or down at + * alternate pixel locations (a simple ordered dither pattern). + */ + +METHODDEF(void) +h2v1_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY output_data) +{ + int outrow; + JDIMENSION outcol; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + JDIMENSION output_cols = compptr->width_in_blocks * data_unit; + register _JSAMPROW inptr, outptr; + register int bias; + + /* Expand input data enough to let all the output samples be generated + * by the standard loop. Special-casing padded output would be more + * efficient. + */ + expand_right_edge(input_data, cinfo->max_v_samp_factor, cinfo->image_width, + output_cols * 2); + + for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) { + outptr = output_data[outrow]; + inptr = input_data[outrow]; + bias = 0; /* bias = 0,1,0,1,... for successive samples */ + for (outcol = 0; outcol < output_cols; outcol++) { + *outptr++ = (_JSAMPLE)((inptr[0] + inptr[1] + bias) >> 1); + bias ^= 1; /* 0=>1, 1=>0 */ + inptr += 2; + } + } +} + + +/* + * Downsample pixel values of a single component. + * This version handles the standard case of 2:1 horizontal and 2:1 vertical, + * without smoothing. + */ + +METHODDEF(void) +h2v2_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY output_data) +{ + int inrow, outrow; + JDIMENSION outcol; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + JDIMENSION output_cols = compptr->width_in_blocks * data_unit; + register _JSAMPROW inptr0, inptr1, outptr; + register int bias; + + /* Expand input data enough to let all the output samples be generated + * by the standard loop. Special-casing padded output would be more + * efficient. + */ + expand_right_edge(input_data, cinfo->max_v_samp_factor, cinfo->image_width, + output_cols * 2); + + inrow = 0; + for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) { + outptr = output_data[outrow]; + inptr0 = input_data[inrow]; + inptr1 = input_data[inrow + 1]; + bias = 1; /* bias = 1,2,1,2,... for successive samples */ + for (outcol = 0; outcol < output_cols; outcol++) { + *outptr++ = (_JSAMPLE) + ((inptr0[0] + inptr0[1] + inptr1[0] + inptr1[1] + bias) >> 2); + bias ^= 3; /* 1=>2, 2=>1 */ + inptr0 += 2; inptr1 += 2; + } + inrow += 2; + } +} + + +#ifdef INPUT_SMOOTHING_SUPPORTED + +/* + * Downsample pixel values of a single component. + * This version handles the standard case of 2:1 horizontal and 2:1 vertical, + * with smoothing. One row of context is required. + */ + +METHODDEF(void) +h2v2_smooth_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY output_data) +{ + int inrow, outrow; + JDIMENSION colctr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + JDIMENSION output_cols = compptr->width_in_blocks * data_unit; + register _JSAMPROW inptr0, inptr1, above_ptr, below_ptr, outptr; + JLONG membersum, neighsum, memberscale, neighscale; + + /* Expand input data enough to let all the output samples be generated + * by the standard loop. Special-casing padded output would be more + * efficient. + */ + expand_right_edge(input_data - 1, cinfo->max_v_samp_factor + 2, + cinfo->image_width, output_cols * 2); + + /* We don't bother to form the individual "smoothed" input pixel values; + * we can directly compute the output which is the average of the four + * smoothed values. Each of the four member pixels contributes a fraction + * (1-8*SF) to its own smoothed image and a fraction SF to each of the three + * other smoothed pixels, therefore a total fraction (1-5*SF)/4 to the final + * output. The four corner-adjacent neighbor pixels contribute a fraction + * SF to just one smoothed pixel, or SF/4 to the final output; while the + * eight edge-adjacent neighbors contribute SF to each of two smoothed + * pixels, or SF/2 overall. In order to use integer arithmetic, these + * factors are scaled by 2^16 = 65536. + * Also recall that SF = smoothing_factor / 1024. + */ + + memberscale = 16384 - cinfo->smoothing_factor * 80; /* scaled (1-5*SF)/4 */ + neighscale = cinfo->smoothing_factor * 16; /* scaled SF/4 */ + + inrow = 0; + for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) { + outptr = output_data[outrow]; + inptr0 = input_data[inrow]; + inptr1 = input_data[inrow + 1]; + above_ptr = input_data[inrow - 1]; + below_ptr = input_data[inrow + 2]; + + /* Special case for first column: pretend column -1 is same as column 0 */ + membersum = inptr0[0] + inptr0[1] + inptr1[0] + inptr1[1]; + neighsum = above_ptr[0] + above_ptr[1] + below_ptr[0] + below_ptr[1] + + inptr0[0] + inptr0[2] + inptr1[0] + inptr1[2]; + neighsum += neighsum; + neighsum += above_ptr[0] + above_ptr[2] + below_ptr[0] + below_ptr[2]; + membersum = membersum * memberscale + neighsum * neighscale; + *outptr++ = (_JSAMPLE)((membersum + 32768) >> 16); + inptr0 += 2; inptr1 += 2; above_ptr += 2; below_ptr += 2; + + for (colctr = output_cols - 2; colctr > 0; colctr--) { + /* sum of pixels directly mapped to this output element */ + membersum = inptr0[0] + inptr0[1] + inptr1[0] + inptr1[1]; + /* sum of edge-neighbor pixels */ + neighsum = above_ptr[0] + above_ptr[1] + below_ptr[0] + below_ptr[1] + + inptr0[-1] + inptr0[2] + inptr1[-1] + inptr1[2]; + /* The edge-neighbors count twice as much as corner-neighbors */ + neighsum += neighsum; + /* Add in the corner-neighbors */ + neighsum += above_ptr[-1] + above_ptr[2] + below_ptr[-1] + below_ptr[2]; + /* form final output scaled up by 2^16 */ + membersum = membersum * memberscale + neighsum * neighscale; + /* round, descale and output it */ + *outptr++ = (_JSAMPLE)((membersum + 32768) >> 16); + inptr0 += 2; inptr1 += 2; above_ptr += 2; below_ptr += 2; + } + + /* Special case for last column */ + membersum = inptr0[0] + inptr0[1] + inptr1[0] + inptr1[1]; + neighsum = above_ptr[0] + above_ptr[1] + below_ptr[0] + below_ptr[1] + + inptr0[-1] + inptr0[1] + inptr1[-1] + inptr1[1]; + neighsum += neighsum; + neighsum += above_ptr[-1] + above_ptr[1] + below_ptr[-1] + below_ptr[1]; + membersum = membersum * memberscale + neighsum * neighscale; + *outptr = (_JSAMPLE)((membersum + 32768) >> 16); + + inrow += 2; + } +} + + +/* + * Downsample pixel values of a single component. + * This version handles the special case of a full-size component, + * with smoothing. One row of context is required. + */ + +METHODDEF(void) +fullsize_smooth_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY output_data) +{ + int outrow; + JDIMENSION colctr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + JDIMENSION output_cols = compptr->width_in_blocks * data_unit; + register _JSAMPROW inptr, above_ptr, below_ptr, outptr; + JLONG membersum, neighsum, memberscale, neighscale; + int colsum, lastcolsum, nextcolsum; + + /* Expand input data enough to let all the output samples be generated + * by the standard loop. Special-casing padded output would be more + * efficient. + */ + expand_right_edge(input_data - 1, cinfo->max_v_samp_factor + 2, + cinfo->image_width, output_cols); + + /* Each of the eight neighbor pixels contributes a fraction SF to the + * smoothed pixel, while the main pixel contributes (1-8*SF). In order + * to use integer arithmetic, these factors are multiplied by 2^16 = 65536. + * Also recall that SF = smoothing_factor / 1024. + */ + + memberscale = 65536L - cinfo->smoothing_factor * 512L; /* scaled 1-8*SF */ + neighscale = cinfo->smoothing_factor * 64; /* scaled SF */ + + for (outrow = 0; outrow < compptr->v_samp_factor; outrow++) { + outptr = output_data[outrow]; + inptr = input_data[outrow]; + above_ptr = input_data[outrow - 1]; + below_ptr = input_data[outrow + 1]; + + /* Special case for first column */ + colsum = (*above_ptr++) + (*below_ptr++) + inptr[0]; + membersum = *inptr++; + nextcolsum = above_ptr[0] + below_ptr[0] + inptr[0]; + neighsum = colsum + (colsum - membersum) + nextcolsum; + membersum = membersum * memberscale + neighsum * neighscale; + *outptr++ = (_JSAMPLE)((membersum + 32768) >> 16); + lastcolsum = colsum; colsum = nextcolsum; + + for (colctr = output_cols - 2; colctr > 0; colctr--) { + membersum = *inptr++; + above_ptr++; below_ptr++; + nextcolsum = above_ptr[0] + below_ptr[0] + inptr[0]; + neighsum = lastcolsum + (colsum - membersum) + nextcolsum; + membersum = membersum * memberscale + neighsum * neighscale; + *outptr++ = (_JSAMPLE)((membersum + 32768) >> 16); + lastcolsum = colsum; colsum = nextcolsum; + } + + /* Special case for last column */ + membersum = *inptr; + neighsum = lastcolsum + (colsum - membersum) + colsum; + membersum = membersum * memberscale + neighsum * neighscale; + *outptr = (_JSAMPLE)((membersum + 32768) >> 16); + + } +} + +#endif /* INPUT_SMOOTHING_SUPPORTED */ + + +/* + * Module initialization routine for downsampling. + * Note that we must select a routine for each component. + */ + +GLOBAL(void) +_jinit_downsampler(j_compress_ptr cinfo) +{ + my_downsample_ptr downsample; + int ci; + jpeg_component_info *compptr; + boolean smoothok = TRUE; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + downsample = (my_downsample_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_downsampler)); + cinfo->downsample = (struct jpeg_downsampler *)downsample; + downsample->pub.start_pass = start_pass_downsample; + downsample->pub._downsample = sep_downsample; + downsample->pub.need_context_rows = FALSE; + + if (cinfo->CCIR601_sampling) + ERREXIT(cinfo, JERR_CCIR601_NOTIMPL); + + /* Verify we can handle the sampling factors, and set up method pointers */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + if (compptr->h_samp_factor == cinfo->max_h_samp_factor && + compptr->v_samp_factor == cinfo->max_v_samp_factor) { +#ifdef INPUT_SMOOTHING_SUPPORTED + if (cinfo->smoothing_factor) { + downsample->methods[ci] = fullsize_smooth_downsample; + downsample->pub.need_context_rows = TRUE; + } else +#endif + downsample->methods[ci] = fullsize_downsample; + } else if (compptr->h_samp_factor * 2 == cinfo->max_h_samp_factor && + compptr->v_samp_factor == cinfo->max_v_samp_factor) { + smoothok = FALSE; +#ifdef WITH_SIMD + if (jsimd_can_h2v1_downsample()) + downsample->methods[ci] = jsimd_h2v1_downsample; + else +#endif + downsample->methods[ci] = h2v1_downsample; + } else if (compptr->h_samp_factor * 2 == cinfo->max_h_samp_factor && + compptr->v_samp_factor * 2 == cinfo->max_v_samp_factor) { +#ifdef INPUT_SMOOTHING_SUPPORTED + if (cinfo->smoothing_factor) { +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_can_h2v2_smooth_downsample()) + downsample->methods[ci] = jsimd_h2v2_smooth_downsample; + else +#endif + downsample->methods[ci] = h2v2_smooth_downsample; + downsample->pub.need_context_rows = TRUE; + } else +#endif + { +#ifdef WITH_SIMD + if (jsimd_can_h2v2_downsample()) + downsample->methods[ci] = jsimd_h2v2_downsample; + else +#endif + downsample->methods[ci] = h2v2_downsample; + } + } else if ((cinfo->max_h_samp_factor % compptr->h_samp_factor) == 0 && + (cinfo->max_v_samp_factor % compptr->v_samp_factor) == 0) { + smoothok = FALSE; + downsample->methods[ci] = int_downsample; + } else + ERREXIT(cinfo, JERR_FRACT_SAMPLE_NOTIMPL); + } + +#ifdef INPUT_SMOOTHING_SUPPORTED + if (cinfo->smoothing_factor && !smoothok) + TRACEMS(cinfo, 0, JTRC_SMOOTH_NOTIMPL); +#endif +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(C_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jctrans.c b/src/spice2x/external/libjpeg-turbo/jctrans.c new file mode 100644 index 0000000..ae52e39 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jctrans.c @@ -0,0 +1,415 @@ +/* + * jctrans.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1995-1998, Thomas G. Lane. + * Modified 2000-2009 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains library routines for transcoding compression, + * that is, writing raw DCT coefficient arrays to an output JPEG file. + * The routines in jcapimin.c will also be needed by a transcoder. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jpegapicomp.h" + + +/* Forward declarations */ +LOCAL(void) transencode_master_selection(j_compress_ptr cinfo, + jvirt_barray_ptr *coef_arrays); +LOCAL(void) transencode_coef_controller(j_compress_ptr cinfo, + jvirt_barray_ptr *coef_arrays); + + +/* + * Compression initialization for writing raw-coefficient data. + * Before calling this, all parameters and a data destination must be set up. + * Call jpeg_finish_compress() to actually write the data. + * + * The number of passed virtual arrays must match cinfo->num_components. + * Note that the virtual arrays need not be filled or even realized at + * the time write_coefficients is called; indeed, if the virtual arrays + * were requested from this compression object's memory manager, they + * typically will be realized during this routine and filled afterwards. + */ + +GLOBAL(void) +jpeg_write_coefficients(j_compress_ptr cinfo, jvirt_barray_ptr *coef_arrays) +{ + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + if (cinfo->global_state != CSTATE_START) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + /* Mark all tables to be written */ + jpeg_suppress_tables(cinfo, FALSE); + /* (Re)initialize error mgr and destination modules */ + (*cinfo->err->reset_error_mgr) ((j_common_ptr)cinfo); + (*cinfo->dest->init_destination) (cinfo); + /* Perform master selection of active modules */ + transencode_master_selection(cinfo, coef_arrays); + /* Wait for jpeg_finish_compress() call */ + cinfo->next_scanline = 0; /* so jpeg_write_marker works */ + cinfo->global_state = CSTATE_WRCOEFS; +} + + +/* + * Initialize the compression object with default parameters, + * then copy from the source object all parameters needed for lossless + * transcoding. Parameters that can be varied without loss (such as + * scan script and Huffman optimization) are left in their default states. + */ + +GLOBAL(void) +jpeg_copy_critical_parameters(j_decompress_ptr srcinfo, j_compress_ptr dstinfo) +{ + JQUANT_TBL **qtblptr; + jpeg_component_info *incomp, *outcomp; + JQUANT_TBL *c_quant, *slot_quant; + int tblno, ci, coefi; + + if (srcinfo->master->lossless) + ERREXIT(dstinfo, JERR_NOTIMPL); + + /* Safety check to ensure start_compress not called yet. */ + if (dstinfo->global_state != CSTATE_START) + ERREXIT1(dstinfo, JERR_BAD_STATE, dstinfo->global_state); + /* Copy fundamental image dimensions */ + dstinfo->image_width = srcinfo->image_width; + dstinfo->image_height = srcinfo->image_height; + dstinfo->input_components = srcinfo->num_components; + dstinfo->in_color_space = srcinfo->jpeg_color_space; +#if JPEG_LIB_VERSION >= 70 + dstinfo->jpeg_width = srcinfo->output_width; + dstinfo->jpeg_height = srcinfo->output_height; + dstinfo->min_DCT_h_scaled_size = srcinfo->min_DCT_h_scaled_size; + dstinfo->min_DCT_v_scaled_size = srcinfo->min_DCT_v_scaled_size; +#endif + /* Initialize all parameters to default values */ + jpeg_set_defaults(dstinfo); + /* jpeg_set_defaults may choose wrong colorspace, eg YCbCr if input is RGB. + * Fix it to get the right header markers for the image colorspace. + */ + jpeg_set_colorspace(dstinfo, srcinfo->jpeg_color_space); + dstinfo->data_precision = srcinfo->data_precision; + dstinfo->CCIR601_sampling = srcinfo->CCIR601_sampling; + /* Copy the source's quantization tables. */ + for (tblno = 0; tblno < NUM_QUANT_TBLS; tblno++) { + if (srcinfo->quant_tbl_ptrs[tblno] != NULL) { + qtblptr = &dstinfo->quant_tbl_ptrs[tblno]; + if (*qtblptr == NULL) + *qtblptr = jpeg_alloc_quant_table((j_common_ptr)dstinfo); + memcpy((*qtblptr)->quantval, srcinfo->quant_tbl_ptrs[tblno]->quantval, + sizeof((*qtblptr)->quantval)); + (*qtblptr)->sent_table = FALSE; + } + } + /* Copy the source's per-component info. + * Note we assume jpeg_set_defaults has allocated the dest comp_info array. + */ + dstinfo->num_components = srcinfo->num_components; + if (dstinfo->num_components < 1 || dstinfo->num_components > MAX_COMPONENTS) + ERREXIT2(dstinfo, JERR_COMPONENT_COUNT, dstinfo->num_components, + MAX_COMPONENTS); + for (ci = 0, incomp = srcinfo->comp_info, outcomp = dstinfo->comp_info; + ci < dstinfo->num_components; ci++, incomp++, outcomp++) { + outcomp->component_id = incomp->component_id; + outcomp->h_samp_factor = incomp->h_samp_factor; + outcomp->v_samp_factor = incomp->v_samp_factor; + outcomp->quant_tbl_no = incomp->quant_tbl_no; + /* Make sure saved quantization table for component matches the qtable + * slot. If not, the input file re-used this qtable slot. + * IJG encoder currently cannot duplicate this. + */ + tblno = outcomp->quant_tbl_no; + if (tblno < 0 || tblno >= NUM_QUANT_TBLS || + srcinfo->quant_tbl_ptrs[tblno] == NULL) + ERREXIT1(dstinfo, JERR_NO_QUANT_TABLE, tblno); + slot_quant = srcinfo->quant_tbl_ptrs[tblno]; + c_quant = incomp->quant_table; + if (c_quant != NULL) { + for (coefi = 0; coefi < DCTSIZE2; coefi++) { + if (c_quant->quantval[coefi] != slot_quant->quantval[coefi]) + ERREXIT1(dstinfo, JERR_MISMATCHED_QUANT_TABLE, tblno); + } + } + /* Note: we do not copy the source's Huffman table assignments; + * instead we rely on jpeg_set_colorspace to have made a suitable choice. + */ + } + /* Also copy JFIF version and resolution information, if available. + * Strictly speaking this isn't "critical" info, but it's nearly + * always appropriate to copy it if available. In particular, + * if the application chooses to copy JFIF 1.02 extension markers from + * the source file, we need to copy the version to make sure we don't + * emit a file that has 1.02 extensions but a claimed version of 1.01. + * We will *not*, however, copy version info from mislabeled "2.01" files. + */ + if (srcinfo->saw_JFIF_marker) { + if (srcinfo->JFIF_major_version == 1) { + dstinfo->JFIF_major_version = srcinfo->JFIF_major_version; + dstinfo->JFIF_minor_version = srcinfo->JFIF_minor_version; + } + dstinfo->density_unit = srcinfo->density_unit; + dstinfo->X_density = srcinfo->X_density; + dstinfo->Y_density = srcinfo->Y_density; + } +} + + +/* + * Master selection of compression modules for transcoding. + * This substitutes for jcinit.c's initialization of the full compressor. + */ + +LOCAL(void) +transencode_master_selection(j_compress_ptr cinfo, + jvirt_barray_ptr *coef_arrays) +{ + /* Although we don't actually use input_components for transcoding, + * jcmaster.c's initial_setup will complain if input_components is 0. + */ + cinfo->input_components = 1; + /* Initialize master control (includes parameter checking/processing) */ + jinit_c_master_control(cinfo, TRUE /* transcode only */); + + /* Entropy encoding: either Huffman or arithmetic coding. */ + if (cinfo->arith_code) { +#ifdef C_ARITH_CODING_SUPPORTED + jinit_arith_encoder(cinfo); +#else + ERREXIT(cinfo, JERR_ARITH_NOTIMPL); +#endif + } else { + if (cinfo->progressive_mode) { +#ifdef C_PROGRESSIVE_SUPPORTED + jinit_phuff_encoder(cinfo); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else + jinit_huff_encoder(cinfo); + } + + /* We need a special coefficient buffer controller. */ + transencode_coef_controller(cinfo, coef_arrays); + + jinit_marker_writer(cinfo); + + /* We can now tell the memory manager to allocate virtual arrays. */ + (*cinfo->mem->realize_virt_arrays) ((j_common_ptr)cinfo); + + /* Write the datastream header (SOI, JFIF) immediately. + * Frame and scan headers are postponed till later. + * This lets application insert special markers after the SOI. + */ + (*cinfo->marker->write_file_header) (cinfo); +} + + +/* + * The rest of this file is a special implementation of the coefficient + * buffer controller. This is similar to jccoefct.c, but it handles only + * output from presupplied virtual arrays. Furthermore, we generate any + * dummy padding blocks on-the-fly rather than expecting them to be present + * in the arrays. + */ + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_c_coef_controller pub; /* public fields */ + + JDIMENSION iMCU_row_num; /* iMCU row # within image */ + JDIMENSION mcu_ctr; /* counts MCUs processed in current row */ + int MCU_vert_offset; /* counts MCU rows within iMCU row */ + int MCU_rows_per_iMCU_row; /* number of such rows needed */ + + /* Virtual block array for each component. */ + jvirt_barray_ptr *whole_image; + + /* Workspace for constructing dummy blocks at right/bottom edges. */ + JBLOCKROW dummy_buffer[C_MAX_BLOCKS_IN_MCU]; +} my_coef_controller; + +typedef my_coef_controller *my_coef_ptr; + + +LOCAL(void) +start_iMCU_row(j_compress_ptr cinfo) +/* Reset within-iMCU-row counters for a new row */ +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + + /* In an interleaved scan, an MCU row is the same as an iMCU row. + * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. + * But at the bottom of the image, process only what's left. + */ + if (cinfo->comps_in_scan > 1) { + coef->MCU_rows_per_iMCU_row = 1; + } else { + if (coef->iMCU_row_num < (cinfo->total_iMCU_rows - 1)) + coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; + else + coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; + } + + coef->mcu_ctr = 0; + coef->MCU_vert_offset = 0; +} + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_coef(j_compress_ptr cinfo, J_BUF_MODE pass_mode) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + + if (pass_mode != JBUF_CRANK_DEST) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + + coef->iMCU_row_num = 0; + start_iMCU_row(cinfo); +} + + +/* + * Process some data. + * We process the equivalent of one fully interleaved MCU row ("iMCU" row) + * per call, ie, v_samp_factor block rows for each component in the scan. + * The data is obtained from the virtual arrays and fed to the entropy coder. + * Returns TRUE if the iMCU row is completed, FALSE if suspended. + * + * NB: input_buf is ignored; it is likely to be a NULL pointer. + */ + +METHODDEF(boolean) +compress_output(j_compress_ptr cinfo, JSAMPIMAGE input_buf) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION MCU_col_num; /* index of current MCU within row */ + JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + int blkn, ci, xindex, yindex, yoffset, blockcnt; + JDIMENSION start_col; + JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; + JBLOCKROW MCU_buffer[C_MAX_BLOCKS_IN_MCU]; + JBLOCKROW buffer_ptr; + jpeg_component_info *compptr; + + /* Align the virtual buffers for the components used in this scan. */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + buffer[ci] = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[compptr->component_index], + coef->iMCU_row_num * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, FALSE); + } + + /* Loop to process one whole iMCU row */ + for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; + yoffset++) { + for (MCU_col_num = coef->mcu_ctr; MCU_col_num < cinfo->MCUs_per_row; + MCU_col_num++) { + /* Construct list of pointers to DCT blocks belonging to this MCU */ + blkn = 0; /* index of current DCT block within MCU */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + start_col = MCU_col_num * compptr->MCU_width; + blockcnt = (MCU_col_num < last_MCU_col) ? compptr->MCU_width : + compptr->last_col_width; + for (yindex = 0; yindex < compptr->MCU_height; yindex++) { + if (coef->iMCU_row_num < last_iMCU_row || + yindex + yoffset < compptr->last_row_height) { + /* Fill in pointers to real blocks in this row */ + buffer_ptr = buffer[ci][yindex + yoffset] + start_col; + for (xindex = 0; xindex < blockcnt; xindex++) + MCU_buffer[blkn++] = buffer_ptr++; + } else { + /* At bottom of image, need a whole row of dummy blocks */ + xindex = 0; + } + /* Fill in any dummy blocks needed in this row. + * Dummy blocks are filled in the same way as in jccoefct.c: + * all zeroes in the AC entries, DC entries equal to previous + * block's DC value. The init routine has already zeroed the + * AC entries, so we need only set the DC entries correctly. + */ + for (; xindex < compptr->MCU_width; xindex++) { + MCU_buffer[blkn] = coef->dummy_buffer[blkn]; + MCU_buffer[blkn][0][0] = MCU_buffer[blkn - 1][0][0]; + blkn++; + } + } + } + /* Try to write the MCU. */ + if (!(*cinfo->entropy->encode_mcu) (cinfo, MCU_buffer)) { + /* Suspension forced; update state counters and exit */ + coef->MCU_vert_offset = yoffset; + coef->mcu_ctr = MCU_col_num; + return FALSE; + } + } + /* Completed an MCU row, but perhaps not an iMCU row */ + coef->mcu_ctr = 0; + } + /* Completed the iMCU row, advance counters for next one */ + coef->iMCU_row_num++; + start_iMCU_row(cinfo); + return TRUE; +} + + +METHODDEF(boolean) +compress_output_12(j_compress_ptr cinfo, J12SAMPIMAGE input_buf) +{ + return compress_output(cinfo, (JSAMPIMAGE)input_buf); +} + + +/* + * Initialize coefficient buffer controller. + * + * Each passed coefficient array must be the right size for that + * coefficient: width_in_blocks wide and height_in_blocks high, + * with unitheight at least v_samp_factor. + */ + +LOCAL(void) +transencode_coef_controller(j_compress_ptr cinfo, + jvirt_barray_ptr *coef_arrays) +{ + my_coef_ptr coef; + JBLOCKROW buffer; + int i; + + coef = (my_coef_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_coef_controller)); + cinfo->coef = (struct jpeg_c_coef_controller *)coef; + coef->pub.start_pass = start_pass_coef; + coef->pub.compress_data = compress_output; + coef->pub.compress_data_12 = compress_output_12; + + /* Save pointer to virtual arrays */ + coef->whole_image = coef_arrays; + + /* Allocate and pre-zero space for dummy DCT blocks. */ + buffer = (JBLOCKROW) + (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, + C_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK)); + jzero_far((void *)buffer, C_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK)); + for (i = 0; i < C_MAX_BLOCKS_IN_MCU; i++) { + coef->dummy_buffer[i] = buffer + i; + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jdapimin.c b/src/spice2x/external/libjpeg-turbo/jdapimin.c new file mode 100644 index 0000000..30d9284 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdapimin.c @@ -0,0 +1,417 @@ +/* + * jdapimin.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1998, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2016, 2022, 2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains application interface code for the decompression half + * of the JPEG library. These are the "minimum" API routines that may be + * needed in either the normal full-decompression case or the + * transcoding-only case. + * + * Most of the routines intended to be called directly by an application + * are in this file or in jdapistd.c. But also see jcomapi.c for routines + * shared by compression and decompression, and jdtrans.c for the transcoding + * case. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdmaster.h" + + +/* + * Initialization of a JPEG decompression object. + * The error manager must already be set up (in case memory manager fails). + */ + +GLOBAL(void) +jpeg_CreateDecompress(j_decompress_ptr cinfo, int version, size_t structsize) +{ + int i; + + /* Guard against version mismatches between library and caller. */ + cinfo->mem = NULL; /* so jpeg_destroy knows mem mgr not called */ + if (version != JPEG_LIB_VERSION) + ERREXIT2(cinfo, JERR_BAD_LIB_VERSION, JPEG_LIB_VERSION, version); + if (structsize != sizeof(struct jpeg_decompress_struct)) + ERREXIT2(cinfo, JERR_BAD_STRUCT_SIZE, + (int)sizeof(struct jpeg_decompress_struct), (int)structsize); + + /* For debugging purposes, we zero the whole master structure. + * But the application has already set the err pointer, and may have set + * client_data, so we have to save and restore those fields. + * Note: if application hasn't set client_data, tools like Purify may + * complain here. + */ + { + struct jpeg_error_mgr *err = cinfo->err; + void *client_data = cinfo->client_data; /* ignore Purify complaint here */ + memset(cinfo, 0, sizeof(struct jpeg_decompress_struct)); + cinfo->err = err; + cinfo->client_data = client_data; + } + cinfo->is_decompressor = TRUE; + + /* Initialize a memory manager instance for this object */ + jinit_memory_mgr((j_common_ptr)cinfo); + + /* Zero out pointers to permanent structures. */ + cinfo->progress = NULL; + cinfo->src = NULL; + + for (i = 0; i < NUM_QUANT_TBLS; i++) + cinfo->quant_tbl_ptrs[i] = NULL; + + for (i = 0; i < NUM_HUFF_TBLS; i++) { + cinfo->dc_huff_tbl_ptrs[i] = NULL; + cinfo->ac_huff_tbl_ptrs[i] = NULL; + } + + /* Initialize marker processor so application can override methods + * for COM, APPn markers before calling jpeg_read_header. + */ + cinfo->marker_list = NULL; + jinit_marker_reader(cinfo); + + /* And initialize the overall input controller. */ + jinit_input_controller(cinfo); + + cinfo->data_precision = BITS_IN_JSAMPLE; + + /* OK, I'm ready */ + cinfo->global_state = DSTATE_START; + + /* The master struct is used to store extension parameters, so we allocate it + * here. + */ + cinfo->master = (struct jpeg_decomp_master *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(my_decomp_master)); + memset(cinfo->master, 0, sizeof(my_decomp_master)); +} + + +/* + * Destruction of a JPEG decompression object + */ + +GLOBAL(void) +jpeg_destroy_decompress(j_decompress_ptr cinfo) +{ + jpeg_destroy((j_common_ptr)cinfo); /* use common routine */ +} + + +/* + * Abort processing of a JPEG decompression operation, + * but don't destroy the object itself. + */ + +GLOBAL(void) +jpeg_abort_decompress(j_decompress_ptr cinfo) +{ + jpeg_abort((j_common_ptr)cinfo); /* use common routine */ +} + + +/* + * Set default decompression parameters. + */ + +LOCAL(void) +default_decompress_parms(j_decompress_ptr cinfo) +{ + /* Guess the input colorspace, and set output colorspace accordingly. */ + /* (Wish JPEG committee had provided a real way to specify this...) */ + /* Note application may override our guesses. */ + switch (cinfo->num_components) { + case 1: + cinfo->jpeg_color_space = JCS_GRAYSCALE; + cinfo->out_color_space = JCS_GRAYSCALE; + break; + + case 3: + if (cinfo->saw_JFIF_marker) { + cinfo->jpeg_color_space = JCS_YCbCr; /* JFIF implies YCbCr */ + } else if (cinfo->saw_Adobe_marker) { + switch (cinfo->Adobe_transform) { + case 0: + cinfo->jpeg_color_space = JCS_RGB; + break; + case 1: + cinfo->jpeg_color_space = JCS_YCbCr; + break; + default: + WARNMS1(cinfo, JWRN_ADOBE_XFORM, cinfo->Adobe_transform); + cinfo->jpeg_color_space = JCS_YCbCr; /* assume it's YCbCr */ + break; + } + } else { + /* Saw no special markers, try to guess from the component IDs */ + int cid0 = cinfo->comp_info[0].component_id; + int cid1 = cinfo->comp_info[1].component_id; + int cid2 = cinfo->comp_info[2].component_id; + + if (cid0 == 1 && cid1 == 2 && cid2 == 3) { + if (cinfo->master->lossless) + cinfo->jpeg_color_space = JCS_RGB; /* assume RGB w/out marker */ + else + cinfo->jpeg_color_space = JCS_YCbCr; /* assume JFIF w/out marker */ + } else if (cid0 == 82 && cid1 == 71 && cid2 == 66) + cinfo->jpeg_color_space = JCS_RGB; /* ASCII 'R', 'G', 'B' */ + else { + TRACEMS3(cinfo, 1, JTRC_UNKNOWN_IDS, cid0, cid1, cid2); + if (cinfo->master->lossless) + cinfo->jpeg_color_space = JCS_RGB; /* assume it's RGB */ + else + cinfo->jpeg_color_space = JCS_YCbCr; /* assume it's YCbCr */ + } + } + /* Always guess RGB is proper output colorspace. */ + cinfo->out_color_space = JCS_RGB; + break; + + case 4: + if (cinfo->saw_Adobe_marker) { + switch (cinfo->Adobe_transform) { + case 0: + cinfo->jpeg_color_space = JCS_CMYK; + break; + case 2: + cinfo->jpeg_color_space = JCS_YCCK; + break; + default: + WARNMS1(cinfo, JWRN_ADOBE_XFORM, cinfo->Adobe_transform); + cinfo->jpeg_color_space = JCS_YCCK; /* assume it's YCCK */ + break; + } + } else { + /* No special markers, assume straight CMYK. */ + cinfo->jpeg_color_space = JCS_CMYK; + } + cinfo->out_color_space = JCS_CMYK; + break; + + default: + cinfo->jpeg_color_space = JCS_UNKNOWN; + cinfo->out_color_space = JCS_UNKNOWN; + break; + } + + /* Set defaults for other decompression parameters. */ + cinfo->scale_num = 1; /* 1:1 scaling */ + cinfo->scale_denom = 1; + cinfo->output_gamma = 1.0; + cinfo->buffered_image = FALSE; + cinfo->raw_data_out = FALSE; + cinfo->dct_method = JDCT_DEFAULT; + cinfo->do_fancy_upsampling = TRUE; + cinfo->do_block_smoothing = TRUE; + cinfo->quantize_colors = FALSE; + /* We set these in case application only sets quantize_colors. */ + cinfo->dither_mode = JDITHER_FS; +#ifdef QUANT_2PASS_SUPPORTED + cinfo->two_pass_quantize = TRUE; +#else + cinfo->two_pass_quantize = FALSE; +#endif + cinfo->desired_number_of_colors = 256; + cinfo->colormap = NULL; + /* Initialize for no mode change in buffered-image mode. */ + cinfo->enable_1pass_quant = FALSE; + cinfo->enable_external_quant = FALSE; + cinfo->enable_2pass_quant = FALSE; +} + + +/* + * Decompression startup: read start of JPEG datastream to see what's there. + * Need only initialize JPEG object and supply a data source before calling. + * + * This routine will read as far as the first SOS marker (ie, actual start of + * compressed data), and will save all tables and parameters in the JPEG + * object. It will also initialize the decompression parameters to default + * values, and finally return JPEG_HEADER_OK. On return, the application may + * adjust the decompression parameters and then call jpeg_start_decompress. + * (Or, if the application only wanted to determine the image parameters, + * the data need not be decompressed. In that case, call jpeg_abort or + * jpeg_destroy to release any temporary space.) + * If an abbreviated (tables only) datastream is presented, the routine will + * return JPEG_HEADER_TABLES_ONLY upon reaching EOI. The application may then + * re-use the JPEG object to read the abbreviated image datastream(s). + * It is unnecessary (but OK) to call jpeg_abort in this case. + * The JPEG_SUSPENDED return code only occurs if the data source module + * requests suspension of the decompressor. In this case the application + * should load more source data and then re-call jpeg_read_header to resume + * processing. + * If a non-suspending data source is used and require_image is TRUE, then the + * return code need not be inspected since only JPEG_HEADER_OK is possible. + * + * This routine is now just a front end to jpeg_consume_input, with some + * extra error checking. + */ + +GLOBAL(int) +jpeg_read_header(j_decompress_ptr cinfo, boolean require_image) +{ + int retcode; + + if (cinfo->global_state != DSTATE_START && + cinfo->global_state != DSTATE_INHEADER) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + retcode = jpeg_consume_input(cinfo); + + switch (retcode) { + case JPEG_REACHED_SOS: + retcode = JPEG_HEADER_OK; + break; + case JPEG_REACHED_EOI: + if (require_image) /* Complain if application wanted an image */ + ERREXIT(cinfo, JERR_NO_IMAGE); + /* Reset to start state; it would be safer to require the application to + * call jpeg_abort, but we can't change it now for compatibility reasons. + * A side effect is to free any temporary memory (there shouldn't be any). + */ + jpeg_abort((j_common_ptr)cinfo); /* sets state = DSTATE_START */ + retcode = JPEG_HEADER_TABLES_ONLY; + break; + case JPEG_SUSPENDED: + /* no work */ + break; + } + + return retcode; +} + + +/* + * Consume data in advance of what the decompressor requires. + * This can be called at any time once the decompressor object has + * been created and a data source has been set up. + * + * This routine is essentially a state machine that handles a couple + * of critical state-transition actions, namely initial setup and + * transition from header scanning to ready-for-start_decompress. + * All the actual input is done via the input controller's consume_input + * method. + */ + +GLOBAL(int) +jpeg_consume_input(j_decompress_ptr cinfo) +{ + int retcode = JPEG_SUSPENDED; + + /* NB: every possible DSTATE value should be listed in this switch */ + switch (cinfo->global_state) { + case DSTATE_START: + /* Start-of-datastream actions: reset appropriate modules */ + (*cinfo->inputctl->reset_input_controller) (cinfo); + /* Initialize application's data source module */ + (*cinfo->src->init_source) (cinfo); + cinfo->global_state = DSTATE_INHEADER; + FALLTHROUGH /*FALLTHROUGH*/ + case DSTATE_INHEADER: + retcode = (*cinfo->inputctl->consume_input) (cinfo); + if (retcode == JPEG_REACHED_SOS) { /* Found SOS, prepare to decompress */ + /* Set up default parameters based on header data */ + default_decompress_parms(cinfo); + /* Set global state: ready for start_decompress */ + cinfo->global_state = DSTATE_READY; + } + break; + case DSTATE_READY: + /* Can't advance past first SOS until start_decompress is called */ + retcode = JPEG_REACHED_SOS; + break; + case DSTATE_PRELOAD: + case DSTATE_PRESCAN: + case DSTATE_SCANNING: + case DSTATE_RAW_OK: + case DSTATE_BUFIMAGE: + case DSTATE_BUFPOST: + case DSTATE_STOPPING: + retcode = (*cinfo->inputctl->consume_input) (cinfo); + break; + default: + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + } + return retcode; +} + + +/* + * Have we finished reading the input file? + */ + +GLOBAL(boolean) +jpeg_input_complete(j_decompress_ptr cinfo) +{ + /* Check for valid jpeg object */ + if (cinfo->global_state < DSTATE_START || + cinfo->global_state > DSTATE_STOPPING) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + return cinfo->inputctl->eoi_reached; +} + + +/* + * Is there more than one scan? + */ + +GLOBAL(boolean) +jpeg_has_multiple_scans(j_decompress_ptr cinfo) +{ + /* Only valid after jpeg_read_header completes */ + if (cinfo->global_state < DSTATE_READY || + cinfo->global_state > DSTATE_STOPPING) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + return cinfo->inputctl->has_multiple_scans; +} + + +/* + * Finish JPEG decompression. + * + * This will normally just verify the file trailer and release temp storage. + * + * Returns FALSE if suspended. The return value need be inspected only if + * a suspending data source is used. + */ + +GLOBAL(boolean) +jpeg_finish_decompress(j_decompress_ptr cinfo) +{ + if ((cinfo->global_state == DSTATE_SCANNING || + cinfo->global_state == DSTATE_RAW_OK) && !cinfo->buffered_image) { + /* Terminate final pass of non-buffered mode */ + if (cinfo->output_scanline < cinfo->output_height) + ERREXIT(cinfo, JERR_TOO_LITTLE_DATA); + (*cinfo->master->finish_output_pass) (cinfo); + cinfo->global_state = DSTATE_STOPPING; + } else if (cinfo->global_state == DSTATE_BUFIMAGE) { + /* Finishing after a buffered-image operation */ + cinfo->global_state = DSTATE_STOPPING; + } else if (cinfo->global_state != DSTATE_STOPPING) { + /* STOPPING = repeat call after a suspension, anything else is error */ + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + } + /* Read until EOI */ + while (!cinfo->inputctl->eoi_reached) { + if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED) + return FALSE; /* Suspend, come back later */ + } + /* Do final cleanup */ + (*cinfo->src->term_source) (cinfo); + /* We can use jpeg_abort to release memory and reset global_state */ + jpeg_abort((j_common_ptr)cinfo); + return TRUE; +} diff --git a/src/spice2x/external/libjpeg-turbo/jdapistd.c b/src/spice2x/external/libjpeg-turbo/jdapistd.c new file mode 100644 index 0000000..38da36b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdapistd.c @@ -0,0 +1,752 @@ +/* + * jdapistd.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2010, 2015-2020, 2022-2024, D. R. Commander. + * Copyright (C) 2015, Google, Inc. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains application interface code for the decompression half + * of the JPEG library. These are the "standard" API routines that are + * used in the normal full-decompression case. They are not used by a + * transcoding-only application. Note that if an application links in + * jpeg_start_decompress, it will end up linking in the entire decompressor. + * We thus must separate this file from jdapimin.c to avoid linking the + * whole decompression library into a transcoder. + */ + +#include "jinclude.h" +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) +#include "jdmainct.h" +#include "jdcoefct.h" +#else +#define JPEG_INTERNALS +#include "jpeglib.h" +#endif +#include "jdmaster.h" +#include "jdmerge.h" +#include "jdsample.h" +#include "jmemsys.h" + +#if BITS_IN_JSAMPLE == 8 + +/* Forward declarations */ +LOCAL(boolean) output_pass_setup(j_decompress_ptr cinfo); + + +/* + * Decompression initialization. + * jpeg_read_header must be completed before calling this. + * + * If a multipass operating mode was selected, this will do all but the + * last pass, and thus may take a great deal of time. + * + * Returns FALSE if suspended. The return value need be inspected only if + * a suspending data source is used. + */ + +GLOBAL(boolean) +jpeg_start_decompress(j_decompress_ptr cinfo) +{ + if (cinfo->global_state == DSTATE_READY) { + /* First call: initialize master control, select active modules */ + jinit_master_decompress(cinfo); + if (cinfo->buffered_image) { + /* No more work here; expecting jpeg_start_output next */ + cinfo->global_state = DSTATE_BUFIMAGE; + return TRUE; + } + cinfo->global_state = DSTATE_PRELOAD; + } + if (cinfo->global_state == DSTATE_PRELOAD) { + /* If file has multiple scans, absorb them all into the coef buffer */ + if (cinfo->inputctl->has_multiple_scans) { +#ifdef D_MULTISCAN_FILES_SUPPORTED + for (;;) { + int retcode; + /* Call progress monitor hook if present */ + if (cinfo->progress != NULL) + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + /* Absorb some more input */ + retcode = (*cinfo->inputctl->consume_input) (cinfo); + if (retcode == JPEG_SUSPENDED) + return FALSE; + if (retcode == JPEG_REACHED_EOI) + break; + /* Advance progress counter if appropriate */ + if (cinfo->progress != NULL && + (retcode == JPEG_ROW_COMPLETED || retcode == JPEG_REACHED_SOS)) { + if (++cinfo->progress->pass_counter >= cinfo->progress->pass_limit) { + /* jdmaster underestimated number of scans; ratchet up one scan */ + cinfo->progress->pass_limit += (long)cinfo->total_iMCU_rows; + } + } + } +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif /* D_MULTISCAN_FILES_SUPPORTED */ + } + cinfo->output_scan_number = cinfo->input_scan_number; + } else if (cinfo->global_state != DSTATE_PRESCAN) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + /* Perform any dummy output passes, and set up for the final pass */ + return output_pass_setup(cinfo); +} + + +/* + * Set up for an output pass, and perform any dummy pass(es) needed. + * Common subroutine for jpeg_start_decompress and jpeg_start_output. + * Entry: global_state = DSTATE_PRESCAN only if previously suspended. + * Exit: If done, returns TRUE and sets global_state for proper output mode. + * If suspended, returns FALSE and sets global_state = DSTATE_PRESCAN. + */ + +LOCAL(boolean) +output_pass_setup(j_decompress_ptr cinfo) +{ + if (cinfo->global_state != DSTATE_PRESCAN) { + /* First call: do pass setup */ + (*cinfo->master->prepare_for_output_pass) (cinfo); + cinfo->output_scanline = 0; + cinfo->global_state = DSTATE_PRESCAN; + } + /* Loop over any required dummy passes */ + while (cinfo->master->is_dummy_pass) { +#ifdef QUANT_2PASS_SUPPORTED + /* Crank through the dummy pass */ + while (cinfo->output_scanline < cinfo->output_height) { + JDIMENSION last_scanline; + /* Call progress monitor hook if present */ + if (cinfo->progress != NULL) { + cinfo->progress->pass_counter = (long)cinfo->output_scanline; + cinfo->progress->pass_limit = (long)cinfo->output_height; + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + } + /* Process some data */ + last_scanline = cinfo->output_scanline; +#ifdef D_LOSSLESS_SUPPORTED + if (cinfo->data_precision == 16) + (*cinfo->main->process_data_16) (cinfo, (J16SAMPARRAY)NULL, + &cinfo->output_scanline, + (JDIMENSION)0); + else +#endif + if (cinfo->data_precision == 12) + (*cinfo->main->process_data_12) (cinfo, (J12SAMPARRAY)NULL, + &cinfo->output_scanline, + (JDIMENSION)0); + else + (*cinfo->main->process_data) (cinfo, (JSAMPARRAY)NULL, + &cinfo->output_scanline, (JDIMENSION)0); + if (cinfo->output_scanline == last_scanline) + return FALSE; /* No progress made, must suspend */ + } + /* Finish up dummy pass, and set up for another one */ + (*cinfo->master->finish_output_pass) (cinfo); + (*cinfo->master->prepare_for_output_pass) (cinfo); + cinfo->output_scanline = 0; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif /* QUANT_2PASS_SUPPORTED */ + } + /* Ready for application to drive output pass through + * _jpeg_read_scanlines or _jpeg_read_raw_data. + */ + cinfo->global_state = cinfo->raw_data_out ? DSTATE_RAW_OK : DSTATE_SCANNING; + return TRUE; +} + +#endif /* BITS_IN_JSAMPLE == 8 */ + + +#if BITS_IN_JSAMPLE != 16 + +/* + * Enable partial scanline decompression + * + * Must be called after jpeg_start_decompress() and before any calls to + * _jpeg_read_scanlines() or _jpeg_skip_scanlines(). + * + * Refer to libjpeg.txt for more information. + */ + +GLOBAL(void) +_jpeg_crop_scanline(j_decompress_ptr cinfo, JDIMENSION *xoffset, + JDIMENSION *width) +{ + int ci, align, orig_downsampled_width; + JDIMENSION input_xoffset; + boolean reinit_upsampler = FALSE; + jpeg_component_info *compptr; +#ifdef UPSAMPLE_MERGING_SUPPORTED + my_master_ptr master = (my_master_ptr)cinfo->master; +#endif + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + if ((cinfo->global_state != DSTATE_SCANNING && + cinfo->global_state != DSTATE_BUFIMAGE) || cinfo->output_scanline != 0) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + if (!xoffset || !width) + ERREXIT(cinfo, JERR_BAD_CROP_SPEC); + + /* xoffset and width must fall within the output image dimensions. */ + if (*width == 0 || + (unsigned long long)(*xoffset) + *width > cinfo->output_width) + ERREXIT(cinfo, JERR_WIDTH_OVERFLOW); + + /* No need to do anything if the caller wants the entire width. */ + if (*width == cinfo->output_width) + return; + + /* Ensuring the proper alignment of xoffset is tricky. At minimum, it + * must align with an MCU boundary, because: + * + * (1) The IDCT is performed in blocks, and it is not feasible to modify + * the algorithm so that it can transform partial blocks. + * (2) Because of the SIMD extensions, any input buffer passed to the + * upsampling and color conversion routines must be aligned to the + * SIMD word size (for instance, 128-bit in the case of SSE2.) The + * easiest way to accomplish this without copying data is to ensure + * that upsampling and color conversion begin at the start of the + * first MCU column that will be inverse transformed. + * + * In practice, we actually impose a stricter alignment requirement. We + * require that xoffset be a multiple of the maximum MCU column width of all + * of the components (the "iMCU column width.") This is to simplify the + * single-pass decompression case, allowing us to use the same MCU column + * width for all of the components. + */ + if (cinfo->comps_in_scan == 1 && cinfo->num_components == 1) + align = cinfo->_min_DCT_scaled_size; + else + align = cinfo->_min_DCT_scaled_size * cinfo->max_h_samp_factor; + + /* Adjust xoffset to the nearest iMCU boundary <= the requested value */ + input_xoffset = *xoffset; + *xoffset = (input_xoffset / align) * align; + + /* Adjust the width so that the right edge of the output image is as + * requested (only the left edge is altered.) It is important that calling + * programs check this value after this function returns, so that they can + * allocate an output buffer with the appropriate size. + */ + *width = *width + input_xoffset - *xoffset; + cinfo->output_width = *width; +#ifdef UPSAMPLE_MERGING_SUPPORTED + if (master->using_merged_upsample && cinfo->max_v_samp_factor == 2) { + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + upsample->out_row_width = + cinfo->output_width * cinfo->out_color_components; + } +#endif + + /* Set the first and last iMCU columns that we must decompress. These values + * will be used in single-scan decompressions. + */ + cinfo->master->first_iMCU_col = (JDIMENSION)(long)(*xoffset) / (long)align; + cinfo->master->last_iMCU_col = + (JDIMENSION)jdiv_round_up((long)(*xoffset + cinfo->output_width), + (long)align) - 1; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + int hsf = (cinfo->comps_in_scan == 1 && cinfo->num_components == 1) ? + 1 : compptr->h_samp_factor; + + /* Set downsampled_width to the new output width. */ + orig_downsampled_width = compptr->downsampled_width; + compptr->downsampled_width = + (JDIMENSION)jdiv_round_up((long)cinfo->output_width * + (long)(compptr->h_samp_factor * + compptr->_DCT_scaled_size), + (long)(cinfo->max_h_samp_factor * + cinfo->_min_DCT_scaled_size)); + if (compptr->downsampled_width < 2 && orig_downsampled_width >= 2) + reinit_upsampler = TRUE; + + /* Set the first and last iMCU columns that we must decompress. These + * values will be used in multi-scan decompressions. + */ + cinfo->master->first_MCU_col[ci] = + (JDIMENSION)(long)(*xoffset * hsf) / (long)align; + cinfo->master->last_MCU_col[ci] = + (JDIMENSION)jdiv_round_up((long)((*xoffset + cinfo->output_width) * hsf), + (long)align) - 1; + } + + if (reinit_upsampler) { + cinfo->master->jinit_upsampler_no_alloc = TRUE; + _jinit_upsampler(cinfo); + cinfo->master->jinit_upsampler_no_alloc = FALSE; + } +} + +#endif /* BITS_IN_JSAMPLE != 16 */ + + +/* + * Read some scanlines of data from the JPEG decompressor. + * + * The return value will be the number of lines actually read. + * This may be less than the number requested in several cases, + * including bottom of image, data source suspension, and operating + * modes that emit multiple scanlines at a time. + * + * Note: we warn about excess calls to _jpeg_read_scanlines() since + * this likely signals an application programmer error. However, + * an oversize buffer (max_lines > scanlines remaining) is not an error. + */ + +GLOBAL(JDIMENSION) +_jpeg_read_scanlines(j_decompress_ptr cinfo, _JSAMPARRAY scanlines, + JDIMENSION max_lines) +{ +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) + JDIMENSION row_ctr; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (cinfo->global_state != DSTATE_SCANNING) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + if (cinfo->output_scanline >= cinfo->output_height) { + WARNMS(cinfo, JWRN_TOO_MUCH_DATA); + return 0; + } + + /* Call progress monitor hook if present */ + if (cinfo->progress != NULL) { + cinfo->progress->pass_counter = (long)cinfo->output_scanline; + cinfo->progress->pass_limit = (long)cinfo->output_height; + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + } + + /* Process some data */ + row_ctr = 0; + (*cinfo->main->_process_data) (cinfo, scanlines, &row_ctr, max_lines); + cinfo->output_scanline += row_ctr; + return row_ctr; +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + return 0; +#endif +} + + +#if BITS_IN_JSAMPLE != 16 + +/* Dummy color convert function used by _jpeg_skip_scanlines() */ +LOCAL(void) +noop_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ +} + + +/* Dummy quantize function used by _jpeg_skip_scanlines() */ +LOCAL(void) +noop_quantize(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +{ +} + + +/* + * In some cases, it is best to call _jpeg_read_scanlines() and discard the + * output, rather than skipping the scanlines, because this allows us to + * maintain the internal state of the context-based upsampler. In these cases, + * we set up and tear down a dummy color converter in order to avoid valgrind + * errors and to achieve the best possible performance. + */ + +LOCAL(void) +read_and_discard_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines) +{ + JDIMENSION n; +#ifdef UPSAMPLE_MERGING_SUPPORTED + my_master_ptr master = (my_master_ptr)cinfo->master; +#endif + _JSAMPLE dummy_sample[1] = { 0 }; + _JSAMPROW dummy_row = dummy_sample; + _JSAMPARRAY scanlines = NULL; + void (*color_convert) (j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) = NULL; + void (*color_quantize) (j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) = NULL; + + if (cinfo->cconvert && cinfo->cconvert->_color_convert) { + color_convert = cinfo->cconvert->_color_convert; + cinfo->cconvert->_color_convert = noop_convert; + /* This just prevents UBSan from complaining about adding 0 to a NULL + * pointer. The pointer isn't actually used. + */ + scanlines = &dummy_row; + } + + if (cinfo->cquantize && cinfo->cquantize->_color_quantize) { + color_quantize = cinfo->cquantize->_color_quantize; + cinfo->cquantize->_color_quantize = noop_quantize; + } + +#ifdef UPSAMPLE_MERGING_SUPPORTED + if (master->using_merged_upsample && cinfo->max_v_samp_factor == 2) { + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + scanlines = &upsample->spare_row; + } +#endif + + for (n = 0; n < num_lines; n++) + _jpeg_read_scanlines(cinfo, scanlines, 1); + + if (color_convert) + cinfo->cconvert->_color_convert = color_convert; + + if (color_quantize) + cinfo->cquantize->_color_quantize = color_quantize; +} + + +/* + * Called by _jpeg_skip_scanlines(). This partially skips a decompress block + * by incrementing the rowgroup counter. + */ + +LOCAL(void) +increment_simple_rowgroup_ctr(j_decompress_ptr cinfo, JDIMENSION rows) +{ + JDIMENSION rows_left; + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + my_master_ptr master = (my_master_ptr)cinfo->master; + + if (master->using_merged_upsample && cinfo->max_v_samp_factor == 2) { + read_and_discard_scanlines(cinfo, rows); + return; + } + + /* Increment the counter to the next row group after the skipped rows. */ + main_ptr->rowgroup_ctr += rows / cinfo->max_v_samp_factor; + + /* Partially skipping a row group would involve modifying the internal state + * of the upsampler, so read the remaining rows into a dummy buffer instead. + */ + rows_left = rows % cinfo->max_v_samp_factor; + cinfo->output_scanline += rows - rows_left; + + read_and_discard_scanlines(cinfo, rows_left); +} + +/* + * Skips some scanlines of data from the JPEG decompressor. + * + * The return value will be the number of lines actually skipped. If skipping + * num_lines would move beyond the end of the image, then the actual number of + * lines remaining in the image is returned. Otherwise, the return value will + * be equal to num_lines. + * + * Refer to libjpeg.txt for more information. + */ + +GLOBAL(JDIMENSION) +_jpeg_skip_scanlines(j_decompress_ptr cinfo, JDIMENSION num_lines) +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + my_master_ptr master = (my_master_ptr)cinfo->master; + my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample; + JDIMENSION i, x; + int y; + JDIMENSION lines_per_iMCU_row, lines_left_in_iMCU_row, lines_after_iMCU_row; + JDIMENSION lines_to_skip, lines_to_read; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + /* Two-pass color quantization is not supported. */ + if (cinfo->quantize_colors && cinfo->two_pass_quantize) + ERREXIT(cinfo, JERR_NOTIMPL); + + if (cinfo->global_state != DSTATE_SCANNING) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + /* Do not skip past the bottom of the image. */ + if ((unsigned long long)cinfo->output_scanline + num_lines >= + cinfo->output_height) { + num_lines = cinfo->output_height - cinfo->output_scanline; + cinfo->output_scanline = cinfo->output_height; + (*cinfo->inputctl->finish_input_pass) (cinfo); + cinfo->inputctl->eoi_reached = TRUE; + return num_lines; + } + + if (num_lines == 0) + return 0; + + lines_per_iMCU_row = cinfo->_min_DCT_scaled_size * cinfo->max_v_samp_factor; + lines_left_in_iMCU_row = + (lines_per_iMCU_row - (cinfo->output_scanline % lines_per_iMCU_row)) % + lines_per_iMCU_row; + lines_after_iMCU_row = num_lines - lines_left_in_iMCU_row; + + /* Skip the lines remaining in the current iMCU row. When upsampling + * requires context rows, we need the previous and next rows in order to read + * the current row. This adds some complexity. + */ + if (cinfo->upsample->need_context_rows) { + /* If the skipped lines would not move us past the current iMCU row, we + * read the lines and ignore them. There might be a faster way of doing + * this, but we are facing increasing complexity for diminishing returns. + * The increasing complexity would be a by-product of meddling with the + * state machine used to skip context rows. Near the end of an iMCU row, + * the next iMCU row may have already been entropy-decoded. In this unique + * case, we will read the next iMCU row if we cannot skip past it as well. + */ + if ((num_lines < lines_left_in_iMCU_row + 1) || + (lines_left_in_iMCU_row <= 1 && main_ptr->buffer_full && + lines_after_iMCU_row < lines_per_iMCU_row + 1)) { + read_and_discard_scanlines(cinfo, num_lines); + return num_lines; + } + + /* If the next iMCU row has already been entropy-decoded, make sure that + * we do not skip too far. + */ + if (lines_left_in_iMCU_row <= 1 && main_ptr->buffer_full) { + cinfo->output_scanline += lines_left_in_iMCU_row + lines_per_iMCU_row; + lines_after_iMCU_row -= lines_per_iMCU_row; + } else { + cinfo->output_scanline += lines_left_in_iMCU_row; + } + + /* If we have just completed the first block, adjust the buffer pointers */ + if (main_ptr->iMCU_row_ctr == 0 || + (main_ptr->iMCU_row_ctr == 1 && lines_left_in_iMCU_row > 2)) + set_wraparound_pointers(cinfo); + main_ptr->buffer_full = FALSE; + main_ptr->rowgroup_ctr = 0; + main_ptr->context_state = CTX_PREPARE_FOR_IMCU; + if (!master->using_merged_upsample) { + upsample->next_row_out = cinfo->max_v_samp_factor; + upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline; + } + } + + /* Skipping is much simpler when context rows are not required. */ + else { + if (num_lines < lines_left_in_iMCU_row) { + increment_simple_rowgroup_ctr(cinfo, num_lines); + return num_lines; + } else { + cinfo->output_scanline += lines_left_in_iMCU_row; + main_ptr->buffer_full = FALSE; + main_ptr->rowgroup_ctr = 0; + if (!master->using_merged_upsample) { + upsample->next_row_out = cinfo->max_v_samp_factor; + upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline; + } + } + } + + /* Calculate how many full iMCU rows we can skip. */ + if (cinfo->upsample->need_context_rows) + lines_to_skip = ((lines_after_iMCU_row - 1) / lines_per_iMCU_row) * + lines_per_iMCU_row; + else + lines_to_skip = (lines_after_iMCU_row / lines_per_iMCU_row) * + lines_per_iMCU_row; + /* Calculate the number of lines that remain to be skipped after skipping all + * of the full iMCU rows that we can. We will not read these lines unless we + * have to. + */ + lines_to_read = lines_after_iMCU_row - lines_to_skip; + + /* For images requiring multiple scans (progressive, non-interleaved, etc.), + * all of the entropy decoding occurs in jpeg_start_decompress(), assuming + * that the input data source is non-suspending. This makes skipping easy. + */ + if (cinfo->inputctl->has_multiple_scans || cinfo->buffered_image) { + if (cinfo->upsample->need_context_rows) { + cinfo->output_scanline += lines_to_skip; + cinfo->output_iMCU_row += lines_to_skip / lines_per_iMCU_row; + main_ptr->iMCU_row_ctr += lines_to_skip / lines_per_iMCU_row; + /* It is complex to properly move to the middle of a context block, so + * read the remaining lines instead of skipping them. + */ + read_and_discard_scanlines(cinfo, lines_to_read); + } else { + cinfo->output_scanline += lines_to_skip; + cinfo->output_iMCU_row += lines_to_skip / lines_per_iMCU_row; + increment_simple_rowgroup_ctr(cinfo, lines_to_read); + } + if (!master->using_merged_upsample) + upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline; + return num_lines; + } + + /* Skip the iMCU rows that we can safely skip. */ + for (i = 0; i < lines_to_skip; i += lines_per_iMCU_row) { + for (y = 0; y < coef->MCU_rows_per_iMCU_row; y++) { + for (x = 0; x < cinfo->MCUs_per_row; x++) { + /* Calling decode_mcu() with a NULL pointer causes it to discard the + * decoded coefficients. This is ~5% faster for large subsets, but + * it's tough to tell a difference for smaller images. + */ + if (!cinfo->entropy->insufficient_data) + cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row; + (*cinfo->entropy->decode_mcu) (cinfo, NULL); + } + } + cinfo->input_iMCU_row++; + cinfo->output_iMCU_row++; + if (cinfo->input_iMCU_row < cinfo->total_iMCU_rows) + start_iMCU_row(cinfo); + else + (*cinfo->inputctl->finish_input_pass) (cinfo); + } + cinfo->output_scanline += lines_to_skip; + + if (cinfo->upsample->need_context_rows) { + /* Context-based upsampling keeps track of iMCU rows. */ + main_ptr->iMCU_row_ctr += lines_to_skip / lines_per_iMCU_row; + + /* It is complex to properly move to the middle of a context block, so + * read the remaining lines instead of skipping them. + */ + read_and_discard_scanlines(cinfo, lines_to_read); + } else { + increment_simple_rowgroup_ctr(cinfo, lines_to_read); + } + + /* Since skipping lines involves skipping the upsampling step, the value of + * "rows_to_go" will become invalid unless we set it here. NOTE: This is a + * bit odd, since "rows_to_go" seems to be redundantly keeping track of + * output_scanline. + */ + if (!master->using_merged_upsample) + upsample->rows_to_go = cinfo->output_height - cinfo->output_scanline; + + /* Always skip the requested number of lines. */ + return num_lines; +} + +/* + * Alternate entry point to read raw data. + * Processes exactly one iMCU row per call, unless suspended. + */ + +GLOBAL(JDIMENSION) +_jpeg_read_raw_data(j_decompress_ptr cinfo, _JSAMPIMAGE data, + JDIMENSION max_lines) +{ + JDIMENSION lines_per_iMCU_row; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + if (cinfo->global_state != DSTATE_RAW_OK) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + if (cinfo->output_scanline >= cinfo->output_height) { + WARNMS(cinfo, JWRN_TOO_MUCH_DATA); + return 0; + } + + /* Call progress monitor hook if present */ + if (cinfo->progress != NULL) { + cinfo->progress->pass_counter = (long)cinfo->output_scanline; + cinfo->progress->pass_limit = (long)cinfo->output_height; + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + } + + /* Verify that at least one iMCU row can be returned. */ + lines_per_iMCU_row = cinfo->max_v_samp_factor * cinfo->_min_DCT_scaled_size; + if (max_lines < lines_per_iMCU_row) + ERREXIT(cinfo, JERR_BUFFER_SIZE); + + /* Decompress directly into user's buffer. */ + if (!(*cinfo->coef->_decompress_data) (cinfo, data)) + return 0; /* suspension forced, can do nothing more */ + + /* OK, we processed one iMCU row. */ + cinfo->output_scanline += lines_per_iMCU_row; + return lines_per_iMCU_row; +} + +#endif /* BITS_IN_JSAMPLE != 16 */ + + +#if BITS_IN_JSAMPLE == 8 + +/* Additional entry points for buffered-image mode. */ + +#ifdef D_MULTISCAN_FILES_SUPPORTED + +/* + * Initialize for an output pass in buffered-image mode. + */ + +GLOBAL(boolean) +jpeg_start_output(j_decompress_ptr cinfo, int scan_number) +{ + if (cinfo->global_state != DSTATE_BUFIMAGE && + cinfo->global_state != DSTATE_PRESCAN) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + /* Limit scan number to valid range */ + if (scan_number <= 0) + scan_number = 1; + if (cinfo->inputctl->eoi_reached && scan_number > cinfo->input_scan_number) + scan_number = cinfo->input_scan_number; + cinfo->output_scan_number = scan_number; + /* Perform any dummy output passes, and set up for the real pass */ + return output_pass_setup(cinfo); +} + + +/* + * Finish up after an output pass in buffered-image mode. + * + * Returns FALSE if suspended. The return value need be inspected only if + * a suspending data source is used. + */ + +GLOBAL(boolean) +jpeg_finish_output(j_decompress_ptr cinfo) +{ + if ((cinfo->global_state == DSTATE_SCANNING || + cinfo->global_state == DSTATE_RAW_OK) && cinfo->buffered_image) { + /* Terminate this pass. */ + /* We do not require the whole pass to have been completed. */ + (*cinfo->master->finish_output_pass) (cinfo); + cinfo->global_state = DSTATE_BUFPOST; + } else if (cinfo->global_state != DSTATE_BUFPOST) { + /* BUFPOST = repeat call after a suspension, anything else is error */ + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + } + /* Read markers looking for SOS or EOI */ + while (cinfo->input_scan_number <= cinfo->output_scan_number && + !cinfo->inputctl->eoi_reached) { + if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED) + return FALSE; /* Suspend, come back later */ + } + cinfo->global_state = DSTATE_BUFIMAGE; + return TRUE; +} + +#endif /* D_MULTISCAN_FILES_SUPPORTED */ + +#endif /* BITS_IN_JSAMPLE == 8 */ diff --git a/src/spice2x/external/libjpeg-turbo/jdarith.c b/src/spice2x/external/libjpeg-turbo/jdarith.c new file mode 100644 index 0000000..21575e8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdarith.c @@ -0,0 +1,782 @@ +/* + * jdarith.c + * + * This file was part of the Independent JPEG Group's software: + * Developed 1997-2015 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2015-2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains portable arithmetic entropy encoding routines for JPEG + * (implementing Recommendation ITU-T T.81 | ISO/IEC 10918-1). + * + * Both sequential and progressive modes are supported in this single module. + * + * Suspension is not currently supported in this module. + * + * NOTE: All referenced figures are from + * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" + + +#define NEG_1 ((unsigned int)-1) + + +/* Expanded entropy decoder object for arithmetic decoding. */ + +typedef struct { + struct jpeg_entropy_decoder pub; /* public fields */ + + JLONG c; /* C register, base of coding interval + input bit buffer */ + JLONG a; /* A register, normalized size of coding interval */ + int ct; /* bit shift counter, # of bits left in bit buffer part of C */ + /* init: ct = -16 */ + /* run: ct = 0..7 */ + /* error: ct = -1 */ + int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */ + int dc_context[MAX_COMPS_IN_SCAN]; /* context index for DC conditioning */ + + unsigned int restarts_to_go; /* MCUs left in this restart interval */ + + /* Pointers to statistics areas (these workspaces have image lifespan) */ + unsigned char *dc_stats[NUM_ARITH_TBLS]; + unsigned char *ac_stats[NUM_ARITH_TBLS]; + + /* Statistics bin for coding with fixed probability 0.5 */ + unsigned char fixed_bin[4]; +} arith_entropy_decoder; + +typedef arith_entropy_decoder *arith_entropy_ptr; + +/* The following two definitions specify the allocation chunk size + * for the statistics area. + * According to sections F.1.4.4.1.3 and F.1.4.4.2, we need at least + * 49 statistics bins for DC, and 245 statistics bins for AC coding. + * + * We use a compact representation with 1 byte per statistics bin, + * thus the numbers directly represent byte sizes. + * This 1 byte per statistics bin contains the meaning of the MPS + * (more probable symbol) in the highest bit (mask 0x80), and the + * index into the probability estimation state machine table + * in the lower bits (mask 0x7F). + */ + +#define DC_STAT_BINS 64 +#define AC_STAT_BINS 256 + + +LOCAL(int) +get_byte(j_decompress_ptr cinfo) +/* Read next input byte; we do not support suspension in this module. */ +{ + struct jpeg_source_mgr *src = cinfo->src; + + if (src->bytes_in_buffer == 0) + if (!(*src->fill_input_buffer) (cinfo)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); + src->bytes_in_buffer--; + return *src->next_input_byte++; +} + + +/* + * The core arithmetic decoding routine (common in JPEG and JBIG). + * This needs to go as fast as possible. + * Machine-dependent optimization facilities + * are not utilized in this portable implementation. + * However, this code should be fairly efficient and + * may be a good base for further optimizations anyway. + * + * Return value is 0 or 1 (binary decision). + * + * Note: I've changed the handling of the code base & bit + * buffer register C compared to other implementations + * based on the standards layout & procedures. + * While it also contains both the actual base of the + * coding interval (16 bits) and the next-bits buffer, + * the cut-point between these two parts is floating + * (instead of fixed) with the bit shift counter CT. + * Thus, we also need only one (variable instead of + * fixed size) shift for the LPS/MPS decision, and + * we can do away with any renormalization update + * of C (except for new data insertion, of course). + * + * I've also introduced a new scheme for accessing + * the probability estimation state machine table, + * derived from Markus Kuhn's JBIG implementation. + */ + +LOCAL(int) +arith_decode(j_decompress_ptr cinfo, unsigned char *st) +{ + register arith_entropy_ptr e = (arith_entropy_ptr)cinfo->entropy; + register unsigned char nl, nm; + register JLONG qe, temp; + register int sv, data; + + /* Renormalization & data input per section D.2.6 */ + while (e->a < 0x8000L) { + if (--e->ct < 0) { + /* Need to fetch next data byte */ + if (cinfo->unread_marker) + data = 0; /* stuff zero data */ + else { + data = get_byte(cinfo); /* read next input byte */ + if (data == 0xFF) { /* zero stuff or marker code */ + do data = get_byte(cinfo); + while (data == 0xFF); /* swallow extra 0xFF bytes */ + if (data == 0) + data = 0xFF; /* discard stuffed zero byte */ + else { + /* Note: Different from the Huffman decoder, hitting + * a marker while processing the compressed data + * segment is legal in arithmetic coding. + * The convention is to supply zero data + * then until decoding is complete. + */ + cinfo->unread_marker = data; + data = 0; + } + } + } + e->c = (e->c << 8) | data; /* insert data into C register */ + if ((e->ct += 8) < 0) /* update bit shift counter */ + /* Need more initial bytes */ + if (++e->ct == 0) + /* Got 2 initial bytes -> re-init A and exit loop */ + e->a = 0x8000L; /* => e->a = 0x10000L after loop exit */ + } + e->a <<= 1; + } + + /* Fetch values from our compact representation of Table D.2: + * Qe values and probability estimation state machine + */ + sv = *st; + qe = jpeg_aritab[sv & 0x7F]; /* => Qe_Value */ + nl = qe & 0xFF; qe >>= 8; /* Next_Index_LPS + Switch_MPS */ + nm = qe & 0xFF; qe >>= 8; /* Next_Index_MPS */ + + /* Decode & estimation procedures per sections D.2.4 & D.2.5 */ + temp = e->a - qe; + e->a = temp; + temp <<= e->ct; + if (e->c >= temp) { + e->c -= temp; + /* Conditional LPS (less probable symbol) exchange */ + if (e->a < qe) { + e->a = qe; + *st = (sv & 0x80) ^ nm; /* Estimate_after_MPS */ + } else { + e->a = qe; + *st = (sv & 0x80) ^ nl; /* Estimate_after_LPS */ + sv ^= 0x80; /* Exchange LPS/MPS */ + } + } else if (e->a < 0x8000L) { + /* Conditional MPS (more probable symbol) exchange */ + if (e->a < qe) { + *st = (sv & 0x80) ^ nl; /* Estimate_after_LPS */ + sv ^= 0x80; /* Exchange LPS/MPS */ + } else { + *st = (sv & 0x80) ^ nm; /* Estimate_after_MPS */ + } + } + + return sv >> 7; +} + + +/* + * Check for a restart marker & resynchronize decoder. + */ + +LOCAL(void) +process_restart(j_decompress_ptr cinfo) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + int ci; + jpeg_component_info *compptr; + + /* Advance past the RSTn marker */ + if (!(*cinfo->marker->read_restart_marker) (cinfo)) + ERREXIT(cinfo, JERR_CANT_SUSPEND); + + /* Re-initialize statistics areas */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + if (!cinfo->progressive_mode || (cinfo->Ss == 0 && cinfo->Ah == 0)) { + memset(entropy->dc_stats[compptr->dc_tbl_no], 0, DC_STAT_BINS); + /* Reset DC predictions to 0 */ + entropy->last_dc_val[ci] = 0; + entropy->dc_context[ci] = 0; + } + if (!cinfo->progressive_mode || cinfo->Ss) { + memset(entropy->ac_stats[compptr->ac_tbl_no], 0, AC_STAT_BINS); + } + } + + /* Reset arithmetic decoding variables */ + entropy->c = 0; + entropy->a = 0; + entropy->ct = -16; /* force reading 2 initial bytes to fill C */ + + /* Reset restart counter */ + entropy->restarts_to_go = cinfo->restart_interval; +} + + +/* + * Arithmetic MCU decoding. + * Each of these routines decodes and returns one MCU's worth of + * arithmetic-compressed coefficients. + * The coefficients are reordered from zigzag order into natural array order, + * but are not dequantized. + * + * The i'th block of the MCU is stored into the block pointed to by + * MCU_data[i]. WE ASSUME THIS AREA IS INITIALLY ZEROED BY THE CALLER. + */ + +/* + * MCU decoding for DC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +METHODDEF(boolean) +decode_mcu_DC_first(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + JBLOCKROW block; + unsigned char *st; + int blkn, ci, tbl, sign; + int v, m; + + /* Process restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + process_restart(cinfo); + entropy->restarts_to_go--; + } + + if (entropy->ct == -1) return TRUE; /* if error do nothing */ + + /* Outer loop handles each block in the MCU */ + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data[blkn]; + ci = cinfo->MCU_membership[blkn]; + tbl = cinfo->cur_comp_info[ci]->dc_tbl_no; + + /* Sections F.2.4.1 & F.1.4.4.1: Decoding of DC coefficients */ + + /* Table F.4: Point to statistics bin S0 for DC coefficient coding */ + st = entropy->dc_stats[tbl] + entropy->dc_context[ci]; + + /* Figure F.19: Decode_DC_DIFF */ + if (arith_decode(cinfo, st) == 0) + entropy->dc_context[ci] = 0; + else { + /* Figure F.21: Decoding nonzero value v */ + /* Figure F.22: Decoding the sign of v */ + sign = arith_decode(cinfo, st + 1); + st += 2; st += sign; + /* Figure F.23: Decoding the magnitude category of v */ + if ((m = arith_decode(cinfo, st)) != 0) { + st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */ + while (arith_decode(cinfo, st)) { + if ((m <<= 1) == 0x8000) { + WARNMS(cinfo, JWRN_ARITH_BAD_CODE); + entropy->ct = -1; /* magnitude overflow */ + return TRUE; + } + st += 1; + } + } + /* Section F.1.4.4.1.2: Establish dc_context conditioning category */ + if (m < (int)((1L << cinfo->arith_dc_L[tbl]) >> 1)) + entropy->dc_context[ci] = 0; /* zero diff category */ + else if (m > (int)((1L << cinfo->arith_dc_U[tbl]) >> 1)) + entropy->dc_context[ci] = 12 + (sign * 4); /* large diff category */ + else + entropy->dc_context[ci] = 4 + (sign * 4); /* small diff category */ + v = m; + /* Figure F.24: Decoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + if (arith_decode(cinfo, st)) v |= m; + v += 1; if (sign) v = -v; + entropy->last_dc_val[ci] = (entropy->last_dc_val[ci] + v) & 0xffff; + } + + /* Scale and output the DC coefficient (assumes jpeg_natural_order[0]=0) */ + (*block)[0] = (JCOEF)LEFT_SHIFT(entropy->last_dc_val[ci], cinfo->Al); + } + + return TRUE; +} + + +/* + * MCU decoding for AC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +METHODDEF(boolean) +decode_mcu_AC_first(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + JBLOCKROW block; + unsigned char *st; + int tbl, sign, k; + int v, m; + + /* Process restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + process_restart(cinfo); + entropy->restarts_to_go--; + } + + if (entropy->ct == -1) return TRUE; /* if error do nothing */ + + /* There is always only one block per MCU */ + block = MCU_data[0]; + tbl = cinfo->cur_comp_info[0]->ac_tbl_no; + + /* Sections F.2.4.2 & F.1.4.4.2: Decoding of AC coefficients */ + + /* Figure F.20: Decode_AC_coefficients */ + for (k = cinfo->Ss; k <= cinfo->Se; k++) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + if (arith_decode(cinfo, st)) break; /* EOB flag */ + while (arith_decode(cinfo, st + 1) == 0) { + st += 3; k++; + if (k > cinfo->Se) { + WARNMS(cinfo, JWRN_ARITH_BAD_CODE); + entropy->ct = -1; /* spectral overflow */ + return TRUE; + } + } + /* Figure F.21: Decoding nonzero value v */ + /* Figure F.22: Decoding the sign of v */ + sign = arith_decode(cinfo, entropy->fixed_bin); + st += 2; + /* Figure F.23: Decoding the magnitude category of v */ + if ((m = arith_decode(cinfo, st)) != 0) { + if (arith_decode(cinfo, st)) { + m <<= 1; + st = entropy->ac_stats[tbl] + + (k <= cinfo->arith_ac_K[tbl] ? 189 : 217); + while (arith_decode(cinfo, st)) { + if ((m <<= 1) == 0x8000) { + WARNMS(cinfo, JWRN_ARITH_BAD_CODE); + entropy->ct = -1; /* magnitude overflow */ + return TRUE; + } + st += 1; + } + } + } + v = m; + /* Figure F.24: Decoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + if (arith_decode(cinfo, st)) v |= m; + v += 1; if (sign) v = -v; + /* Scale and output coefficient in natural (dezigzagged) order */ + (*block)[jpeg_natural_order[k]] = (JCOEF)((unsigned)v << cinfo->Al); + } + + return TRUE; +} + + +/* + * MCU decoding for DC successive approximation refinement scan. + */ + +METHODDEF(boolean) +decode_mcu_DC_refine(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + unsigned char *st; + int p1, blkn; + + /* Process restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + process_restart(cinfo); + entropy->restarts_to_go--; + } + + st = entropy->fixed_bin; /* use fixed probability estimation */ + p1 = 1 << cinfo->Al; /* 1 in the bit position being coded */ + + /* Outer loop handles each block in the MCU */ + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + /* Encoded data is simply the next bit of the two's-complement DC value */ + if (arith_decode(cinfo, st)) + MCU_data[blkn][0][0] |= p1; + } + + return TRUE; +} + + +/* + * MCU decoding for AC successive approximation refinement scan. + */ + +METHODDEF(boolean) +decode_mcu_AC_refine(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + JBLOCKROW block; + JCOEFPTR thiscoef; + unsigned char *st; + int tbl, k, kex; + int p1, m1; + + /* Process restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + process_restart(cinfo); + entropy->restarts_to_go--; + } + + if (entropy->ct == -1) return TRUE; /* if error do nothing */ + + /* There is always only one block per MCU */ + block = MCU_data[0]; + tbl = cinfo->cur_comp_info[0]->ac_tbl_no; + + p1 = 1 << cinfo->Al; /* 1 in the bit position being coded */ + m1 = (NEG_1) << cinfo->Al; /* -1 in the bit position being coded */ + + /* Establish EOBx (previous stage end-of-block) index */ + for (kex = cinfo->Se; kex > 0; kex--) + if ((*block)[jpeg_natural_order[kex]]) break; + + for (k = cinfo->Ss; k <= cinfo->Se; k++) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + if (k > kex) + if (arith_decode(cinfo, st)) break; /* EOB flag */ + for (;;) { + thiscoef = *block + jpeg_natural_order[k]; + if (*thiscoef) { /* previously nonzero coef */ + if (arith_decode(cinfo, st + 2)) { + if (*thiscoef < 0) + *thiscoef += (JCOEF)m1; + else + *thiscoef += (JCOEF)p1; + } + break; + } + if (arith_decode(cinfo, st + 1)) { /* newly nonzero coef */ + if (arith_decode(cinfo, entropy->fixed_bin)) + *thiscoef = (JCOEF)m1; + else + *thiscoef = (JCOEF)p1; + break; + } + st += 3; k++; + if (k > cinfo->Se) { + WARNMS(cinfo, JWRN_ARITH_BAD_CODE); + entropy->ct = -1; /* spectral overflow */ + return TRUE; + } + } + } + + return TRUE; +} + + +/* + * Decode one MCU's worth of arithmetic-compressed coefficients. + */ + +METHODDEF(boolean) +decode_mcu(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + jpeg_component_info *compptr; + JBLOCKROW block; + unsigned char *st; + int blkn, ci, tbl, sign, k; + int v, m; + + /* Process restart marker if needed */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + process_restart(cinfo); + entropy->restarts_to_go--; + } + + if (entropy->ct == -1) return TRUE; /* if error do nothing */ + + /* Outer loop handles each block in the MCU */ + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data ? MCU_data[blkn] : NULL; + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + + /* Sections F.2.4.1 & F.1.4.4.1: Decoding of DC coefficients */ + + tbl = compptr->dc_tbl_no; + + /* Table F.4: Point to statistics bin S0 for DC coefficient coding */ + st = entropy->dc_stats[tbl] + entropy->dc_context[ci]; + + /* Figure F.19: Decode_DC_DIFF */ + if (arith_decode(cinfo, st) == 0) + entropy->dc_context[ci] = 0; + else { + /* Figure F.21: Decoding nonzero value v */ + /* Figure F.22: Decoding the sign of v */ + sign = arith_decode(cinfo, st + 1); + st += 2; st += sign; + /* Figure F.23: Decoding the magnitude category of v */ + if ((m = arith_decode(cinfo, st)) != 0) { + st = entropy->dc_stats[tbl] + 20; /* Table F.4: X1 = 20 */ + while (arith_decode(cinfo, st)) { + if ((m <<= 1) == 0x8000) { + WARNMS(cinfo, JWRN_ARITH_BAD_CODE); + entropy->ct = -1; /* magnitude overflow */ + return TRUE; + } + st += 1; + } + } + /* Section F.1.4.4.1.2: Establish dc_context conditioning category */ + if (m < (int)((1L << cinfo->arith_dc_L[tbl]) >> 1)) + entropy->dc_context[ci] = 0; /* zero diff category */ + else if (m > (int)((1L << cinfo->arith_dc_U[tbl]) >> 1)) + entropy->dc_context[ci] = 12 + (sign * 4); /* large diff category */ + else + entropy->dc_context[ci] = 4 + (sign * 4); /* small diff category */ + v = m; + /* Figure F.24: Decoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + if (arith_decode(cinfo, st)) v |= m; + v += 1; if (sign) v = -v; + entropy->last_dc_val[ci] = (entropy->last_dc_val[ci] + v) & 0xffff; + } + + if (block) + (*block)[0] = (JCOEF)entropy->last_dc_val[ci]; + + /* Sections F.2.4.2 & F.1.4.4.2: Decoding of AC coefficients */ + + tbl = compptr->ac_tbl_no; + + /* Figure F.20: Decode_AC_coefficients */ + for (k = 1; k <= DCTSIZE2 - 1; k++) { + st = entropy->ac_stats[tbl] + 3 * (k - 1); + if (arith_decode(cinfo, st)) break; /* EOB flag */ + while (arith_decode(cinfo, st + 1) == 0) { + st += 3; k++; + if (k > DCTSIZE2 - 1) { + WARNMS(cinfo, JWRN_ARITH_BAD_CODE); + entropy->ct = -1; /* spectral overflow */ + return TRUE; + } + } + /* Figure F.21: Decoding nonzero value v */ + /* Figure F.22: Decoding the sign of v */ + sign = arith_decode(cinfo, entropy->fixed_bin); + st += 2; + /* Figure F.23: Decoding the magnitude category of v */ + if ((m = arith_decode(cinfo, st)) != 0) { + if (arith_decode(cinfo, st)) { + m <<= 1; + st = entropy->ac_stats[tbl] + + (k <= cinfo->arith_ac_K[tbl] ? 189 : 217); + while (arith_decode(cinfo, st)) { + if ((m <<= 1) == 0x8000) { + WARNMS(cinfo, JWRN_ARITH_BAD_CODE); + entropy->ct = -1; /* magnitude overflow */ + return TRUE; + } + st += 1; + } + } + } + v = m; + /* Figure F.24: Decoding the magnitude bit pattern of v */ + st += 14; + while (m >>= 1) + if (arith_decode(cinfo, st)) v |= m; + v += 1; if (sign) v = -v; + if (block) + (*block)[jpeg_natural_order[k]] = (JCOEF)v; + } + } + + return TRUE; +} + + +/* + * Initialize for an arithmetic-compressed scan. + */ + +METHODDEF(void) +start_pass(j_decompress_ptr cinfo) +{ + arith_entropy_ptr entropy = (arith_entropy_ptr)cinfo->entropy; + int ci, tbl; + jpeg_component_info *compptr; + + if (cinfo->progressive_mode) { + /* Validate progressive scan parameters */ + if (cinfo->Ss == 0) { + if (cinfo->Se != 0) + goto bad; + } else { + /* need not check Ss/Se < 0 since they came from unsigned bytes */ + if (cinfo->Se < cinfo->Ss || cinfo->Se > DCTSIZE2 - 1) + goto bad; + /* AC scans may have only one component */ + if (cinfo->comps_in_scan != 1) + goto bad; + } + if (cinfo->Ah != 0) { + /* Successive approximation refinement scan: must have Al = Ah-1. */ + if (cinfo->Ah - 1 != cinfo->Al) + goto bad; + } + if (cinfo->Al > 13) { /* need not check for < 0 */ +bad: + ERREXIT4(cinfo, JERR_BAD_PROGRESSION, + cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al); + } + /* Update progression status, and verify that scan order is legal. + * Note that inter-scan inconsistencies are treated as warnings + * not fatal errors ... not clear if this is right way to behave. + */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + int coefi, cindex = cinfo->cur_comp_info[ci]->component_index; + int *coef_bit_ptr = &cinfo->coef_bits[cindex][0]; + int *prev_coef_bit_ptr = + &cinfo->coef_bits[cindex + cinfo->num_components][0]; + if (cinfo->Ss && coef_bit_ptr[0] < 0) /* AC without prior DC scan */ + WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0); + for (coefi = MIN(cinfo->Ss, 1); coefi <= MAX(cinfo->Se, 9); coefi++) { + if (cinfo->input_scan_number > 1) + prev_coef_bit_ptr[coefi] = coef_bit_ptr[coefi]; + else + prev_coef_bit_ptr[coefi] = 0; + } + for (coefi = cinfo->Ss; coefi <= cinfo->Se; coefi++) { + int expected = (coef_bit_ptr[coefi] < 0) ? 0 : coef_bit_ptr[coefi]; + if (cinfo->Ah != expected) + WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, coefi); + coef_bit_ptr[coefi] = cinfo->Al; + } + } + /* Select MCU decoding routine */ + if (cinfo->Ah == 0) { + if (cinfo->Ss == 0) + entropy->pub.decode_mcu = decode_mcu_DC_first; + else + entropy->pub.decode_mcu = decode_mcu_AC_first; + } else { + if (cinfo->Ss == 0) + entropy->pub.decode_mcu = decode_mcu_DC_refine; + else + entropy->pub.decode_mcu = decode_mcu_AC_refine; + } + } else { + /* Check that the scan parameters Ss, Se, Ah/Al are OK for sequential JPEG. + * This ought to be an error condition, but we make it a warning. + */ + if (cinfo->Ss != 0 || cinfo->Se != DCTSIZE2 - 1 || + cinfo->Ah != 0 || cinfo->Al != 0) + WARNMS(cinfo, JWRN_NOT_SEQUENTIAL); + /* Select MCU decoding routine */ + entropy->pub.decode_mcu = decode_mcu; + } + + /* Allocate & initialize requested statistics areas */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + if (!cinfo->progressive_mode || (cinfo->Ss == 0 && cinfo->Ah == 0)) { + tbl = compptr->dc_tbl_no; + if (tbl < 0 || tbl >= NUM_ARITH_TBLS) + ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl); + if (entropy->dc_stats[tbl] == NULL) + entropy->dc_stats[tbl] = (unsigned char *)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, DC_STAT_BINS); + memset(entropy->dc_stats[tbl], 0, DC_STAT_BINS); + /* Initialize DC predictions to 0 */ + entropy->last_dc_val[ci] = 0; + entropy->dc_context[ci] = 0; + } + if (!cinfo->progressive_mode || cinfo->Ss) { + tbl = compptr->ac_tbl_no; + if (tbl < 0 || tbl >= NUM_ARITH_TBLS) + ERREXIT1(cinfo, JERR_NO_ARITH_TABLE, tbl); + if (entropy->ac_stats[tbl] == NULL) + entropy->ac_stats[tbl] = (unsigned char *)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, AC_STAT_BINS); + memset(entropy->ac_stats[tbl], 0, AC_STAT_BINS); + } + } + + /* Initialize arithmetic decoding variables */ + entropy->c = 0; + entropy->a = 0; + entropy->ct = -16; /* force reading 2 initial bytes to fill C */ + entropy->pub.insufficient_data = FALSE; + + /* Initialize restart counter */ + entropy->restarts_to_go = cinfo->restart_interval; +} + + +/* + * Module initialization routine for arithmetic entropy decoding. + */ + +GLOBAL(void) +jinit_arith_decoder(j_decompress_ptr cinfo) +{ + arith_entropy_ptr entropy; + int i; + + entropy = (arith_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(arith_entropy_decoder)); + cinfo->entropy = (struct jpeg_entropy_decoder *)entropy; + entropy->pub.start_pass = start_pass; + + /* Mark tables unallocated */ + for (i = 0; i < NUM_ARITH_TBLS; i++) { + entropy->dc_stats[i] = NULL; + entropy->ac_stats[i] = NULL; + } + + /* Initialize index for fixed probability estimation */ + entropy->fixed_bin[0] = 113; + + if (cinfo->progressive_mode) { + /* Create progression status table */ + int *coef_bit_ptr, ci; + cinfo->coef_bits = (int (*)[DCTSIZE2]) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + cinfo->num_components * 2 * DCTSIZE2 * + sizeof(int)); + coef_bit_ptr = &cinfo->coef_bits[0][0]; + for (ci = 0; ci < cinfo->num_components; ci++) + for (i = 0; i < DCTSIZE2; i++) + *coef_bit_ptr++ = -1; + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jdatadst.c b/src/spice2x/external/libjpeg-turbo/jdatadst.c new file mode 100644 index 0000000..529f93b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdatadst.c @@ -0,0 +1,277 @@ +/* + * jdatadst.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * Modified 2009-2012 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2013, 2016, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains compression data destination routines for the case of + * emitting JPEG data to memory or to a file (or any stdio stream). + * While these routines are sufficient for most applications, + * some will want to use a different destination manager. + * IMPORTANT: we assume that fwrite() will correctly transcribe an array of + * JOCTETs into 8-bit-wide elements on external storage. If char is wider + * than 8 bits on your machine, you may need to do some tweaking. + */ + +/* this is not a core library module, so it doesn't define JPEG_INTERNALS */ +#include "jinclude.h" +#include "jpeglib.h" +#include "jerror.h" + + +/* Expanded data destination object for stdio output */ + +typedef struct { + struct jpeg_destination_mgr pub; /* public fields */ + + FILE *outfile; /* target stream */ + JOCTET *buffer; /* start of buffer */ +} my_destination_mgr; + +typedef my_destination_mgr *my_dest_ptr; + +#define OUTPUT_BUF_SIZE 4096 /* choose an efficiently fwrite'able size */ + + +/* Expanded data destination object for memory output */ + +typedef struct { + struct jpeg_destination_mgr pub; /* public fields */ + + unsigned char **outbuffer; /* target buffer */ + unsigned long *outsize; + unsigned char *newbuffer; /* newly allocated buffer */ + JOCTET *buffer; /* start of buffer */ + size_t bufsize; +} my_mem_destination_mgr; + +typedef my_mem_destination_mgr *my_mem_dest_ptr; + + +/* + * Initialize destination --- called by jpeg_start_compress + * before any data is actually written. + */ + +METHODDEF(void) +init_destination(j_compress_ptr cinfo) +{ + my_dest_ptr dest = (my_dest_ptr)cinfo->dest; + + /* Allocate the output buffer --- it will be released when done with image */ + dest->buffer = (JOCTET *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + OUTPUT_BUF_SIZE * sizeof(JOCTET)); + + dest->pub.next_output_byte = dest->buffer; + dest->pub.free_in_buffer = OUTPUT_BUF_SIZE; +} + +METHODDEF(void) +init_mem_destination(j_compress_ptr cinfo) +{ + /* no work necessary here */ +} + + +/* + * Empty the output buffer --- called whenever buffer fills up. + * + * In typical applications, this should write the entire output buffer + * (ignoring the current state of next_output_byte & free_in_buffer), + * reset the pointer & count to the start of the buffer, and return TRUE + * indicating that the buffer has been dumped. + * + * In applications that need to be able to suspend compression due to output + * overrun, a FALSE return indicates that the buffer cannot be emptied now. + * In this situation, the compressor will return to its caller (possibly with + * an indication that it has not accepted all the supplied scanlines). The + * application should resume compression after it has made more room in the + * output buffer. Note that there are substantial restrictions on the use of + * suspension --- see the documentation. + * + * When suspending, the compressor will back up to a convenient restart point + * (typically the start of the current MCU). next_output_byte & free_in_buffer + * indicate where the restart point will be if the current call returns FALSE. + * Data beyond this point will be regenerated after resumption, so do not + * write it out when emptying the buffer externally. + */ + +METHODDEF(boolean) +empty_output_buffer(j_compress_ptr cinfo) +{ + my_dest_ptr dest = (my_dest_ptr)cinfo->dest; + + if (fwrite(dest->buffer, 1, OUTPUT_BUF_SIZE, dest->outfile) != + (size_t)OUTPUT_BUF_SIZE) + ERREXIT(cinfo, JERR_FILE_WRITE); + + dest->pub.next_output_byte = dest->buffer; + dest->pub.free_in_buffer = OUTPUT_BUF_SIZE; + + return TRUE; +} + +METHODDEF(boolean) +empty_mem_output_buffer(j_compress_ptr cinfo) +{ + size_t nextsize; + JOCTET *nextbuffer; + my_mem_dest_ptr dest = (my_mem_dest_ptr)cinfo->dest; + + /* Try to allocate new buffer with double size */ + nextsize = dest->bufsize * 2; + nextbuffer = (JOCTET *)malloc(nextsize); + + if (nextbuffer == NULL) + ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 10); + + memcpy(nextbuffer, dest->buffer, dest->bufsize); + + free(dest->newbuffer); + + dest->newbuffer = nextbuffer; + + dest->pub.next_output_byte = nextbuffer + dest->bufsize; + dest->pub.free_in_buffer = dest->bufsize; + + dest->buffer = nextbuffer; + dest->bufsize = nextsize; + + return TRUE; +} + + +/* + * Terminate destination --- called by jpeg_finish_compress + * after all data has been written. Usually needs to flush buffer. + * + * NB: *not* called by jpeg_abort or jpeg_destroy; surrounding + * application must deal with any cleanup that should happen even + * for error exit. + */ + +METHODDEF(void) +term_destination(j_compress_ptr cinfo) +{ + my_dest_ptr dest = (my_dest_ptr)cinfo->dest; + size_t datacount = OUTPUT_BUF_SIZE - dest->pub.free_in_buffer; + + /* Write any data remaining in the buffer */ + if (datacount > 0) { + if (fwrite(dest->buffer, 1, datacount, dest->outfile) != datacount) + ERREXIT(cinfo, JERR_FILE_WRITE); + } + fflush(dest->outfile); + /* Make sure we wrote the output file OK */ + if (ferror(dest->outfile)) + ERREXIT(cinfo, JERR_FILE_WRITE); +} + +METHODDEF(void) +term_mem_destination(j_compress_ptr cinfo) +{ + my_mem_dest_ptr dest = (my_mem_dest_ptr)cinfo->dest; + + *dest->outbuffer = dest->buffer; + *dest->outsize = (unsigned long)(dest->bufsize - dest->pub.free_in_buffer); +} + + +/* + * Prepare for output to a stdio stream. + * The caller must have already opened the stream, and is responsible + * for closing it after finishing compression. + */ + +GLOBAL(void) +jpeg_stdio_dest(j_compress_ptr cinfo, FILE *outfile) +{ + my_dest_ptr dest; + + /* The destination object is made permanent so that multiple JPEG images + * can be written to the same file without re-executing jpeg_stdio_dest. + */ + if (cinfo->dest == NULL) { /* first time for this JPEG object? */ + cinfo->dest = (struct jpeg_destination_mgr *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(my_destination_mgr)); + } else if (cinfo->dest->init_destination != init_destination) { + /* It is unsafe to reuse the existing destination manager unless it was + * created by this function. Otherwise, there is no guarantee that the + * opaque structure is the right size. Note that we could just create a + * new structure, but the old structure would not be freed until + * jpeg_destroy_compress() was called. + */ + ERREXIT(cinfo, JERR_BUFFER_SIZE); + } + + dest = (my_dest_ptr)cinfo->dest; + dest->pub.init_destination = init_destination; + dest->pub.empty_output_buffer = empty_output_buffer; + dest->pub.term_destination = term_destination; + dest->outfile = outfile; +} + + +/* + * Prepare for output to a memory buffer. + * The caller may supply an own initial buffer with appropriate size. + * Otherwise, or when the actual data output exceeds the given size, + * the library adapts the buffer size as necessary. + * The standard library functions malloc/free are used for allocating + * larger memory, so the buffer is available to the application after + * finishing compression, and then the application is responsible for + * freeing the requested memory. + * Note: An initial buffer supplied by the caller is expected to be + * managed by the application. The library does not free such buffer + * when allocating a larger buffer. + */ + +GLOBAL(void) +jpeg_mem_dest(j_compress_ptr cinfo, unsigned char **outbuffer, + unsigned long *outsize) +{ + my_mem_dest_ptr dest; + + if (outbuffer == NULL || outsize == NULL) /* sanity check */ + ERREXIT(cinfo, JERR_BUFFER_SIZE); + + /* The destination object is made permanent so that multiple JPEG images + * can be written to the same buffer without re-executing jpeg_mem_dest. + */ + if (cinfo->dest == NULL) { /* first time for this JPEG object? */ + cinfo->dest = (struct jpeg_destination_mgr *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(my_mem_destination_mgr)); + } else if (cinfo->dest->init_destination != init_mem_destination) { + /* It is unsafe to reuse the existing destination manager unless it was + * created by this function. + */ + ERREXIT(cinfo, JERR_BUFFER_SIZE); + } + + dest = (my_mem_dest_ptr)cinfo->dest; + dest->pub.init_destination = init_mem_destination; + dest->pub.empty_output_buffer = empty_mem_output_buffer; + dest->pub.term_destination = term_mem_destination; + dest->outbuffer = outbuffer; + dest->outsize = outsize; + dest->newbuffer = NULL; + + if (*outbuffer == NULL || *outsize == 0) { + /* Allocate initial buffer */ + dest->newbuffer = *outbuffer = (unsigned char *)malloc(OUTPUT_BUF_SIZE); + if (dest->newbuffer == NULL) + ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 10); + *outsize = OUTPUT_BUF_SIZE; + } + + dest->pub.next_output_byte = dest->buffer = *outbuffer; + dest->pub.free_in_buffer = dest->bufsize = *outsize; +} diff --git a/src/spice2x/external/libjpeg-turbo/jdatasrc.c b/src/spice2x/external/libjpeg-turbo/jdatasrc.c new file mode 100644 index 0000000..dc135f4 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdatasrc.c @@ -0,0 +1,289 @@ +/* + * jdatasrc.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * Modified 2009-2011 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2013, 2016, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains decompression data source routines for the case of + * reading JPEG data from memory or from a file (or any stdio stream). + * While these routines are sufficient for most applications, + * some will want to use a different source manager. + * IMPORTANT: we assume that fread() will correctly transcribe an array of + * JOCTETs from 8-bit-wide elements on external storage. If char is wider + * than 8 bits on your machine, you may need to do some tweaking. + */ + +/* this is not a core library module, so it doesn't define JPEG_INTERNALS */ +#include "jinclude.h" +#include "jpeglib.h" +#include "jerror.h" + + +/* Expanded data source object for stdio input */ + +typedef struct { + struct jpeg_source_mgr pub; /* public fields */ + + FILE *infile; /* source stream */ + JOCTET *buffer; /* start of buffer */ + boolean start_of_file; /* have we gotten any data yet? */ +} my_source_mgr; + +typedef my_source_mgr *my_src_ptr; + +#define INPUT_BUF_SIZE 4096 /* choose an efficiently fread'able size */ + + +/* + * Initialize source --- called by jpeg_read_header + * before any data is actually read. + */ + +METHODDEF(void) +init_source(j_decompress_ptr cinfo) +{ + my_src_ptr src = (my_src_ptr)cinfo->src; + + /* We reset the empty-input-file flag for each image, + * but we don't clear the input buffer. + * This is correct behavior for reading a series of images from one source. + */ + src->start_of_file = TRUE; +} + +METHODDEF(void) +init_mem_source(j_decompress_ptr cinfo) +{ + /* no work necessary here */ +} + + +/* + * Fill the input buffer --- called whenever buffer is emptied. + * + * In typical applications, this should read fresh data into the buffer + * (ignoring the current state of next_input_byte & bytes_in_buffer), + * reset the pointer & count to the start of the buffer, and return TRUE + * indicating that the buffer has been reloaded. It is not necessary to + * fill the buffer entirely, only to obtain at least one more byte. + * + * There is no such thing as an EOF return. If the end of the file has been + * reached, the routine has a choice of ERREXIT() or inserting fake data into + * the buffer. In most cases, generating a warning message and inserting a + * fake EOI marker is the best course of action --- this will allow the + * decompressor to output however much of the image is there. However, + * the resulting error message is misleading if the real problem is an empty + * input file, so we handle that case specially. + * + * In applications that need to be able to suspend compression due to input + * not being available yet, a FALSE return indicates that no more data can be + * obtained right now, but more may be forthcoming later. In this situation, + * the decompressor will return to its caller (with an indication of the + * number of scanlines it has read, if any). The application should resume + * decompression after it has loaded more data into the input buffer. Note + * that there are substantial restrictions on the use of suspension --- see + * the documentation. + * + * When suspending, the decompressor will back up to a convenient restart point + * (typically the start of the current MCU). next_input_byte & bytes_in_buffer + * indicate where the restart point will be if the current call returns FALSE. + * Data beyond this point must be rescanned after resumption, so move it to + * the front of the buffer rather than discarding it. + */ + +METHODDEF(boolean) +fill_input_buffer(j_decompress_ptr cinfo) +{ + my_src_ptr src = (my_src_ptr)cinfo->src; + size_t nbytes; + + nbytes = fread(src->buffer, 1, INPUT_BUF_SIZE, src->infile); + + if (nbytes <= 0) { + if (src->start_of_file) /* Treat empty input file as fatal error */ + ERREXIT(cinfo, JERR_INPUT_EMPTY); + WARNMS(cinfo, JWRN_JPEG_EOF); + /* Insert a fake EOI marker */ + src->buffer[0] = (JOCTET)0xFF; + src->buffer[1] = (JOCTET)JPEG_EOI; + nbytes = 2; + } + + src->pub.next_input_byte = src->buffer; + src->pub.bytes_in_buffer = nbytes; + src->start_of_file = FALSE; + + return TRUE; +} + +METHODDEF(boolean) +fill_mem_input_buffer(j_decompress_ptr cinfo) +{ + static const JOCTET mybuffer[4] = { + (JOCTET)0xFF, (JOCTET)JPEG_EOI, 0, 0 + }; + + /* The whole JPEG data is expected to reside in the supplied memory + * buffer, so any request for more data beyond the given buffer size + * is treated as an error. + */ + WARNMS(cinfo, JWRN_JPEG_EOF); + + /* Insert a fake EOI marker */ + + cinfo->src->next_input_byte = mybuffer; + cinfo->src->bytes_in_buffer = 2; + + return TRUE; +} + + +/* + * Skip data --- used to skip over a potentially large amount of + * uninteresting data (such as an APPn marker). + * + * Writers of suspendable-input applications must note that skip_input_data + * is not granted the right to give a suspension return. If the skip extends + * beyond the data currently in the buffer, the buffer can be marked empty so + * that the next read will cause a fill_input_buffer call that can suspend. + * Arranging for additional bytes to be discarded before reloading the input + * buffer is the application writer's problem. + */ + +METHODDEF(void) +skip_input_data(j_decompress_ptr cinfo, long num_bytes) +{ + struct jpeg_source_mgr *src = cinfo->src; + + /* Just a dumb implementation for now. Could use fseek() except + * it doesn't work on pipes. Not clear that being smart is worth + * any trouble anyway --- large skips are infrequent. + */ + if (num_bytes > 0) { + while (num_bytes > (long)src->bytes_in_buffer) { + num_bytes -= (long)src->bytes_in_buffer; + (void)(*src->fill_input_buffer) (cinfo); + /* note we assume that fill_input_buffer will never return FALSE, + * so suspension need not be handled. + */ + } + src->next_input_byte += (size_t)num_bytes; + src->bytes_in_buffer -= (size_t)num_bytes; + } +} + + +/* + * An additional method that can be provided by data source modules is the + * resync_to_restart method for error recovery in the presence of RST markers. + * For the moment, this source module just uses the default resync method + * provided by the JPEG library. That method assumes that no backtracking + * is possible. + */ + + +/* + * Terminate source --- called by jpeg_finish_decompress + * after all data has been read. Often a no-op. + * + * NB: *not* called by jpeg_abort or jpeg_destroy; surrounding + * application must deal with any cleanup that should happen even + * for error exit. + */ + +METHODDEF(void) +term_source(j_decompress_ptr cinfo) +{ + /* no work necessary here */ +} + + +/* + * Prepare for input from a stdio stream. + * The caller must have already opened the stream, and is responsible + * for closing it after finishing decompression. + */ + +GLOBAL(void) +jpeg_stdio_src(j_decompress_ptr cinfo, FILE *infile) +{ + my_src_ptr src; + + /* The source object and input buffer are made permanent so that a series + * of JPEG images can be read from the same file by calling jpeg_stdio_src + * only before the first one. (If we discarded the buffer at the end of + * one image, we'd likely lose the start of the next one.) + */ + if (cinfo->src == NULL) { /* first time for this JPEG object? */ + cinfo->src = (struct jpeg_source_mgr *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(my_source_mgr)); + src = (my_src_ptr)cinfo->src; + src->buffer = (JOCTET *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + INPUT_BUF_SIZE * sizeof(JOCTET)); + } else if (cinfo->src->init_source != init_source) { + /* It is unsafe to reuse the existing source manager unless it was created + * by this function. Otherwise, there is no guarantee that the opaque + * structure is the right size. Note that we could just create a new + * structure, but the old structure would not be freed until + * jpeg_destroy_decompress() was called. + */ + ERREXIT(cinfo, JERR_BUFFER_SIZE); + } + + src = (my_src_ptr)cinfo->src; + src->pub.init_source = init_source; + src->pub.fill_input_buffer = fill_input_buffer; + src->pub.skip_input_data = skip_input_data; + src->pub.resync_to_restart = jpeg_resync_to_restart; /* use default method */ + src->pub.term_source = term_source; + src->infile = infile; + src->pub.bytes_in_buffer = 0; /* forces fill_input_buffer on first read */ + src->pub.next_input_byte = NULL; /* until buffer loaded */ +} + + +/* + * Prepare for input from a supplied memory buffer. + * The buffer must contain the whole JPEG data. + */ + +GLOBAL(void) +jpeg_mem_src(j_decompress_ptr cinfo, const unsigned char *inbuffer, + unsigned long insize) +{ + struct jpeg_source_mgr *src; + + if (inbuffer == NULL || insize == 0) /* Treat empty input as fatal error */ + ERREXIT(cinfo, JERR_INPUT_EMPTY); + + /* The source object is made permanent so that a series of JPEG images + * can be read from the same buffer by calling jpeg_mem_src only before + * the first one. + */ + if (cinfo->src == NULL) { /* first time for this JPEG object? */ + cinfo->src = (struct jpeg_source_mgr *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(struct jpeg_source_mgr)); + } else if (cinfo->src->init_source != init_mem_source) { + /* It is unsafe to reuse the existing source manager unless it was created + * by this function. + */ + ERREXIT(cinfo, JERR_BUFFER_SIZE); + } + + src = cinfo->src; + src->init_source = init_mem_source; + src->fill_input_buffer = fill_mem_input_buffer; + src->skip_input_data = skip_input_data; + src->resync_to_restart = jpeg_resync_to_restart; /* use default method */ + src->term_source = term_source; + src->bytes_in_buffer = (size_t)insize; + src->next_input_byte = (const JOCTET *)inbuffer; +} diff --git a/src/spice2x/external/libjpeg-turbo/jdcoefct.c b/src/spice2x/external/libjpeg-turbo/jdcoefct.c new file mode 100644 index 0000000..40ce272 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdcoefct.c @@ -0,0 +1,885 @@ +/* + * jdcoefct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2010, 2015-2016, 2019-2020, 2022-2023, D. R. Commander. + * Copyright (C) 2015, 2020, Google, Inc. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the coefficient buffer controller for decompression. + * This controller is the top level of the lossy JPEG decompressor proper. + * The coefficient buffer lies between entropy decoding and inverse-DCT steps. + * + * In buffered-image mode, this controller is the interface between + * input-oriented processing and output-oriented processing. + * Also, the input side (only) is used when reading a file for transcoding. + */ + +#include "jinclude.h" +#include "jdcoefct.h" +#include "jpegapicomp.h" +#include "jsamplecomp.h" + + +/* Forward declarations */ +METHODDEF(int) decompress_onepass(j_decompress_ptr cinfo, + _JSAMPIMAGE output_buf); +#ifdef D_MULTISCAN_FILES_SUPPORTED +METHODDEF(int) decompress_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf); +#endif +#ifdef BLOCK_SMOOTHING_SUPPORTED +LOCAL(boolean) smoothing_ok(j_decompress_ptr cinfo); +METHODDEF(int) decompress_smooth_data(j_decompress_ptr cinfo, + _JSAMPIMAGE output_buf); +#endif + + +/* + * Initialize for an input processing pass. + */ + +METHODDEF(void) +start_input_pass(j_decompress_ptr cinfo) +{ + cinfo->input_iMCU_row = 0; + start_iMCU_row(cinfo); +} + + +/* + * Initialize for an output processing pass. + */ + +METHODDEF(void) +start_output_pass(j_decompress_ptr cinfo) +{ +#ifdef BLOCK_SMOOTHING_SUPPORTED + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + + /* If multipass, check to see whether to use block smoothing on this pass */ + if (coef->pub.coef_arrays != NULL) { + if (cinfo->do_block_smoothing && smoothing_ok(cinfo)) + coef->pub._decompress_data = decompress_smooth_data; + else + coef->pub._decompress_data = decompress_data; + } +#endif + cinfo->output_iMCU_row = 0; +} + + +/* + * Decompress and return some data in the single-pass case. + * Always attempts to emit one fully interleaved MCU row ("iMCU" row). + * Input and output must run in lockstep since we have only a one-MCU buffer. + * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. + * + * NB: output_buf contains a plane for each component in image, + * which we index according to the component's SOF position. + */ + +METHODDEF(int) +decompress_onepass(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION MCU_col_num; /* index of current MCU within row */ + JDIMENSION last_MCU_col = cinfo->MCUs_per_row - 1; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + int blkn, ci, xindex, yindex, yoffset, useful_width; + _JSAMPARRAY output_ptr; + JDIMENSION start_col, output_col; + jpeg_component_info *compptr; + _inverse_DCT_method_ptr inverse_DCT; + + /* Loop to process as much as one whole iMCU row */ + for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; + yoffset++) { + for (MCU_col_num = coef->MCU_ctr; MCU_col_num <= last_MCU_col; + MCU_col_num++) { + /* Try to fetch an MCU. Entropy decoder expects buffer to be zeroed. */ + jzero_far((void *)coef->MCU_buffer[0], + (size_t)(cinfo->blocks_in_MCU * sizeof(JBLOCK))); + if (!cinfo->entropy->insufficient_data) + cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row; + if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) { + /* Suspension forced; update state counters and exit */ + coef->MCU_vert_offset = yoffset; + coef->MCU_ctr = MCU_col_num; + return JPEG_SUSPENDED; + } + + /* Only perform the IDCT on blocks that are contained within the desired + * cropping region. + */ + if (MCU_col_num >= cinfo->master->first_iMCU_col && + MCU_col_num <= cinfo->master->last_iMCU_col) { + /* Determine where data should go in output_buf and do the IDCT thing. + * We skip dummy blocks at the right and bottom edges (but blkn gets + * incremented past them!). Note the inner loop relies on having + * allocated the MCU_buffer[] blocks sequentially. + */ + blkn = 0; /* index of current DCT block within MCU */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* Don't bother to IDCT an uninteresting component. */ + if (!compptr->component_needed) { + blkn += compptr->MCU_blocks; + continue; + } + inverse_DCT = cinfo->idct->_inverse_DCT[compptr->component_index]; + useful_width = (MCU_col_num < last_MCU_col) ? + compptr->MCU_width : compptr->last_col_width; + output_ptr = output_buf[compptr->component_index] + + yoffset * compptr->_DCT_scaled_size; + start_col = (MCU_col_num - cinfo->master->first_iMCU_col) * + compptr->MCU_sample_width; + for (yindex = 0; yindex < compptr->MCU_height; yindex++) { + if (cinfo->input_iMCU_row < last_iMCU_row || + yoffset + yindex < compptr->last_row_height) { + output_col = start_col; + for (xindex = 0; xindex < useful_width; xindex++) { + (*inverse_DCT) (cinfo, compptr, + (JCOEFPTR)coef->MCU_buffer[blkn + xindex], + output_ptr, output_col); + output_col += compptr->_DCT_scaled_size; + } + } + blkn += compptr->MCU_width; + output_ptr += compptr->_DCT_scaled_size; + } + } + } + } + /* Completed an MCU row, but perhaps not an iMCU row */ + coef->MCU_ctr = 0; + } + /* Completed the iMCU row, advance counters for next one */ + cinfo->output_iMCU_row++; + if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) { + start_iMCU_row(cinfo); + return JPEG_ROW_COMPLETED; + } + /* Completed the scan */ + (*cinfo->inputctl->finish_input_pass) (cinfo); + return JPEG_SCAN_COMPLETED; +} + + +/* + * Dummy consume-input routine for single-pass operation. + */ + +METHODDEF(int) +dummy_consume_data(j_decompress_ptr cinfo) +{ + return JPEG_SUSPENDED; /* Always indicate nothing was done */ +} + + +#ifdef D_MULTISCAN_FILES_SUPPORTED + +/* + * Consume input data and store it in the full-image coefficient buffer. + * We read as much as one fully interleaved MCU row ("iMCU" row) per call, + * ie, v_samp_factor block rows for each component in the scan. + * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. + */ + +METHODDEF(int) +consume_data(j_decompress_ptr cinfo) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION MCU_col_num; /* index of current MCU within row */ + int blkn, ci, xindex, yindex, yoffset; + JDIMENSION start_col; + JBLOCKARRAY buffer[MAX_COMPS_IN_SCAN]; + JBLOCKROW buffer_ptr; + jpeg_component_info *compptr; + + /* Align the virtual buffers for the components used in this scan. */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + buffer[ci] = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[compptr->component_index], + cinfo->input_iMCU_row * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, TRUE); + /* Note: entropy decoder expects buffer to be zeroed, + * but this is handled automatically by the memory manager + * because we requested a pre-zeroed array. + */ + } + + /* Loop to process one whole iMCU row */ + for (yoffset = coef->MCU_vert_offset; yoffset < coef->MCU_rows_per_iMCU_row; + yoffset++) { + for (MCU_col_num = coef->MCU_ctr; MCU_col_num < cinfo->MCUs_per_row; + MCU_col_num++) { + /* Construct list of pointers to DCT blocks belonging to this MCU */ + blkn = 0; /* index of current DCT block within MCU */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + start_col = MCU_col_num * compptr->MCU_width; + for (yindex = 0; yindex < compptr->MCU_height; yindex++) { + buffer_ptr = buffer[ci][yindex + yoffset] + start_col; + for (xindex = 0; xindex < compptr->MCU_width; xindex++) { + coef->MCU_buffer[blkn++] = buffer_ptr++; + } + } + } + if (!cinfo->entropy->insufficient_data) + cinfo->master->last_good_iMCU_row = cinfo->input_iMCU_row; + /* Try to fetch the MCU. */ + if (!(*cinfo->entropy->decode_mcu) (cinfo, coef->MCU_buffer)) { + /* Suspension forced; update state counters and exit */ + coef->MCU_vert_offset = yoffset; + coef->MCU_ctr = MCU_col_num; + return JPEG_SUSPENDED; + } + } + /* Completed an MCU row, but perhaps not an iMCU row */ + coef->MCU_ctr = 0; + } + /* Completed the iMCU row, advance counters for next one */ + if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) { + start_iMCU_row(cinfo); + return JPEG_ROW_COMPLETED; + } + /* Completed the scan */ + (*cinfo->inputctl->finish_input_pass) (cinfo); + return JPEG_SCAN_COMPLETED; +} + + +/* + * Decompress and return some data in the multi-pass case. + * Always attempts to emit one fully interleaved MCU row ("iMCU" row). + * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. + * + * NB: output_buf contains a plane for each component in image. + */ + +METHODDEF(int) +decompress_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + JDIMENSION block_num; + int ci, block_row, block_rows; + JBLOCKARRAY buffer; + JBLOCKROW buffer_ptr; + _JSAMPARRAY output_ptr; + JDIMENSION output_col; + jpeg_component_info *compptr; + _inverse_DCT_method_ptr inverse_DCT; + + /* Force some input to be done if we are getting ahead of the input. */ + while (cinfo->input_scan_number < cinfo->output_scan_number || + (cinfo->input_scan_number == cinfo->output_scan_number && + cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) { + if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED) + return JPEG_SUSPENDED; + } + + /* OK, output from the virtual arrays. */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Don't bother to IDCT an uninteresting component. */ + if (!compptr->component_needed) + continue; + /* Align the virtual buffer for this component. */ + buffer = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[ci], + cinfo->output_iMCU_row * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, FALSE); + /* Count non-dummy DCT block rows in this iMCU row. */ + if (cinfo->output_iMCU_row < last_iMCU_row) + block_rows = compptr->v_samp_factor; + else { + /* NB: can't use last_row_height here; it is input-side-dependent! */ + block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (block_rows == 0) block_rows = compptr->v_samp_factor; + } + inverse_DCT = cinfo->idct->_inverse_DCT[ci]; + output_ptr = output_buf[ci]; + /* Loop over all DCT blocks to be processed. */ + for (block_row = 0; block_row < block_rows; block_row++) { + buffer_ptr = buffer[block_row] + cinfo->master->first_MCU_col[ci]; + output_col = 0; + for (block_num = cinfo->master->first_MCU_col[ci]; + block_num <= cinfo->master->last_MCU_col[ci]; block_num++) { + (*inverse_DCT) (cinfo, compptr, (JCOEFPTR)buffer_ptr, output_ptr, + output_col); + buffer_ptr++; + output_col += compptr->_DCT_scaled_size; + } + output_ptr += compptr->_DCT_scaled_size; + } + } + + if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows) + return JPEG_ROW_COMPLETED; + return JPEG_SCAN_COMPLETED; +} + +#endif /* D_MULTISCAN_FILES_SUPPORTED */ + + +#ifdef BLOCK_SMOOTHING_SUPPORTED + +/* + * This code applies interblock smoothing; the first 9 AC coefficients are + * estimated from the DC values of a DCT block and its 24 neighboring blocks. + * We apply smoothing only for progressive JPEG decoding, and only if + * the coefficients it can estimate are not yet known to full precision. + */ + +/* Natural-order array positions of the first 9 zigzag-order coefficients */ +#define Q01_POS 1 +#define Q10_POS 8 +#define Q20_POS 16 +#define Q11_POS 9 +#define Q02_POS 2 +#define Q03_POS 3 +#define Q12_POS 10 +#define Q21_POS 17 +#define Q30_POS 24 + +/* + * Determine whether block smoothing is applicable and safe. + * We also latch the current states of the coef_bits[] entries for the + * AC coefficients; otherwise, if the input side of the decompressor + * advances into a new scan, we might think the coefficients are known + * more accurately than they really are. + */ + +LOCAL(boolean) +smoothing_ok(j_decompress_ptr cinfo) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + boolean smoothing_useful = FALSE; + int ci, coefi; + jpeg_component_info *compptr; + JQUANT_TBL *qtable; + int *coef_bits, *prev_coef_bits; + int *coef_bits_latch, *prev_coef_bits_latch; + + if (!cinfo->progressive_mode || cinfo->coef_bits == NULL) + return FALSE; + + /* Allocate latch area if not already done */ + if (coef->coef_bits_latch == NULL) + coef->coef_bits_latch = (int *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + cinfo->num_components * 2 * + (SAVED_COEFS * sizeof(int))); + coef_bits_latch = coef->coef_bits_latch; + prev_coef_bits_latch = + &coef->coef_bits_latch[cinfo->num_components * SAVED_COEFS]; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* All components' quantization values must already be latched. */ + if ((qtable = compptr->quant_table) == NULL) + return FALSE; + /* Verify DC & first 9 AC quantizers are nonzero to avoid zero-divide. */ + if (qtable->quantval[0] == 0 || + qtable->quantval[Q01_POS] == 0 || + qtable->quantval[Q10_POS] == 0 || + qtable->quantval[Q20_POS] == 0 || + qtable->quantval[Q11_POS] == 0 || + qtable->quantval[Q02_POS] == 0 || + qtable->quantval[Q03_POS] == 0 || + qtable->quantval[Q12_POS] == 0 || + qtable->quantval[Q21_POS] == 0 || + qtable->quantval[Q30_POS] == 0) + return FALSE; + /* DC values must be at least partly known for all components. */ + coef_bits = cinfo->coef_bits[ci]; + prev_coef_bits = cinfo->coef_bits[ci + cinfo->num_components]; + if (coef_bits[0] < 0) + return FALSE; + coef_bits_latch[0] = coef_bits[0]; + /* Block smoothing is helpful if some AC coefficients remain inaccurate. */ + for (coefi = 1; coefi < SAVED_COEFS; coefi++) { + if (cinfo->input_scan_number > 1) + prev_coef_bits_latch[coefi] = prev_coef_bits[coefi]; + else + prev_coef_bits_latch[coefi] = -1; + coef_bits_latch[coefi] = coef_bits[coefi]; + if (coef_bits[coefi] != 0) + smoothing_useful = TRUE; + } + coef_bits_latch += SAVED_COEFS; + prev_coef_bits_latch += SAVED_COEFS; + } + + return smoothing_useful; +} + + +/* + * Variant of decompress_data for use when doing block smoothing. + */ + +METHODDEF(int) +decompress_smooth_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf) +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + JDIMENSION block_num, last_block_column; + int ci, block_row, block_rows, access_rows, image_block_row, + image_block_rows; + JBLOCKARRAY buffer; + JBLOCKROW buffer_ptr, prev_prev_block_row, prev_block_row; + JBLOCKROW next_block_row, next_next_block_row; + _JSAMPARRAY output_ptr; + JDIMENSION output_col; + jpeg_component_info *compptr; + _inverse_DCT_method_ptr inverse_DCT; + boolean change_dc; + JCOEF *workspace; + int *coef_bits; + JQUANT_TBL *quanttbl; + JLONG Q00, Q01, Q02, Q03 = 0, Q10, Q11, Q12 = 0, Q20, Q21 = 0, Q30 = 0, num; + int DC01, DC02, DC03, DC04, DC05, DC06, DC07, DC08, DC09, DC10, DC11, DC12, + DC13, DC14, DC15, DC16, DC17, DC18, DC19, DC20, DC21, DC22, DC23, DC24, + DC25; + int Al, pred; + + /* Keep a local variable to avoid looking it up more than once */ + workspace = coef->workspace; + + /* Force some input to be done if we are getting ahead of the input. */ + while (cinfo->input_scan_number <= cinfo->output_scan_number && + !cinfo->inputctl->eoi_reached) { + if (cinfo->input_scan_number == cinfo->output_scan_number) { + /* If input is working on current scan, we ordinarily want it to + * have completed the current row. But if input scan is DC, + * we want it to keep two rows ahead so that next two block rows' DC + * values are up to date. + */ + JDIMENSION delta = (cinfo->Ss == 0) ? 2 : 0; + if (cinfo->input_iMCU_row > cinfo->output_iMCU_row + delta) + break; + } + if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED) + return JPEG_SUSPENDED; + } + + /* OK, output from the virtual arrays. */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Don't bother to IDCT an uninteresting component. */ + if (!compptr->component_needed) + continue; + /* Count non-dummy DCT block rows in this iMCU row. */ + if (cinfo->output_iMCU_row + 1 < last_iMCU_row) { + block_rows = compptr->v_samp_factor; + access_rows = block_rows * 3; /* this and next two iMCU rows */ + } else if (cinfo->output_iMCU_row < last_iMCU_row) { + block_rows = compptr->v_samp_factor; + access_rows = block_rows * 2; /* this and next iMCU row */ + } else { + /* NB: can't use last_row_height here; it is input-side-dependent! */ + block_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (block_rows == 0) block_rows = compptr->v_samp_factor; + access_rows = block_rows; /* this iMCU row only */ + } + /* Align the virtual buffer for this component. */ + if (cinfo->output_iMCU_row > 1) { + access_rows += 2 * compptr->v_samp_factor; /* prior two iMCU rows too */ + buffer = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[ci], + (cinfo->output_iMCU_row - 2) * compptr->v_samp_factor, + (JDIMENSION)access_rows, FALSE); + buffer += 2 * compptr->v_samp_factor; /* point to current iMCU row */ + } else if (cinfo->output_iMCU_row > 0) { + access_rows += compptr->v_samp_factor; /* prior iMCU row too */ + buffer = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[ci], + (cinfo->output_iMCU_row - 1) * compptr->v_samp_factor, + (JDIMENSION)access_rows, FALSE); + buffer += compptr->v_samp_factor; /* point to current iMCU row */ + } else { + buffer = (*cinfo->mem->access_virt_barray) + ((j_common_ptr)cinfo, coef->whole_image[ci], + (JDIMENSION)0, (JDIMENSION)access_rows, FALSE); + } + /* Fetch component-dependent info. + * If the current scan is incomplete, then we use the component-dependent + * info from the previous scan. + */ + if (cinfo->output_iMCU_row > cinfo->master->last_good_iMCU_row) + coef_bits = + coef->coef_bits_latch + ((ci + cinfo->num_components) * SAVED_COEFS); + else + coef_bits = coef->coef_bits_latch + (ci * SAVED_COEFS); + + /* We only do DC interpolation if no AC coefficient data is available. */ + change_dc = + coef_bits[1] == -1 && coef_bits[2] == -1 && coef_bits[3] == -1 && + coef_bits[4] == -1 && coef_bits[5] == -1 && coef_bits[6] == -1 && + coef_bits[7] == -1 && coef_bits[8] == -1 && coef_bits[9] == -1; + + quanttbl = compptr->quant_table; + Q00 = quanttbl->quantval[0]; + Q01 = quanttbl->quantval[Q01_POS]; + Q10 = quanttbl->quantval[Q10_POS]; + Q20 = quanttbl->quantval[Q20_POS]; + Q11 = quanttbl->quantval[Q11_POS]; + Q02 = quanttbl->quantval[Q02_POS]; + if (change_dc) { + Q03 = quanttbl->quantval[Q03_POS]; + Q12 = quanttbl->quantval[Q12_POS]; + Q21 = quanttbl->quantval[Q21_POS]; + Q30 = quanttbl->quantval[Q30_POS]; + } + inverse_DCT = cinfo->idct->_inverse_DCT[ci]; + output_ptr = output_buf[ci]; + /* Loop over all DCT blocks to be processed. */ + image_block_rows = block_rows * cinfo->total_iMCU_rows; + for (block_row = 0; block_row < block_rows; block_row++) { + image_block_row = cinfo->output_iMCU_row * block_rows + block_row; + buffer_ptr = buffer[block_row] + cinfo->master->first_MCU_col[ci]; + + if (image_block_row > 0) + prev_block_row = + buffer[block_row - 1] + cinfo->master->first_MCU_col[ci]; + else + prev_block_row = buffer_ptr; + + if (image_block_row > 1) + prev_prev_block_row = + buffer[block_row - 2] + cinfo->master->first_MCU_col[ci]; + else + prev_prev_block_row = prev_block_row; + + if (image_block_row < image_block_rows - 1) + next_block_row = + buffer[block_row + 1] + cinfo->master->first_MCU_col[ci]; + else + next_block_row = buffer_ptr; + + if (image_block_row < image_block_rows - 2) + next_next_block_row = + buffer[block_row + 2] + cinfo->master->first_MCU_col[ci]; + else + next_next_block_row = next_block_row; + + /* We fetch the surrounding DC values using a sliding-register approach. + * Initialize all 25 here so as to do the right thing on narrow pics. + */ + DC01 = DC02 = DC03 = DC04 = DC05 = (int)prev_prev_block_row[0][0]; + DC06 = DC07 = DC08 = DC09 = DC10 = (int)prev_block_row[0][0]; + DC11 = DC12 = DC13 = DC14 = DC15 = (int)buffer_ptr[0][0]; + DC16 = DC17 = DC18 = DC19 = DC20 = (int)next_block_row[0][0]; + DC21 = DC22 = DC23 = DC24 = DC25 = (int)next_next_block_row[0][0]; + output_col = 0; + last_block_column = compptr->width_in_blocks - 1; + for (block_num = cinfo->master->first_MCU_col[ci]; + block_num <= cinfo->master->last_MCU_col[ci]; block_num++) { + /* Fetch current DCT block into workspace so we can modify it. */ + jcopy_block_row(buffer_ptr, (JBLOCKROW)workspace, (JDIMENSION)1); + /* Update DC values */ + if (block_num == cinfo->master->first_MCU_col[ci] && + block_num < last_block_column) { + DC04 = DC05 = (int)prev_prev_block_row[1][0]; + DC09 = DC10 = (int)prev_block_row[1][0]; + DC14 = DC15 = (int)buffer_ptr[1][0]; + DC19 = DC20 = (int)next_block_row[1][0]; + DC24 = DC25 = (int)next_next_block_row[1][0]; + } + if (block_num + 1 < last_block_column) { + DC05 = (int)prev_prev_block_row[2][0]; + DC10 = (int)prev_block_row[2][0]; + DC15 = (int)buffer_ptr[2][0]; + DC20 = (int)next_block_row[2][0]; + DC25 = (int)next_next_block_row[2][0]; + } + /* If DC interpolation is enabled, compute coefficient estimates using + * a Gaussian-like kernel, keeping the averages of the DC values. + * + * If DC interpolation is disabled, compute coefficient estimates using + * an algorithm similar to the one described in Section K.8 of the JPEG + * standard, except applied to a 5x5 window rather than a 3x3 window. + * + * An estimate is applied only if the coefficient is still zero and is + * not known to be fully accurate. + */ + /* AC01 */ + if ((Al = coef_bits[1]) != 0 && workspace[1] == 0) { + num = Q00 * (change_dc ? + (-DC01 - DC02 + DC04 + DC05 - 3 * DC06 + 13 * DC07 - + 13 * DC09 + 3 * DC10 - 3 * DC11 + 38 * DC12 - 38 * DC14 + + 3 * DC15 - 3 * DC16 + 13 * DC17 - 13 * DC19 + 3 * DC20 - + DC21 - DC22 + DC24 + DC25) : + (-7 * DC11 + 50 * DC12 - 50 * DC14 + 7 * DC15)); + if (num >= 0) { + pred = (int)(((Q01 << 7) + num) / (Q01 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q01 << 7) - num) / (Q01 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[1] = (JCOEF)pred; + } + /* AC10 */ + if ((Al = coef_bits[2]) != 0 && workspace[8] == 0) { + num = Q00 * (change_dc ? + (-DC01 - 3 * DC02 - 3 * DC03 - 3 * DC04 - DC05 - DC06 + + 13 * DC07 + 38 * DC08 + 13 * DC09 - DC10 + DC16 - + 13 * DC17 - 38 * DC18 - 13 * DC19 + DC20 + DC21 + + 3 * DC22 + 3 * DC23 + 3 * DC24 + DC25) : + (-7 * DC03 + 50 * DC08 - 50 * DC18 + 7 * DC23)); + if (num >= 0) { + pred = (int)(((Q10 << 7) + num) / (Q10 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q10 << 7) - num) / (Q10 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[8] = (JCOEF)pred; + } + /* AC20 */ + if ((Al = coef_bits[3]) != 0 && workspace[16] == 0) { + num = Q00 * (change_dc ? + (DC03 + 2 * DC07 + 7 * DC08 + 2 * DC09 - 5 * DC12 - 14 * DC13 - + 5 * DC14 + 2 * DC17 + 7 * DC18 + 2 * DC19 + DC23) : + (-DC03 + 13 * DC08 - 24 * DC13 + 13 * DC18 - DC23)); + if (num >= 0) { + pred = (int)(((Q20 << 7) + num) / (Q20 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q20 << 7) - num) / (Q20 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[16] = (JCOEF)pred; + } + /* AC11 */ + if ((Al = coef_bits[4]) != 0 && workspace[9] == 0) { + num = Q00 * (change_dc ? + (-DC01 + DC05 + 9 * DC07 - 9 * DC09 - 9 * DC17 + + 9 * DC19 + DC21 - DC25) : + (DC10 + DC16 - 10 * DC17 + 10 * DC19 - DC02 - DC20 + DC22 - + DC24 + DC04 - DC06 + 10 * DC07 - 10 * DC09)); + if (num >= 0) { + pred = (int)(((Q11 << 7) + num) / (Q11 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q11 << 7) - num) / (Q11 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[9] = (JCOEF)pred; + } + /* AC02 */ + if ((Al = coef_bits[5]) != 0 && workspace[2] == 0) { + num = Q00 * (change_dc ? + (2 * DC07 - 5 * DC08 + 2 * DC09 + DC11 + 7 * DC12 - 14 * DC13 + + 7 * DC14 + DC15 + 2 * DC17 - 5 * DC18 + 2 * DC19) : + (-DC11 + 13 * DC12 - 24 * DC13 + 13 * DC14 - DC15)); + if (num >= 0) { + pred = (int)(((Q02 << 7) + num) / (Q02 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q02 << 7) - num) / (Q02 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[2] = (JCOEF)pred; + } + if (change_dc) { + /* AC03 */ + if ((Al = coef_bits[6]) != 0 && workspace[3] == 0) { + num = Q00 * (DC07 - DC09 + 2 * DC12 - 2 * DC14 + DC17 - DC19); + if (num >= 0) { + pred = (int)(((Q03 << 7) + num) / (Q03 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q03 << 7) - num) / (Q03 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[3] = (JCOEF)pred; + } + /* AC12 */ + if ((Al = coef_bits[7]) != 0 && workspace[10] == 0) { + num = Q00 * (DC07 - 3 * DC08 + DC09 - DC17 + 3 * DC18 - DC19); + if (num >= 0) { + pred = (int)(((Q12 << 7) + num) / (Q12 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q12 << 7) - num) / (Q12 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[10] = (JCOEF)pred; + } + /* AC21 */ + if ((Al = coef_bits[8]) != 0 && workspace[17] == 0) { + num = Q00 * (DC07 - DC09 - 3 * DC12 + 3 * DC14 + DC17 - DC19); + if (num >= 0) { + pred = (int)(((Q21 << 7) + num) / (Q21 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q21 << 7) - num) / (Q21 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[17] = (JCOEF)pred; + } + /* AC30 */ + if ((Al = coef_bits[9]) != 0 && workspace[24] == 0) { + num = Q00 * (DC07 + 2 * DC08 + DC09 - DC17 - 2 * DC18 - DC19); + if (num >= 0) { + pred = (int)(((Q30 << 7) + num) / (Q30 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + } else { + pred = (int)(((Q30 << 7) - num) / (Q30 << 8)); + if (Al > 0 && pred >= (1 << Al)) + pred = (1 << Al) - 1; + pred = -pred; + } + workspace[24] = (JCOEF)pred; + } + /* coef_bits[0] is non-negative. Otherwise this function would not + * be called. + */ + num = Q00 * + (-2 * DC01 - 6 * DC02 - 8 * DC03 - 6 * DC04 - 2 * DC05 - + 6 * DC06 + 6 * DC07 + 42 * DC08 + 6 * DC09 - 6 * DC10 - + 8 * DC11 + 42 * DC12 + 152 * DC13 + 42 * DC14 - 8 * DC15 - + 6 * DC16 + 6 * DC17 + 42 * DC18 + 6 * DC19 - 6 * DC20 - + 2 * DC21 - 6 * DC22 - 8 * DC23 - 6 * DC24 - 2 * DC25); + if (num >= 0) { + pred = (int)(((Q00 << 7) + num) / (Q00 << 8)); + } else { + pred = (int)(((Q00 << 7) - num) / (Q00 << 8)); + pred = -pred; + } + workspace[0] = (JCOEF)pred; + } /* change_dc */ + + /* OK, do the IDCT */ + (*inverse_DCT) (cinfo, compptr, (JCOEFPTR)workspace, output_ptr, + output_col); + /* Advance for next column */ + DC01 = DC02; DC02 = DC03; DC03 = DC04; DC04 = DC05; + DC06 = DC07; DC07 = DC08; DC08 = DC09; DC09 = DC10; + DC11 = DC12; DC12 = DC13; DC13 = DC14; DC14 = DC15; + DC16 = DC17; DC17 = DC18; DC18 = DC19; DC19 = DC20; + DC21 = DC22; DC22 = DC23; DC23 = DC24; DC24 = DC25; + buffer_ptr++, prev_block_row++, next_block_row++, + prev_prev_block_row++, next_next_block_row++; + output_col += compptr->_DCT_scaled_size; + } + output_ptr += compptr->_DCT_scaled_size; + } + } + + if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows) + return JPEG_ROW_COMPLETED; + return JPEG_SCAN_COMPLETED; +} + +#endif /* BLOCK_SMOOTHING_SUPPORTED */ + + +/* + * Initialize coefficient buffer controller. + */ + +GLOBAL(void) +_jinit_d_coef_controller(j_decompress_ptr cinfo, boolean need_full_buffer) +{ + my_coef_ptr coef; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + coef = (my_coef_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_coef_controller)); + cinfo->coef = (struct jpeg_d_coef_controller *)coef; + coef->pub.start_input_pass = start_input_pass; + coef->pub.start_output_pass = start_output_pass; +#ifdef BLOCK_SMOOTHING_SUPPORTED + coef->coef_bits_latch = NULL; +#endif + + /* Create the coefficient buffer. */ + if (need_full_buffer) { +#ifdef D_MULTISCAN_FILES_SUPPORTED + /* Allocate a full-image virtual array for each component, */ + /* padded to a multiple of samp_factor DCT blocks in each direction. */ + /* Note we ask for a pre-zeroed array. */ + int ci, access_rows; + jpeg_component_info *compptr; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + access_rows = compptr->v_samp_factor; +#ifdef BLOCK_SMOOTHING_SUPPORTED + /* If block smoothing could be used, need a bigger window */ + if (cinfo->progressive_mode) + access_rows *= 5; +#endif + coef->whole_image[ci] = (*cinfo->mem->request_virt_barray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, TRUE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)jround_up((long)compptr->height_in_blocks, + (long)compptr->v_samp_factor), + (JDIMENSION)access_rows); + } + coef->pub.consume_data = consume_data; + coef->pub._decompress_data = decompress_data; + coef->pub.coef_arrays = coef->whole_image; /* link to virtual arrays */ +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + /* We only need a single-MCU buffer. */ + JBLOCKROW buffer; + int i; + + buffer = (JBLOCKROW) + (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, + D_MAX_BLOCKS_IN_MCU * sizeof(JBLOCK)); + for (i = 0; i < D_MAX_BLOCKS_IN_MCU; i++) { + coef->MCU_buffer[i] = buffer + i; + } + coef->pub.consume_data = dummy_consume_data; + coef->pub._decompress_data = decompress_onepass; + coef->pub.coef_arrays = NULL; /* flag for no virtual arrays */ + } + + /* Allocate the workspace buffer */ + coef->workspace = (JCOEF *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(JCOEF) * DCTSIZE2); +} diff --git a/src/spice2x/external/libjpeg-turbo/jdcoefct.h b/src/spice2x/external/libjpeg-turbo/jdcoefct.h new file mode 100644 index 0000000..bbe9e97 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdcoefct.h @@ -0,0 +1,88 @@ +/* + * jdcoefct.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2020, Google, Inc. + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + */ + +#define JPEG_INTERNALS +#include "jpeglib.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) + +/* Block smoothing is only applicable for progressive JPEG, so: */ +#ifndef D_PROGRESSIVE_SUPPORTED +#undef BLOCK_SMOOTHING_SUPPORTED +#endif + + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_d_coef_controller pub; /* public fields */ + + /* These variables keep track of the current location of the input side. */ + /* cinfo->input_iMCU_row is also used for this. */ + JDIMENSION MCU_ctr; /* counts MCUs processed in current row */ + int MCU_vert_offset; /* counts MCU rows within iMCU row */ + int MCU_rows_per_iMCU_row; /* number of such rows needed */ + + /* The output side's location is represented by cinfo->output_iMCU_row. */ + + /* In single-pass modes, it's sufficient to buffer just one MCU. + * We allocate a workspace of D_MAX_BLOCKS_IN_MCU coefficient blocks, + * and let the entropy decoder write into that workspace each time. + * In multi-pass modes, this array points to the current MCU's blocks + * within the virtual arrays; it is used only by the input side. + */ + JBLOCKROW MCU_buffer[D_MAX_BLOCKS_IN_MCU]; + + /* Temporary workspace for one MCU */ + JCOEF *workspace; + +#ifdef D_MULTISCAN_FILES_SUPPORTED + /* In multi-pass modes, we need a virtual block array for each component. */ + jvirt_barray_ptr whole_image[MAX_COMPONENTS]; +#endif + +#ifdef BLOCK_SMOOTHING_SUPPORTED + /* When doing block smoothing, we latch coefficient Al values here */ + int *coef_bits_latch; +#define SAVED_COEFS 10 /* we save coef_bits[0..9] */ +#endif +} my_coef_controller; + +typedef my_coef_controller *my_coef_ptr; + + +LOCAL(void) +start_iMCU_row(j_decompress_ptr cinfo) +/* Reset within-iMCU-row counters for a new row (input side) */ +{ + my_coef_ptr coef = (my_coef_ptr)cinfo->coef; + + /* In an interleaved scan, an MCU row is the same as an iMCU row. + * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. + * But at the bottom of the image, process only what's left. + */ + if (cinfo->comps_in_scan > 1) { + coef->MCU_rows_per_iMCU_row = 1; + } else { + if (cinfo->input_iMCU_row < (cinfo->total_iMCU_rows - 1)) + coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; + else + coef->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; + } + + coef->MCU_ctr = 0; + coef->MCU_vert_offset = 0; +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jdcol565.c b/src/spice2x/external/libjpeg-turbo/jdcol565.c new file mode 100644 index 0000000..2172d98 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdcol565.c @@ -0,0 +1,392 @@ +/* + * jdcol565.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Modifications: + * Copyright (C) 2013, Linaro Limited. + * Copyright (C) 2014-2015, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains output colorspace conversion routines. + */ + +/* This file is included by jdcolor.c */ + + +INLINE +LOCAL(void) +ycc_rgb565_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int y, cb, cr; + register _JSAMPROW outptr; + register _JSAMPROW inptr0, inptr1, inptr2; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + register int *Crrtab = cconvert->Cr_r_tab; + register int *Cbbtab = cconvert->Cb_b_tab; + register JLONG *Crgtab = cconvert->Cr_g_tab; + register JLONG *Cbgtab = cconvert->Cb_g_tab; + SHIFT_TEMPS + + while (--num_rows >= 0) { + JLONG rgb; + unsigned int r, g, b; + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + + if (PACK_NEED_ALIGNMENT(outptr)) { + y = *inptr0++; + cb = *inptr1++; + cr = *inptr2++; + r = range_limit[y + Crrtab[cr]]; + g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS))]; + b = range_limit[y + Cbbtab[cb]]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + outptr += 2; + num_cols--; + } + for (col = 0; col < (num_cols >> 1); col++) { + y = *inptr0++; + cb = *inptr1++; + cr = *inptr2++; + r = range_limit[y + Crrtab[cr]]; + g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS))]; + b = range_limit[y + Cbbtab[cb]]; + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr0++; + cb = *inptr1++; + cr = *inptr2++; + r = range_limit[y + Crrtab[cr]]; + g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS))]; + b = range_limit[y + Cbbtab[cb]]; + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_ALIGNED_PIXELS(outptr, rgb); + outptr += 4; + } + if (num_cols & 1) { + y = *inptr0; + cb = *inptr1; + cr = *inptr2; + r = range_limit[y + Crrtab[cr]]; + g = range_limit[y + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS))]; + b = range_limit[y + Cbbtab[cb]]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +INLINE +LOCAL(void) +ycc_rgb565D_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int y, cb, cr; + register _JSAMPROW outptr; + register _JSAMPROW inptr0, inptr1, inptr2; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + register int *Crrtab = cconvert->Cr_r_tab; + register int *Cbbtab = cconvert->Cb_b_tab; + register JLONG *Crgtab = cconvert->Cr_g_tab; + register JLONG *Cbgtab = cconvert->Cb_g_tab; + JLONG d0 = dither_matrix[cinfo->output_scanline & DITHER_MASK]; + SHIFT_TEMPS + + while (--num_rows >= 0) { + JLONG rgb; + unsigned int r, g, b; + + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + if (PACK_NEED_ALIGNMENT(outptr)) { + y = *inptr0++; + cb = *inptr1++; + cr = *inptr2++; + r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)]; + g = range_limit[DITHER_565_G(y + + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS)), d0)]; + b = range_limit[DITHER_565_B(y + Cbbtab[cb], d0)]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + outptr += 2; + num_cols--; + } + for (col = 0; col < (num_cols >> 1); col++) { + y = *inptr0++; + cb = *inptr1++; + cr = *inptr2++; + r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)]; + g = range_limit[DITHER_565_G(y + + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS)), d0)]; + b = range_limit[DITHER_565_B(y + Cbbtab[cb], d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr0++; + cb = *inptr1++; + cr = *inptr2++; + r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)]; + g = range_limit[DITHER_565_G(y + + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS)), d0)]; + b = range_limit[DITHER_565_B(y + Cbbtab[cb], d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_ALIGNED_PIXELS(outptr, rgb); + outptr += 4; + } + if (num_cols & 1) { + y = *inptr0; + cb = *inptr1; + cr = *inptr2; + r = range_limit[DITHER_565_R(y + Crrtab[cr], d0)]; + g = range_limit[DITHER_565_G(y + + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS)), d0)]; + b = range_limit[DITHER_565_B(y + Cbbtab[cb], d0)]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +INLINE +LOCAL(void) +rgb_rgb565_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ + register _JSAMPROW outptr; + register _JSAMPROW inptr0, inptr1, inptr2; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + SHIFT_TEMPS + + while (--num_rows >= 0) { + JLONG rgb; + unsigned int r, g, b; + + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + if (PACK_NEED_ALIGNMENT(outptr)) { + r = *inptr0++; + g = *inptr1++; + b = *inptr2++; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + outptr += 2; + num_cols--; + } + for (col = 0; col < (num_cols >> 1); col++) { + r = *inptr0++; + g = *inptr1++; + b = *inptr2++; + rgb = PACK_SHORT_565(r, g, b); + + r = *inptr0++; + g = *inptr1++; + b = *inptr2++; + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_ALIGNED_PIXELS(outptr, rgb); + outptr += 4; + } + if (num_cols & 1) { + r = *inptr0; + g = *inptr1; + b = *inptr2; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + } + } +} + + +INLINE +LOCAL(void) +rgb_rgb565D_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ + register _JSAMPROW outptr; + register _JSAMPROW inptr0, inptr1, inptr2; + register JDIMENSION col; + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + JDIMENSION num_cols = cinfo->output_width; + JLONG d0 = dither_matrix[cinfo->output_scanline & DITHER_MASK]; + SHIFT_TEMPS + + while (--num_rows >= 0) { + JLONG rgb; + unsigned int r, g, b; + + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + if (PACK_NEED_ALIGNMENT(outptr)) { + r = range_limit[DITHER_565_R(*inptr0++, d0)]; + g = range_limit[DITHER_565_G(*inptr1++, d0)]; + b = range_limit[DITHER_565_B(*inptr2++, d0)]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + outptr += 2; + num_cols--; + } + for (col = 0; col < (num_cols >> 1); col++) { + r = range_limit[DITHER_565_R(*inptr0++, d0)]; + g = range_limit[DITHER_565_G(*inptr1++, d0)]; + b = range_limit[DITHER_565_B(*inptr2++, d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_SHORT_565(r, g, b); + + r = range_limit[DITHER_565_R(*inptr0++, d0)]; + g = range_limit[DITHER_565_G(*inptr1++, d0)]; + b = range_limit[DITHER_565_B(*inptr2++, d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_ALIGNED_PIXELS(outptr, rgb); + outptr += 4; + } + if (num_cols & 1) { + r = range_limit[DITHER_565_R(*inptr0, d0)]; + g = range_limit[DITHER_565_G(*inptr1, d0)]; + b = range_limit[DITHER_565_B(*inptr2, d0)]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + } + } +} + + +INLINE +LOCAL(void) +gray_rgb565_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ + register _JSAMPROW inptr, outptr; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + + while (--num_rows >= 0) { + JLONG rgb; + unsigned int g; + + inptr = input_buf[0][input_row++]; + outptr = *output_buf++; + if (PACK_NEED_ALIGNMENT(outptr)) { + g = *inptr++; + rgb = PACK_SHORT_565(g, g, g); + *(INT16 *)outptr = (INT16)rgb; + outptr += 2; + num_cols--; + } + for (col = 0; col < (num_cols >> 1); col++) { + g = *inptr++; + rgb = PACK_SHORT_565(g, g, g); + g = *inptr++; + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(g, g, g)); + WRITE_TWO_ALIGNED_PIXELS(outptr, rgb); + outptr += 4; + } + if (num_cols & 1) { + g = *inptr; + rgb = PACK_SHORT_565(g, g, g); + *(INT16 *)outptr = (INT16)rgb; + } + } +} + + +INLINE +LOCAL(void) +gray_rgb565D_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ + register _JSAMPROW inptr, outptr; + register JDIMENSION col; + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + JDIMENSION num_cols = cinfo->output_width; + JLONG d0 = dither_matrix[cinfo->output_scanline & DITHER_MASK]; + + while (--num_rows >= 0) { + JLONG rgb; + unsigned int g; + + inptr = input_buf[0][input_row++]; + outptr = *output_buf++; + if (PACK_NEED_ALIGNMENT(outptr)) { + g = *inptr++; + g = range_limit[DITHER_565_R(g, d0)]; + rgb = PACK_SHORT_565(g, g, g); + *(INT16 *)outptr = (INT16)rgb; + outptr += 2; + num_cols--; + } + for (col = 0; col < (num_cols >> 1); col++) { + g = *inptr++; + g = range_limit[DITHER_565_R(g, d0)]; + rgb = PACK_SHORT_565(g, g, g); + d0 = DITHER_ROTATE(d0); + + g = *inptr++; + g = range_limit[DITHER_565_R(g, d0)]; + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(g, g, g)); + d0 = DITHER_ROTATE(d0); + + WRITE_TWO_ALIGNED_PIXELS(outptr, rgb); + outptr += 4; + } + if (num_cols & 1) { + g = *inptr; + g = range_limit[DITHER_565_R(g, d0)]; + rgb = PACK_SHORT_565(g, g, g); + *(INT16 *)outptr = (INT16)rgb; + } + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jdcolext.c b/src/spice2x/external/libjpeg-turbo/jdcolext.c new file mode 100644 index 0000000..f22e29d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdcolext.c @@ -0,0 +1,145 @@ +/* + * jdcolext.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2009, 2011, 2015, 2022-2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains output colorspace conversion routines. + */ + + +/* This file is included by jdcolor.c */ + + +/* + * Convert some rows of samples to the output colorspace. + * + * Note that we change from noninterleaved, one-plane-per-component format + * to interleaved-pixel format. The output buffer is therefore three times + * as wide as the input buffer. + * A starting row offset is provided only for the input buffer. The caller + * can easily adjust the passed output_buf value to accommodate any row + * offset required on that side. + */ + +INLINE +LOCAL(void) +ycc_rgb_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int y, cb, cr; + register _JSAMPROW outptr; + register _JSAMPROW inptr0, inptr1, inptr2; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + register int *Crrtab = cconvert->Cr_r_tab; + register int *Cbbtab = cconvert->Cb_b_tab; + register JLONG *Crgtab = cconvert->Cr_g_tab; + register JLONG *Cbgtab = cconvert->Cb_g_tab; + SHIFT_TEMPS + + while (--num_rows >= 0) { + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + for (col = 0; col < num_cols; col++) { + y = inptr0[col]; + cb = inptr1[col]; + cr = inptr2[col]; + /* Range-limiting is essential due to noise introduced by DCT losses. */ + outptr[RGB_RED] = range_limit[y + Crrtab[cr]]; + outptr[RGB_GREEN] = range_limit[y + + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS))]; + outptr[RGB_BLUE] = range_limit[y + Cbbtab[cb]]; + /* Set unused byte to _MAXJSAMPLE so it can be interpreted as an */ + /* opaque alpha channel value */ +#ifdef RGB_ALPHA + outptr[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr += RGB_PIXELSIZE; + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * Convert grayscale to RGB: just duplicate the graylevel three times. + * This is provided to support applications that don't want to cope + * with grayscale as a separate case. + */ + +INLINE +LOCAL(void) +gray_rgb_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ + register _JSAMPROW inptr, outptr; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + + while (--num_rows >= 0) { + inptr = input_buf[0][input_row++]; + outptr = *output_buf++; + for (col = 0; col < num_cols; col++) { + outptr[RGB_RED] = outptr[RGB_GREEN] = outptr[RGB_BLUE] = inptr[col]; + /* Set unused byte to _MAXJSAMPLE so it can be interpreted as an */ + /* opaque alpha channel value */ +#ifdef RGB_ALPHA + outptr[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr += RGB_PIXELSIZE; + } + } +} + + +/* + * Convert RGB to extended RGB: just swap the order of source pixels + */ + +INLINE +LOCAL(void) +rgb_rgb_convert_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, + int num_rows) +{ + register _JSAMPROW inptr0, inptr1, inptr2; + register _JSAMPROW outptr; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + + while (--num_rows >= 0) { + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + for (col = 0; col < num_cols; col++) { + outptr[RGB_RED] = inptr0[col]; + outptr[RGB_GREEN] = inptr1[col]; + outptr[RGB_BLUE] = inptr2[col]; + /* Set unused byte to _MAXJSAMPLE so it can be interpreted as an */ + /* opaque alpha channel value */ +#ifdef RGB_ALPHA + outptr[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr += RGB_PIXELSIZE; + } + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jdcolor.c b/src/spice2x/external/libjpeg-turbo/jdcolor.c new file mode 100644 index 0000000..e5c7b58 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdcolor.c @@ -0,0 +1,925 @@ +/* + * jdcolor.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Modified 2011 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2009, 2011-2012, 2014-2015, 2022, D. R. Commander. + * Copyright (C) 2013, Linaro Limited. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains output colorspace conversion routines. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsimd.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) + +/* Private subobject */ + +typedef struct { + struct jpeg_color_deconverter pub; /* public fields */ + +#if BITS_IN_JSAMPLE != 16 + /* Private state for YCC->RGB conversion */ + int *Cr_r_tab; /* => table for Cr to R conversion */ + int *Cb_b_tab; /* => table for Cb to B conversion */ + JLONG *Cr_g_tab; /* => table for Cr to G conversion */ + JLONG *Cb_g_tab; /* => table for Cb to G conversion */ + + /* Private state for RGB->Y conversion */ + JLONG *rgb_y_tab; /* => table for RGB to Y conversion */ +#endif +} my_color_deconverter; + +typedef my_color_deconverter *my_cconvert_ptr; + + +/**************** YCbCr -> RGB conversion: most common case **************/ +/**************** RGB -> Y conversion: less common case **************/ + +/* + * YCbCr is defined per CCIR 601-1, except that Cb and Cr are + * normalized to the range 0.._MAXJSAMPLE rather than -0.5 .. 0.5. + * The conversion equations to be implemented are therefore + * + * R = Y + 1.40200 * Cr + * G = Y - 0.34414 * Cb - 0.71414 * Cr + * B = Y + 1.77200 * Cb + * + * Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + * + * where Cb and Cr represent the incoming values less _CENTERJSAMPLE. + * (These numbers are derived from TIFF 6.0 section 21, dated 3-June-92.) + * + * To avoid floating-point arithmetic, we represent the fractional constants + * as integers scaled up by 2^16 (about 4 digits precision); we have to divide + * the products by 2^16, with appropriate rounding, to get the correct answer. + * Notice that Y, being an integral input, does not contribute any fraction + * so it need not participate in the rounding. + * + * For even more speed, we avoid doing any multiplications in the inner loop + * by precalculating the constants times Cb and Cr for all possible values. + * For 8-bit samples this is very reasonable (only 256 entries per table); + * for 12-bit samples it is still acceptable. It's not very reasonable for + * 16-bit samples, but if you want lossless storage you shouldn't be changing + * colorspace anyway. + * The Cr=>R and Cb=>B values can be rounded to integers in advance; the + * values for the G calculation are left scaled up, since we must add them + * together before rounding. + */ + +#define SCALEBITS 16 /* speediest right-shift on some machines */ +#define ONE_HALF ((JLONG)1 << (SCALEBITS - 1)) +#define FIX(x) ((JLONG)((x) * (1L << SCALEBITS) + 0.5)) + +/* We allocate one big table for RGB->Y conversion and divide it up into + * three parts, instead of doing three alloc_small requests. This lets us + * use a single table base address, which can be held in a register in the + * inner loops on many machines (more than can hold all three addresses, + * anyway). + */ + +#define R_Y_OFF 0 /* offset to R => Y section */ +#define G_Y_OFF (1 * (_MAXJSAMPLE + 1)) /* offset to G => Y section */ +#define B_Y_OFF (2 * (_MAXJSAMPLE + 1)) /* etc. */ +#define TABLE_SIZE (3 * (_MAXJSAMPLE + 1)) + + +/* Include inline routines for colorspace extensions */ + +#include "jdcolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE + +#define RGB_RED EXT_RGB_RED +#define RGB_GREEN EXT_RGB_GREEN +#define RGB_BLUE EXT_RGB_BLUE +#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +#define ycc_rgb_convert_internal ycc_extrgb_convert_internal +#define gray_rgb_convert_internal gray_extrgb_convert_internal +#define rgb_rgb_convert_internal rgb_extrgb_convert_internal +#include "jdcolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef ycc_rgb_convert_internal +#undef gray_rgb_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_RGBX_RED +#define RGB_GREEN EXT_RGBX_GREEN +#define RGB_BLUE EXT_RGBX_BLUE +#define RGB_ALPHA 3 +#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +#define ycc_rgb_convert_internal ycc_extrgbx_convert_internal +#define gray_rgb_convert_internal gray_extrgbx_convert_internal +#define rgb_rgb_convert_internal rgb_extrgbx_convert_internal +#include "jdcolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef ycc_rgb_convert_internal +#undef gray_rgb_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_BGR_RED +#define RGB_GREEN EXT_BGR_GREEN +#define RGB_BLUE EXT_BGR_BLUE +#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +#define ycc_rgb_convert_internal ycc_extbgr_convert_internal +#define gray_rgb_convert_internal gray_extbgr_convert_internal +#define rgb_rgb_convert_internal rgb_extbgr_convert_internal +#include "jdcolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef ycc_rgb_convert_internal +#undef gray_rgb_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_BGRX_RED +#define RGB_GREEN EXT_BGRX_GREEN +#define RGB_BLUE EXT_BGRX_BLUE +#define RGB_ALPHA 3 +#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +#define ycc_rgb_convert_internal ycc_extbgrx_convert_internal +#define gray_rgb_convert_internal gray_extbgrx_convert_internal +#define rgb_rgb_convert_internal rgb_extbgrx_convert_internal +#include "jdcolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef ycc_rgb_convert_internal +#undef gray_rgb_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_XBGR_RED +#define RGB_GREEN EXT_XBGR_GREEN +#define RGB_BLUE EXT_XBGR_BLUE +#define RGB_ALPHA 0 +#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +#define ycc_rgb_convert_internal ycc_extxbgr_convert_internal +#define gray_rgb_convert_internal gray_extxbgr_convert_internal +#define rgb_rgb_convert_internal rgb_extxbgr_convert_internal +#include "jdcolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef ycc_rgb_convert_internal +#undef gray_rgb_convert_internal +#undef rgb_rgb_convert_internal + +#define RGB_RED EXT_XRGB_RED +#define RGB_GREEN EXT_XRGB_GREEN +#define RGB_BLUE EXT_XRGB_BLUE +#define RGB_ALPHA 0 +#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +#define ycc_rgb_convert_internal ycc_extxrgb_convert_internal +#define gray_rgb_convert_internal gray_extxrgb_convert_internal +#define rgb_rgb_convert_internal rgb_extxrgb_convert_internal +#include "jdcolext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef ycc_rgb_convert_internal +#undef gray_rgb_convert_internal +#undef rgb_rgb_convert_internal + + +/* + * Initialize tables for YCC->RGB colorspace conversion. + */ + +LOCAL(void) +build_ycc_rgb_table(j_decompress_ptr cinfo) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + int i; + JLONG x; + SHIFT_TEMPS + + cconvert->Cr_r_tab = (int *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(int)); + cconvert->Cb_b_tab = (int *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(int)); + cconvert->Cr_g_tab = (JLONG *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(JLONG)); + cconvert->Cb_g_tab = (JLONG *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(JLONG)); + + for (i = 0, x = -_CENTERJSAMPLE; i <= _MAXJSAMPLE; i++, x++) { + /* i is the actual input pixel value, in the range 0.._MAXJSAMPLE */ + /* The Cb or Cr value we are thinking of is x = i - _CENTERJSAMPLE */ + /* Cr=>R value is nearest int to 1.40200 * x */ + cconvert->Cr_r_tab[i] = (int) + RIGHT_SHIFT(FIX(1.40200) * x + ONE_HALF, SCALEBITS); + /* Cb=>B value is nearest int to 1.77200 * x */ + cconvert->Cb_b_tab[i] = (int) + RIGHT_SHIFT(FIX(1.77200) * x + ONE_HALF, SCALEBITS); + /* Cr=>G value is scaled-up -0.71414 * x */ + cconvert->Cr_g_tab[i] = (-FIX(0.71414)) * x; + /* Cb=>G value is scaled-up -0.34414 * x */ + /* We also add in ONE_HALF so that need not do it in inner loop */ + cconvert->Cb_g_tab[i] = (-FIX(0.34414)) * x + ONE_HALF; + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * Convert some rows of samples to the output colorspace. + */ + +METHODDEF(void) +ycc_rgb_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + ycc_extrgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + ycc_extrgbx_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_BGR: + ycc_extbgr_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + ycc_extbgrx_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + ycc_extxbgr_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + ycc_extxrgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + default: + ycc_rgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + } +} + + +/**************** Cases other than YCbCr -> RGB **************/ + + +/* + * Initialize for RGB->grayscale colorspace conversion. + */ + +LOCAL(void) +build_rgb_y_table(j_decompress_ptr cinfo) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + JLONG *rgb_y_tab; + JLONG i; + + /* Allocate and fill in the conversion tables. */ + cconvert->rgb_y_tab = rgb_y_tab = (JLONG *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (TABLE_SIZE * sizeof(JLONG))); + + for (i = 0; i <= _MAXJSAMPLE; i++) { + rgb_y_tab[i + R_Y_OFF] = FIX(0.29900) * i; + rgb_y_tab[i + G_Y_OFF] = FIX(0.58700) * i; + rgb_y_tab[i + B_Y_OFF] = FIX(0.11400) * i + ONE_HALF; + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * Convert RGB to grayscale. + */ + +METHODDEF(void) +rgb_gray_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int r, g, b; + register JLONG *ctab = cconvert->rgb_y_tab; + register _JSAMPROW outptr; + register _JSAMPROW inptr0, inptr1, inptr2; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + + while (--num_rows >= 0) { + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + for (col = 0; col < num_cols; col++) { + r = inptr0[col]; + g = inptr1[col]; + b = inptr2[col]; + /* Y */ + outptr[col] = (_JSAMPLE)((ctab[r + R_Y_OFF] + ctab[g + G_Y_OFF] + + ctab[b + B_Y_OFF]) >> SCALEBITS); + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * Color conversion for no colorspace change: just copy the data, + * converting from separate-planes to interleaved representation. + */ + +METHODDEF(void) +null_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + register _JSAMPROW inptr, inptr0, inptr1, inptr2, inptr3, outptr; + register JDIMENSION col; + register int num_components = cinfo->num_components; + JDIMENSION num_cols = cinfo->output_width; + int ci; + + if (num_components == 3) { + while (--num_rows >= 0) { + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + input_row++; + outptr = *output_buf++; + for (col = 0; col < num_cols; col++) { + *outptr++ = inptr0[col]; + *outptr++ = inptr1[col]; + *outptr++ = inptr2[col]; + } + } + } else if (num_components == 4) { + while (--num_rows >= 0) { + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + inptr3 = input_buf[3][input_row]; + input_row++; + outptr = *output_buf++; + for (col = 0; col < num_cols; col++) { + *outptr++ = inptr0[col]; + *outptr++ = inptr1[col]; + *outptr++ = inptr2[col]; + *outptr++ = inptr3[col]; + } + } + } else { + while (--num_rows >= 0) { + for (ci = 0; ci < num_components; ci++) { + inptr = input_buf[ci][input_row]; + outptr = *output_buf; + for (col = 0; col < num_cols; col++) { + outptr[ci] = inptr[col]; + outptr += num_components; + } + } + output_buf++; + input_row++; + } + } +} + + +/* + * Color conversion for grayscale: just copy the data. + * This also works for YCbCr -> grayscale conversion, in which + * we just copy the Y (luminance) component and ignore chrominance. + */ + +METHODDEF(void) +grayscale_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + _jcopy_sample_rows(input_buf[0], (int)input_row, output_buf, 0, num_rows, + cinfo->output_width); +} + + +/* + * Convert grayscale to RGB + */ + +METHODDEF(void) +gray_rgb_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + gray_extrgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + gray_extrgbx_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_BGR: + gray_extbgr_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + gray_extbgrx_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + gray_extxbgr_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + gray_extxrgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + default: + gray_rgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + } +} + + +/* + * Convert plain RGB to extended RGB + */ + +METHODDEF(void) +rgb_rgb_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + rgb_extrgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + rgb_extrgbx_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_BGR: + rgb_extbgr_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + rgb_extbgrx_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + rgb_extxbgr_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + rgb_extxrgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + default: + rgb_rgb_convert_internal(cinfo, input_buf, input_row, output_buf, + num_rows); + break; + } +} + + +/* + * Adobe-style YCCK->CMYK conversion. + * We convert YCbCr to R=1-C, G=1-M, and B=1-Y using the same + * conversion as above, while passing K (black) unchanged. + * We assume build_ycc_rgb_table has been called. + */ + +METHODDEF(void) +ycck_cmyk_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ +#if BITS_IN_JSAMPLE != 16 + my_cconvert_ptr cconvert = (my_cconvert_ptr)cinfo->cconvert; + register int y, cb, cr; + register _JSAMPROW outptr; + register _JSAMPROW inptr0, inptr1, inptr2, inptr3; + register JDIMENSION col; + JDIMENSION num_cols = cinfo->output_width; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + register int *Crrtab = cconvert->Cr_r_tab; + register int *Cbbtab = cconvert->Cb_b_tab; + register JLONG *Crgtab = cconvert->Cr_g_tab; + register JLONG *Cbgtab = cconvert->Cb_g_tab; + SHIFT_TEMPS + + while (--num_rows >= 0) { + inptr0 = input_buf[0][input_row]; + inptr1 = input_buf[1][input_row]; + inptr2 = input_buf[2][input_row]; + inptr3 = input_buf[3][input_row]; + input_row++; + outptr = *output_buf++; + for (col = 0; col < num_cols; col++) { + y = inptr0[col]; + cb = inptr1[col]; + cr = inptr2[col]; + /* Range-limiting is essential due to noise introduced by DCT losses. */ + outptr[0] = range_limit[_MAXJSAMPLE - (y + Crrtab[cr])]; /* red */ + outptr[1] = range_limit[_MAXJSAMPLE - (y + /* green */ + ((int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], + SCALEBITS)))]; + outptr[2] = range_limit[_MAXJSAMPLE - (y + Cbbtab[cb])]; /* blue */ + /* K passes through unchanged */ + outptr[3] = inptr3[col]; + outptr += 4; + } + } +#else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); +#endif +} + + +/* + * RGB565 conversion + */ + +#define PACK_SHORT_565_LE(r, g, b) \ + ((((r) << 8) & 0xF800) | (((g) << 3) & 0x7E0) | ((b) >> 3)) +#define PACK_SHORT_565_BE(r, g, b) \ + (((r) & 0xF8) | ((g) >> 5) | (((g) << 11) & 0xE000) | (((b) << 5) & 0x1F00)) + +#define PACK_TWO_PIXELS_LE(l, r) ((r << 16) | l) +#define PACK_TWO_PIXELS_BE(l, r) ((l << 16) | r) + +#define PACK_NEED_ALIGNMENT(ptr) (((size_t)(ptr)) & 3) + +#define WRITE_TWO_ALIGNED_PIXELS(addr, pixels) ((*(int *)(addr)) = pixels) + +#define DITHER_565_R(r, dither) ((r) + ((dither) & 0xFF)) +#define DITHER_565_G(g, dither) ((g) + (((dither) & 0xFF) >> 1)) +#define DITHER_565_B(b, dither) ((b) + ((dither) & 0xFF)) + + +/* Declarations for ordered dithering + * + * We use a 4x4 ordered dither array packed into 32 bits. This array is + * sufficient for dithering RGB888 to RGB565. + */ + +#define DITHER_MASK 0x3 +#define DITHER_ROTATE(x) ((((x) & 0xFF) << 24) | (((x) >> 8) & 0x00FFFFFF)) +static const JLONG dither_matrix[4] = { + 0x0008020A, + 0x0C040E06, + 0x030B0109, + 0x0F070D05 +}; + + +static INLINE boolean is_big_endian(void) +{ + int test_value = 1; + if (*(char *)&test_value != 1) + return TRUE; + return FALSE; +} + + +/* Include inline routines for RGB565 conversion */ + +#define PACK_SHORT_565 PACK_SHORT_565_LE +#define PACK_TWO_PIXELS PACK_TWO_PIXELS_LE +#define ycc_rgb565_convert_internal ycc_rgb565_convert_le +#define ycc_rgb565D_convert_internal ycc_rgb565D_convert_le +#define rgb_rgb565_convert_internal rgb_rgb565_convert_le +#define rgb_rgb565D_convert_internal rgb_rgb565D_convert_le +#define gray_rgb565_convert_internal gray_rgb565_convert_le +#define gray_rgb565D_convert_internal gray_rgb565D_convert_le +#include "jdcol565.c" +#undef PACK_SHORT_565 +#undef PACK_TWO_PIXELS +#undef ycc_rgb565_convert_internal +#undef ycc_rgb565D_convert_internal +#undef rgb_rgb565_convert_internal +#undef rgb_rgb565D_convert_internal +#undef gray_rgb565_convert_internal +#undef gray_rgb565D_convert_internal + +#define PACK_SHORT_565 PACK_SHORT_565_BE +#define PACK_TWO_PIXELS PACK_TWO_PIXELS_BE +#define ycc_rgb565_convert_internal ycc_rgb565_convert_be +#define ycc_rgb565D_convert_internal ycc_rgb565D_convert_be +#define rgb_rgb565_convert_internal rgb_rgb565_convert_be +#define rgb_rgb565D_convert_internal rgb_rgb565D_convert_be +#define gray_rgb565_convert_internal gray_rgb565_convert_be +#define gray_rgb565D_convert_internal gray_rgb565D_convert_be +#include "jdcol565.c" +#undef PACK_SHORT_565 +#undef PACK_TWO_PIXELS +#undef ycc_rgb565_convert_internal +#undef ycc_rgb565D_convert_internal +#undef rgb_rgb565_convert_internal +#undef rgb_rgb565D_convert_internal +#undef gray_rgb565_convert_internal +#undef gray_rgb565D_convert_internal + + +METHODDEF(void) +ycc_rgb565_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + if (is_big_endian()) + ycc_rgb565_convert_be(cinfo, input_buf, input_row, output_buf, num_rows); + else + ycc_rgb565_convert_le(cinfo, input_buf, input_row, output_buf, num_rows); +} + + +METHODDEF(void) +ycc_rgb565D_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + if (is_big_endian()) + ycc_rgb565D_convert_be(cinfo, input_buf, input_row, output_buf, num_rows); + else + ycc_rgb565D_convert_le(cinfo, input_buf, input_row, output_buf, num_rows); +} + + +METHODDEF(void) +rgb_rgb565_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + if (is_big_endian()) + rgb_rgb565_convert_be(cinfo, input_buf, input_row, output_buf, num_rows); + else + rgb_rgb565_convert_le(cinfo, input_buf, input_row, output_buf, num_rows); +} + + +METHODDEF(void) +rgb_rgb565D_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + if (is_big_endian()) + rgb_rgb565D_convert_be(cinfo, input_buf, input_row, output_buf, num_rows); + else + rgb_rgb565D_convert_le(cinfo, input_buf, input_row, output_buf, num_rows); +} + + +METHODDEF(void) +gray_rgb565_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + if (is_big_endian()) + gray_rgb565_convert_be(cinfo, input_buf, input_row, output_buf, num_rows); + else + gray_rgb565_convert_le(cinfo, input_buf, input_row, output_buf, num_rows); +} + + +METHODDEF(void) +gray_rgb565D_convert(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION input_row, _JSAMPARRAY output_buf, int num_rows) +{ + if (is_big_endian()) + gray_rgb565D_convert_be(cinfo, input_buf, input_row, output_buf, num_rows); + else + gray_rgb565D_convert_le(cinfo, input_buf, input_row, output_buf, num_rows); +} + + +/* + * Empty method for start_pass. + */ + +METHODDEF(void) +start_pass_dcolor(j_decompress_ptr cinfo) +{ + /* no work needed */ +} + + +/* + * Module initialization routine for output colorspace conversion. + */ + +GLOBAL(void) +_jinit_color_deconverter(j_decompress_ptr cinfo) +{ + my_cconvert_ptr cconvert; + int ci; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + cconvert = (my_cconvert_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_color_deconverter)); + cinfo->cconvert = (struct jpeg_color_deconverter *)cconvert; + cconvert->pub.start_pass = start_pass_dcolor; + + /* Make sure num_components agrees with jpeg_color_space */ + switch (cinfo->jpeg_color_space) { + case JCS_GRAYSCALE: + if (cinfo->num_components != 1) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + break; + + case JCS_RGB: + case JCS_YCbCr: + if (cinfo->num_components != 3) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + break; + + case JCS_CMYK: + case JCS_YCCK: + if (cinfo->num_components != 4) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + break; + + default: /* JCS_UNKNOWN can be anything */ + if (cinfo->num_components < 1) + ERREXIT(cinfo, JERR_BAD_J_COLORSPACE); + break; + } + + /* Set out_color_components and conversion method based on requested space. + * Also clear the component_needed flags for any unused components, + * so that earlier pipeline stages can avoid useless computation. + * NOTE: We do not allow any lossy color conversion algorithms in lossless + * mode. + */ + + switch (cinfo->out_color_space) { + case JCS_GRAYSCALE: + if (cinfo->master->lossless && + cinfo->jpeg_color_space != cinfo->out_color_space) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + cinfo->out_color_components = 1; + if (cinfo->jpeg_color_space == JCS_GRAYSCALE || + cinfo->jpeg_color_space == JCS_YCbCr) { + cconvert->pub._color_convert = grayscale_convert; + /* For color->grayscale conversion, only the Y (0) component is needed */ + for (ci = 1; ci < cinfo->num_components; ci++) + cinfo->comp_info[ci].component_needed = FALSE; + } else if (cinfo->jpeg_color_space == JCS_RGB) { + cconvert->pub._color_convert = rgb_gray_convert; + build_rgb_y_table(cinfo); + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + case JCS_RGB: + case JCS_EXT_RGB: + case JCS_EXT_RGBX: + case JCS_EXT_BGR: + case JCS_EXT_BGRX: + case JCS_EXT_XBGR: + case JCS_EXT_XRGB: + case JCS_EXT_RGBA: + case JCS_EXT_BGRA: + case JCS_EXT_ABGR: + case JCS_EXT_ARGB: + if (cinfo->master->lossless && cinfo->jpeg_color_space != JCS_RGB) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + cinfo->out_color_components = rgb_pixelsize[cinfo->out_color_space]; + if (cinfo->jpeg_color_space == JCS_YCbCr) { +#ifdef WITH_SIMD + if (jsimd_can_ycc_rgb()) + cconvert->pub._color_convert = jsimd_ycc_rgb_convert; + else +#endif + { + cconvert->pub._color_convert = ycc_rgb_convert; + build_ycc_rgb_table(cinfo); + } + } else if (cinfo->jpeg_color_space == JCS_GRAYSCALE) { + cconvert->pub._color_convert = gray_rgb_convert; + } else if (cinfo->jpeg_color_space == JCS_RGB) { + if (rgb_red[cinfo->out_color_space] == 0 && + rgb_green[cinfo->out_color_space] == 1 && + rgb_blue[cinfo->out_color_space] == 2 && + rgb_pixelsize[cinfo->out_color_space] == 3) + cconvert->pub._color_convert = null_convert; + else + cconvert->pub._color_convert = rgb_rgb_convert; + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + case JCS_RGB565: + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + cinfo->out_color_components = 3; + if (cinfo->dither_mode == JDITHER_NONE) { + if (cinfo->jpeg_color_space == JCS_YCbCr) { +#ifdef WITH_SIMD + if (jsimd_can_ycc_rgb565()) + cconvert->pub._color_convert = jsimd_ycc_rgb565_convert; + else +#endif + { + cconvert->pub._color_convert = ycc_rgb565_convert; + build_ycc_rgb_table(cinfo); + } + } else if (cinfo->jpeg_color_space == JCS_GRAYSCALE) { + cconvert->pub._color_convert = gray_rgb565_convert; + } else if (cinfo->jpeg_color_space == JCS_RGB) { + cconvert->pub._color_convert = rgb_rgb565_convert; + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + } else { + /* only ordered dithering is supported */ + if (cinfo->jpeg_color_space == JCS_YCbCr) { + cconvert->pub._color_convert = ycc_rgb565D_convert; + build_ycc_rgb_table(cinfo); + } else if (cinfo->jpeg_color_space == JCS_GRAYSCALE) { + cconvert->pub._color_convert = gray_rgb565D_convert; + } else if (cinfo->jpeg_color_space == JCS_RGB) { + cconvert->pub._color_convert = rgb_rgb565D_convert; + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + } + break; + + case JCS_CMYK: + if (cinfo->master->lossless && + cinfo->jpeg_color_space != cinfo->out_color_space) + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + cinfo->out_color_components = 4; + if (cinfo->jpeg_color_space == JCS_YCCK) { + cconvert->pub._color_convert = ycck_cmyk_convert; + build_ycc_rgb_table(cinfo); + } else if (cinfo->jpeg_color_space == JCS_CMYK) { + cconvert->pub._color_convert = null_convert; + } else + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + + default: + /* Permit null conversion to same output space */ + if (cinfo->out_color_space == cinfo->jpeg_color_space) { + cinfo->out_color_components = cinfo->num_components; + cconvert->pub._color_convert = null_convert; + } else /* unsupported non-null conversion */ + ERREXIT(cinfo, JERR_CONVERSION_NOTIMPL); + break; + } + + if (cinfo->quantize_colors) + cinfo->output_components = 1; /* single colormapped output component */ + else + cinfo->output_components = cinfo->out_color_components; +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jdct.h b/src/spice2x/external/libjpeg-turbo/jdct.h new file mode 100644 index 0000000..0411a79 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdct.h @@ -0,0 +1,221 @@ +/* + * jdct.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This include file contains common declarations for the forward and + * inverse DCT modules. These declarations are private to the DCT managers + * (jcdctmgr.c, jddctmgr.c) and the individual DCT algorithms. + * The individual DCT algorithms are kept in separate files to ease + * machine-dependent tuning (e.g., assembly coding). + */ + +#include "jsamplecomp.h" + + +/* + * A forward DCT routine is given a pointer to a work area of type DCTELEM[]; + * the DCT is to be performed in-place in that buffer. Type DCTELEM is int + * for 8-bit samples, JLONG for 12-bit samples. (NOTE: Floating-point DCT + * implementations use an array of type FAST_FLOAT, instead.) + * The DCT inputs are expected to be signed (range +-_CENTERJSAMPLE). + * The DCT outputs are returned scaled up by a factor of 8; they therefore + * have a range of +-8K for 8-bit data, +-128K for 12-bit data. This + * convention improves accuracy in integer implementations and saves some + * work in floating-point ones. + * Quantization of the output coefficients is done by jcdctmgr.c. This + * step requires an unsigned type and also one with twice the bits. + */ + +#if BITS_IN_JSAMPLE == 8 +#ifndef WITH_SIMD +typedef int DCTELEM; /* 16 or 32 bits is fine */ +typedef unsigned int UDCTELEM; +typedef unsigned long long UDCTELEM2; +#else +typedef short DCTELEM; /* prefer 16 bit with SIMD for parellelism */ +typedef unsigned short UDCTELEM; +typedef unsigned int UDCTELEM2; +#endif +#else +typedef JLONG DCTELEM; /* must have 32 bits */ +typedef unsigned long long UDCTELEM2; +#endif + + +/* + * An inverse DCT routine is given a pointer to the input JBLOCK and a pointer + * to an output sample array. The routine must dequantize the input data as + * well as perform the IDCT; for dequantization, it uses the multiplier table + * pointed to by compptr->dct_table. The output data is to be placed into the + * sample array starting at a specified column. (Any row offset needed will + * be applied to the array pointer before it is passed to the IDCT code.) + * Note that the number of samples emitted by the IDCT routine is + * DCT_scaled_size * DCT_scaled_size. + */ + +/* typedef inverse_DCT_method_ptr is declared in jpegint.h */ + +/* + * Each IDCT routine has its own ideas about the best dct_table element type. + */ + +typedef MULTIPLIER ISLOW_MULT_TYPE; /* short or int, whichever is faster */ +#if BITS_IN_JSAMPLE == 8 +typedef MULTIPLIER IFAST_MULT_TYPE; /* 16 bits is OK, use short if faster */ +#define IFAST_SCALE_BITS 2 /* fractional bits in scale factors */ +#else +typedef JLONG IFAST_MULT_TYPE; /* need 32 bits for scaled quantizers */ +#define IFAST_SCALE_BITS 13 /* fractional bits in scale factors */ +#endif +typedef FAST_FLOAT FLOAT_MULT_TYPE; /* preferred floating type */ + + +/* + * Each IDCT routine is responsible for range-limiting its results and + * converting them to unsigned form (0.._MAXJSAMPLE). The raw outputs could + * be quite far out of range if the input data is corrupt, so a bulletproof + * range-limiting step is required. We use a mask-and-table-lookup method + * to do the combined operations quickly. See the comments with + * prepare_range_limit_table (in jdmaster.c) for more info. + */ + +#define IDCT_range_limit(cinfo) \ + ((_JSAMPLE *)((cinfo)->sample_range_limit) + _CENTERJSAMPLE) + +#define RANGE_MASK (_MAXJSAMPLE * 4 + 3) /* 2 bits wider than legal samples */ + + +/* Extern declarations for the forward and inverse DCT routines. */ + +EXTERN(void) _jpeg_fdct_islow(DCTELEM *data); +EXTERN(void) _jpeg_fdct_ifast(DCTELEM *data); +EXTERN(void) jpeg_fdct_float(FAST_FLOAT *data); + +EXTERN(void) _jpeg_idct_islow(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_ifast(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_float(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_7x7(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_6x6(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_5x5(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_4x4(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_3x3(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_2x2(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_1x1(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_9x9(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + _JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) _jpeg_idct_10x10(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_11x11(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_12x12(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_13x13(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_14x14(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_15x15(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) _jpeg_idct_16x16(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col); + + +/* + * Macros for handling fixed-point arithmetic; these are used by many + * but not all of the DCT/IDCT modules. + * + * All values are expected to be of type JLONG. + * Fractional constants are scaled left by CONST_BITS bits. + * CONST_BITS is defined within each module using these macros, + * and may differ from one module to the next. + */ + +#define ONE ((JLONG)1) +#define CONST_SCALE (ONE << CONST_BITS) + +/* Convert a positive real constant to an integer scaled by CONST_SCALE. + * Caution: some C compilers fail to reduce "FIX(constant)" at compile time, + * thus causing a lot of useless floating-point operations at run time. + */ + +#define FIX(x) ((JLONG)((x) * CONST_SCALE + 0.5)) + +/* Descale and correctly round a JLONG value that's scaled by N bits. + * We assume RIGHT_SHIFT rounds towards minus infinity, so adding + * the fudge factor is correct for either sign of X. + */ + +#define DESCALE(x, n) RIGHT_SHIFT((x) + (ONE << ((n) - 1)), n) + +/* Multiply a JLONG variable by a JLONG constant to yield a JLONG result. + * This macro is used only when the two inputs will actually be no more than + * 16 bits wide, so that a 16x16->32 bit multiply can be used instead of a + * full 32x32 multiply. This provides a useful speedup on many machines. + * Unfortunately there is no way to specify a 16x16->32 multiply portably + * in C, but some C compilers will do the right thing if you provide the + * correct combination of casts. + */ + +#ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ +#define MULTIPLY16C16(var, const) (((INT16)(var)) * ((INT16)(const))) +#endif +#ifdef SHORTxLCONST_32 /* known to work with Microsoft C 6.0 */ +#define MULTIPLY16C16(var, const) (((INT16)(var)) * ((JLONG)(const))) +#endif + +#ifndef MULTIPLY16C16 /* default definition */ +#define MULTIPLY16C16(var, const) ((var) * (const)) +#endif + +/* Same except both inputs are variables. */ + +#ifdef SHORTxSHORT_32 /* may work if 'int' is 32 bits */ +#define MULTIPLY16V16(var1, var2) (((INT16)(var1)) * ((INT16)(var2))) +#endif + +#ifndef MULTIPLY16V16 /* default definition */ +#define MULTIPLY16V16(var1, var2) ((var1) * (var2)) +#endif diff --git a/src/spice2x/external/libjpeg-turbo/jddctmgr.c b/src/spice2x/external/libjpeg-turbo/jddctmgr.c new file mode 100644 index 0000000..0bd8c2b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jddctmgr.c @@ -0,0 +1,365 @@ +/* + * jddctmgr.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * Modified 2002-2010 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2010, 2015, 2022, D. R. Commander. + * Copyright (C) 2013, MIPS Technologies, Inc., California. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the inverse-DCT management logic. + * This code selects a particular IDCT implementation to be used, + * and it performs related housekeeping chores. No code in this file + * is executed per IDCT step, only during output pass setup. + * + * Note that the IDCT routines are responsible for performing coefficient + * dequantization as well as the IDCT proper. This module sets up the + * dequantization multiplier table needed by the IDCT routine. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ +#include "jsimddct.h" +#include "jpegapicomp.h" + + +/* + * The decompressor input side (jdinput.c) saves away the appropriate + * quantization table for each component at the start of the first scan + * involving that component. (This is necessary in order to correctly + * decode files that reuse Q-table slots.) + * When we are ready to make an output pass, the saved Q-table is converted + * to a multiplier table that will actually be used by the IDCT routine. + * The multiplier table contents are IDCT-method-dependent. To support + * application changes in IDCT method between scans, we can remake the + * multiplier tables if necessary. + * In buffered-image mode, the first output pass may occur before any data + * has been seen for some components, and thus before their Q-tables have + * been saved away. To handle this case, multiplier tables are preset + * to zeroes; the result of the IDCT will be a neutral gray level. + */ + + +/* Private subobject for this module */ + +typedef struct { + struct jpeg_inverse_dct pub; /* public fields */ + + /* This array contains the IDCT method code that each multiplier table + * is currently set up for, or -1 if it's not yet set up. + * The actual multiplier tables are pointed to by dct_table in the + * per-component comp_info structures. + */ + int cur_method[MAX_COMPONENTS]; +} my_idct_controller; + +typedef my_idct_controller *my_idct_ptr; + + +/* Allocated multiplier tables: big enough for any supported variant */ + +typedef union { + ISLOW_MULT_TYPE islow_array[DCTSIZE2]; +#ifdef DCT_IFAST_SUPPORTED + IFAST_MULT_TYPE ifast_array[DCTSIZE2]; +#endif +#ifdef DCT_FLOAT_SUPPORTED + FLOAT_MULT_TYPE float_array[DCTSIZE2]; +#endif +} multiplier_table; + + +/* The current scaled-IDCT routines require ISLOW-style multiplier tables, + * so be sure to compile that code if either ISLOW or SCALING is requested. + */ +#ifdef DCT_ISLOW_SUPPORTED +#define PROVIDE_ISLOW_TABLES +#else +#ifdef IDCT_SCALING_SUPPORTED +#define PROVIDE_ISLOW_TABLES +#endif +#endif + + +/* + * Prepare for an output pass. + * Here we select the proper IDCT routine for each component and build + * a matching multiplier table. + */ + +METHODDEF(void) +start_pass(j_decompress_ptr cinfo) +{ + my_idct_ptr idct = (my_idct_ptr)cinfo->idct; + int ci, i; + jpeg_component_info *compptr; + int method = 0; + _inverse_DCT_method_ptr method_ptr = NULL; + JQUANT_TBL *qtbl; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Select the proper IDCT routine for this component's scaling */ + switch (compptr->_DCT_scaled_size) { +#ifdef IDCT_SCALING_SUPPORTED + case 1: + method_ptr = _jpeg_idct_1x1; + method = JDCT_ISLOW; /* jidctred uses islow-style table */ + break; + case 2: +#ifdef WITH_SIMD + if (jsimd_can_idct_2x2()) + method_ptr = jsimd_idct_2x2; + else +#endif + method_ptr = _jpeg_idct_2x2; + method = JDCT_ISLOW; /* jidctred uses islow-style table */ + break; + case 3: + method_ptr = _jpeg_idct_3x3; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 4: +#ifdef WITH_SIMD + if (jsimd_can_idct_4x4()) + method_ptr = jsimd_idct_4x4; + else +#endif + method_ptr = _jpeg_idct_4x4; + method = JDCT_ISLOW; /* jidctred uses islow-style table */ + break; + case 5: + method_ptr = _jpeg_idct_5x5; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 6: +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_can_idct_6x6()) + method_ptr = jsimd_idct_6x6; + else +#endif + method_ptr = _jpeg_idct_6x6; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 7: + method_ptr = _jpeg_idct_7x7; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; +#endif + case DCTSIZE: + switch (cinfo->dct_method) { +#ifdef DCT_ISLOW_SUPPORTED + case JDCT_ISLOW: +#ifdef WITH_SIMD + if (jsimd_can_idct_islow()) + method_ptr = jsimd_idct_islow; + else +#endif + method_ptr = _jpeg_idct_islow; + method = JDCT_ISLOW; + break; +#endif +#ifdef DCT_IFAST_SUPPORTED + case JDCT_IFAST: +#ifdef WITH_SIMD + if (jsimd_can_idct_ifast()) + method_ptr = jsimd_idct_ifast; + else +#endif + method_ptr = _jpeg_idct_ifast; + method = JDCT_IFAST; + break; +#endif +#ifdef DCT_FLOAT_SUPPORTED + case JDCT_FLOAT: +#ifdef WITH_SIMD + if (jsimd_can_idct_float()) + method_ptr = jsimd_idct_float; + else +#endif + method_ptr = _jpeg_idct_float; + method = JDCT_FLOAT; + break; +#endif + default: + ERREXIT(cinfo, JERR_NOT_COMPILED); + break; + } + break; +#ifdef IDCT_SCALING_SUPPORTED + case 9: + method_ptr = _jpeg_idct_9x9; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 10: + method_ptr = _jpeg_idct_10x10; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 11: + method_ptr = _jpeg_idct_11x11; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 12: +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_can_idct_12x12()) + method_ptr = jsimd_idct_12x12; + else +#endif + method_ptr = _jpeg_idct_12x12; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 13: + method_ptr = _jpeg_idct_13x13; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 14: + method_ptr = _jpeg_idct_14x14; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 15: + method_ptr = _jpeg_idct_15x15; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; + case 16: + method_ptr = _jpeg_idct_16x16; + method = JDCT_ISLOW; /* jidctint uses islow-style table */ + break; +#endif + default: + ERREXIT1(cinfo, JERR_BAD_DCTSIZE, compptr->_DCT_scaled_size); + break; + } + idct->pub._inverse_DCT[ci] = method_ptr; + /* Create multiplier table from quant table. + * However, we can skip this if the component is uninteresting + * or if we already built the table. Also, if no quant table + * has yet been saved for the component, we leave the + * multiplier table all-zero; we'll be reading zeroes from the + * coefficient controller's buffer anyway. + */ + if (!compptr->component_needed || idct->cur_method[ci] == method) + continue; + qtbl = compptr->quant_table; + if (qtbl == NULL) /* happens if no data yet for component */ + continue; + idct->cur_method[ci] = method; + switch (method) { +#ifdef PROVIDE_ISLOW_TABLES + case JDCT_ISLOW: + { + /* For LL&M IDCT method, multipliers are equal to raw quantization + * coefficients, but are stored as ints to ensure access efficiency. + */ + ISLOW_MULT_TYPE *ismtbl = (ISLOW_MULT_TYPE *)compptr->dct_table; + for (i = 0; i < DCTSIZE2; i++) { + ismtbl[i] = (ISLOW_MULT_TYPE)qtbl->quantval[i]; + } + } + break; +#endif +#ifdef DCT_IFAST_SUPPORTED + case JDCT_IFAST: + { + /* For AA&N IDCT method, multipliers are equal to quantization + * coefficients scaled by scalefactor[row]*scalefactor[col], where + * scalefactor[0] = 1 + * scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7 + * For integer operation, the multiplier table is to be scaled by + * IFAST_SCALE_BITS. + */ + IFAST_MULT_TYPE *ifmtbl = (IFAST_MULT_TYPE *)compptr->dct_table; +#define CONST_BITS 14 + static const INT16 aanscales[DCTSIZE2] = { + /* precomputed values scaled up by 14 bits */ + 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, + 22725, 31521, 29692, 26722, 22725, 17855, 12299, 6270, + 21407, 29692, 27969, 25172, 21407, 16819, 11585, 5906, + 19266, 26722, 25172, 22654, 19266, 15137, 10426, 5315, + 16384, 22725, 21407, 19266, 16384, 12873, 8867, 4520, + 12873, 17855, 16819, 15137, 12873, 10114, 6967, 3552, + 8867, 12299, 11585, 10426, 8867, 6967, 4799, 2446, + 4520, 6270, 5906, 5315, 4520, 3552, 2446, 1247 + }; + SHIFT_TEMPS + + for (i = 0; i < DCTSIZE2; i++) { + ifmtbl[i] = (IFAST_MULT_TYPE) + DESCALE(MULTIPLY16V16((JLONG)qtbl->quantval[i], + (JLONG)aanscales[i]), + CONST_BITS - IFAST_SCALE_BITS); + } + } + break; +#endif +#ifdef DCT_FLOAT_SUPPORTED + case JDCT_FLOAT: + { + /* For float AA&N IDCT method, multipliers are equal to quantization + * coefficients scaled by scalefactor[row]*scalefactor[col], where + * scalefactor[0] = 1 + * scalefactor[k] = cos(k*PI/16) * sqrt(2) for k=1..7 + */ + FLOAT_MULT_TYPE *fmtbl = (FLOAT_MULT_TYPE *)compptr->dct_table; + int row, col; + static const double aanscalefactor[DCTSIZE] = { + 1.0, 1.387039845, 1.306562965, 1.175875602, + 1.0, 0.785694958, 0.541196100, 0.275899379 + }; + + i = 0; + for (row = 0; row < DCTSIZE; row++) { + for (col = 0; col < DCTSIZE; col++) { + fmtbl[i] = (FLOAT_MULT_TYPE) + ((double)qtbl->quantval[i] * + aanscalefactor[row] * aanscalefactor[col]); + i++; + } + } + } + break; +#endif + default: + ERREXIT(cinfo, JERR_NOT_COMPILED); + break; + } + } +} + + +/* + * Initialize IDCT manager. + */ + +GLOBAL(void) +_jinit_inverse_dct(j_decompress_ptr cinfo) +{ + my_idct_ptr idct; + int ci; + jpeg_component_info *compptr; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + idct = (my_idct_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_idct_controller)); + cinfo->idct = (struct jpeg_inverse_dct *)idct; + idct->pub.start_pass = start_pass; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Allocate and pre-zero a multiplier table for each component */ + compptr->dct_table = + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(multiplier_table)); + memset(compptr->dct_table, 0, sizeof(multiplier_table)); + /* Mark multiplier table not yet set up for any method */ + idct->cur_method[ci] = -1; + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jddiffct.c b/src/spice2x/external/libjpeg-turbo/jddiffct.c new file mode 100644 index 0000000..f1d7f61 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jddiffct.c @@ -0,0 +1,403 @@ +/* + * jddiffct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the [un]difference buffer controller for decompression. + * This controller is the top level of the lossless JPEG decompressor proper. + * The difference buffer lies between the entropy decoding and + * prediction/undifferencing steps. The undifference buffer lies between the + * prediction/undifferencing and scaling steps. + * + * In buffered-image mode, this controller is the interface between + * input-oriented processing and output-oriented processing. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jlossls.h" /* Private declarations for lossless codec */ + + +#ifdef D_LOSSLESS_SUPPORTED + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_d_coef_controller pub; /* public fields */ + + /* These variables keep track of the current location of the input side. */ + /* cinfo->input_iMCU_row is also used for this. */ + JDIMENSION MCU_ctr; /* counts MCUs processed in current row */ + unsigned int restart_rows_to_go; /* MCU rows left in this restart + interval */ + unsigned int MCU_vert_offset; /* counts MCU rows within iMCU row */ + unsigned int MCU_rows_per_iMCU_row; /* number of such rows needed */ + + /* The output side's location is represented by cinfo->output_iMCU_row. */ + + JDIFFARRAY diff_buf[MAX_COMPONENTS]; /* iMCU row of differences */ + JDIFFARRAY undiff_buf[MAX_COMPONENTS]; /* iMCU row of undiff'd samples */ + +#ifdef D_MULTISCAN_FILES_SUPPORTED + /* In multi-pass modes, we need a virtual sample array for each component. */ + jvirt_sarray_ptr whole_image[MAX_COMPONENTS]; +#endif +} my_diff_controller; + +typedef my_diff_controller *my_diff_ptr; + +/* Forward declarations */ +METHODDEF(int) decompress_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf); +#ifdef D_MULTISCAN_FILES_SUPPORTED +METHODDEF(int) output_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf); +#endif + + +LOCAL(void) +start_iMCU_row(j_decompress_ptr cinfo) +/* Reset within-iMCU-row counters for a new row (input side) */ +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + + /* In an interleaved scan, an MCU row is the same as an iMCU row. + * In a noninterleaved scan, an iMCU row has v_samp_factor MCU rows. + * But at the bottom of the image, process only what's left. + */ + if (cinfo->comps_in_scan > 1) { + diff->MCU_rows_per_iMCU_row = 1; + } else { + if (cinfo->input_iMCU_row < (cinfo->total_iMCU_rows-1)) + diff->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->v_samp_factor; + else + diff->MCU_rows_per_iMCU_row = cinfo->cur_comp_info[0]->last_row_height; + } + + diff->MCU_ctr = 0; + diff->MCU_vert_offset = 0; +} + + +/* + * Initialize for an input processing pass. + */ + +METHODDEF(void) +start_input_pass(j_decompress_ptr cinfo) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + + /* Because it is hitching a ride on the jpeg_inverse_dct struct, + * start_pass_lossless() will be called at the start of the output pass. + * This ensures that it will be called at the start of the input pass as + * well. + */ + (*cinfo->idct->start_pass) (cinfo); + + /* Check that the restart interval is an integer multiple of the number + * of MCUs in an MCU row. + */ + if (cinfo->restart_interval % cinfo->MCUs_per_row != 0) + ERREXIT2(cinfo, JERR_BAD_RESTART, + cinfo->restart_interval, cinfo->MCUs_per_row); + + /* Initialize restart counter */ + diff->restart_rows_to_go = cinfo->restart_interval / cinfo->MCUs_per_row; + + cinfo->input_iMCU_row = 0; + start_iMCU_row(cinfo); +} + + +/* + * Check for a restart marker & resynchronize decoder, undifferencer. + * Returns FALSE if must suspend. + */ + +METHODDEF(boolean) +process_restart(j_decompress_ptr cinfo) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + + if (!(*cinfo->entropy->process_restart) (cinfo)) + return FALSE; + + (*cinfo->idct->start_pass) (cinfo); + + /* Reset restart counter */ + diff->restart_rows_to_go = cinfo->restart_interval / cinfo->MCUs_per_row; + + return TRUE; +} + + +/* + * Initialize for an output processing pass. + */ + +METHODDEF(void) +start_output_pass(j_decompress_ptr cinfo) +{ + cinfo->output_iMCU_row = 0; +} + + +/* + * Decompress and return some data in the supplied buffer. + * Always attempts to emit one fully interleaved MCU row ("iMCU" row). + * Input and output must run in lockstep since we have only a one-MCU buffer. + * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. + * + * NB: output_buf contains a plane for each component in image, + * which we index according to the component's SOF position. + */ + +METHODDEF(int) +decompress_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + lossless_decomp_ptr losslessd = (lossless_decomp_ptr)cinfo->idct; + JDIMENSION MCU_col_num; /* index of current MCU within row */ + JDIMENSION MCU_count; /* number of MCUs decoded */ + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + int ci, compi, row, prev_row; + unsigned int yoffset; + jpeg_component_info *compptr; + + /* Loop to process as much as one whole iMCU row */ + for (yoffset = diff->MCU_vert_offset; yoffset < diff->MCU_rows_per_iMCU_row; + yoffset++) { + + /* Process restart marker if needed; may have to suspend */ + if (cinfo->restart_interval) { + if (diff->restart_rows_to_go == 0) + if (!process_restart(cinfo)) + return JPEG_SUSPENDED; + } + + MCU_col_num = diff->MCU_ctr; + /* Try to fetch an MCU row (or remaining portion of suspended MCU row). */ + MCU_count = + (*cinfo->entropy->decode_mcus) (cinfo, + diff->diff_buf, yoffset, MCU_col_num, + cinfo->MCUs_per_row - MCU_col_num); + if (MCU_count != cinfo->MCUs_per_row - MCU_col_num) { + /* Suspension forced; update state counters and exit */ + diff->MCU_vert_offset = yoffset; + diff->MCU_ctr += MCU_count; + return JPEG_SUSPENDED; + } + + /* Account for restart interval (no-op if not using restarts) */ + if (cinfo->restart_interval) + diff->restart_rows_to_go--; + + /* Completed an MCU row, but perhaps not an iMCU row */ + diff->MCU_ctr = 0; + } + + /* + * Undifference and scale each scanline of the disassembled MCU row + * separately. We do not process dummy samples at the end of a scanline + * or dummy rows at the end of the image. + */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + compi = compptr->component_index; + for (row = 0, prev_row = compptr->v_samp_factor - 1; + row < (cinfo->input_iMCU_row == last_iMCU_row ? + compptr->last_row_height : compptr->v_samp_factor); + prev_row = row, row++) { + (*losslessd->predict_undifference[compi]) + (cinfo, compi, diff->diff_buf[compi][row], + diff->undiff_buf[compi][prev_row], diff->undiff_buf[compi][row], + compptr->width_in_blocks); + (*losslessd->scaler_scale) (cinfo, diff->undiff_buf[compi][row], + output_buf[compi][row], + compptr->width_in_blocks); + } + } + + /* Completed the iMCU row, advance counters for next one. + * + * NB: output_data will increment output_iMCU_row. + * This counter is not needed for the single-pass case + * or the input side of the multi-pass case. + */ + if (++(cinfo->input_iMCU_row) < cinfo->total_iMCU_rows) { + start_iMCU_row(cinfo); + return JPEG_ROW_COMPLETED; + } + /* Completed the scan */ + (*cinfo->inputctl->finish_input_pass) (cinfo); + return JPEG_SCAN_COMPLETED; +} + + +/* + * Dummy consume-input routine for single-pass operation. + */ + +METHODDEF(int) +dummy_consume_data(j_decompress_ptr cinfo) +{ + return JPEG_SUSPENDED; /* Always indicate nothing was done */ +} + + +#ifdef D_MULTISCAN_FILES_SUPPORTED + +/* + * Consume input data and store it in the full-image sample buffer. + * We read as much as one fully interleaved MCU row ("iMCU" row) per call, + * ie, v_samp_factor rows for each component in the scan. + * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. + */ + +METHODDEF(int) +consume_data(j_decompress_ptr cinfo) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + int ci, compi; + _JSAMPARRAY buffer[MAX_COMPS_IN_SCAN]; + jpeg_component_info *compptr; + + /* Align the virtual buffers for the components used in this scan. */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + compi = compptr->component_index; + buffer[compi] = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray) + ((j_common_ptr)cinfo, diff->whole_image[compi], + cinfo->input_iMCU_row * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, TRUE); + } + + return decompress_data(cinfo, buffer); +} + + +/* + * Output some data from the full-image sample buffer in the multi-pass case. + * Always attempts to emit one fully interleaved MCU row ("iMCU" row). + * Return value is JPEG_ROW_COMPLETED, JPEG_SCAN_COMPLETED, or JPEG_SUSPENDED. + * + * NB: output_buf contains a plane for each component in image. + */ + +METHODDEF(int) +output_data(j_decompress_ptr cinfo, _JSAMPIMAGE output_buf) +{ + my_diff_ptr diff = (my_diff_ptr)cinfo->coef; + JDIMENSION last_iMCU_row = cinfo->total_iMCU_rows - 1; + int ci, samp_rows, row; + _JSAMPARRAY buffer; + jpeg_component_info *compptr; + + /* Force some input to be done if we are getting ahead of the input. */ + while (cinfo->input_scan_number < cinfo->output_scan_number || + (cinfo->input_scan_number == cinfo->output_scan_number && + cinfo->input_iMCU_row <= cinfo->output_iMCU_row)) { + if ((*cinfo->inputctl->consume_input) (cinfo) == JPEG_SUSPENDED) + return JPEG_SUSPENDED; + } + + /* OK, output from the virtual arrays. */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Align the virtual buffer for this component. */ + buffer = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray) + ((j_common_ptr)cinfo, diff->whole_image[ci], + cinfo->output_iMCU_row * compptr->v_samp_factor, + (JDIMENSION)compptr->v_samp_factor, FALSE); + + if (cinfo->output_iMCU_row < last_iMCU_row) + samp_rows = compptr->v_samp_factor; + else { + /* NB: can't use last_row_height here; it is input-side-dependent! */ + samp_rows = (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (samp_rows == 0) samp_rows = compptr->v_samp_factor; + } + + for (row = 0; row < samp_rows; row++) { + memcpy(output_buf[ci][row], buffer[row], + compptr->width_in_blocks * sizeof(_JSAMPLE)); + } + } + + if (++(cinfo->output_iMCU_row) < cinfo->total_iMCU_rows) + return JPEG_ROW_COMPLETED; + return JPEG_SCAN_COMPLETED; +} + +#endif /* D_MULTISCAN_FILES_SUPPORTED */ + + +/* + * Initialize difference buffer controller. + */ + +GLOBAL(void) +_jinit_d_diff_controller(j_decompress_ptr cinfo, boolean need_full_buffer) +{ + my_diff_ptr diff; + int ci; + jpeg_component_info *compptr; + + diff = (my_diff_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_diff_controller)); + cinfo->coef = (struct jpeg_d_coef_controller *)diff; + diff->pub.start_input_pass = start_input_pass; + diff->pub.start_output_pass = start_output_pass; + + /* Create the [un]difference buffers. */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + diff->diff_buf[ci] = + ALLOC_DARRAY(JPOOL_IMAGE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)compptr->v_samp_factor); + diff->undiff_buf[ci] = + ALLOC_DARRAY(JPOOL_IMAGE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)compptr->v_samp_factor); + } + + if (need_full_buffer) { +#ifdef D_MULTISCAN_FILES_SUPPORTED + /* Allocate a full-image virtual array for each component. */ + int access_rows; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + access_rows = compptr->v_samp_factor; + diff->whole_image[ci] = (*cinfo->mem->request_virt_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, FALSE, + (JDIMENSION)jround_up((long)compptr->width_in_blocks, + (long)compptr->h_samp_factor), + (JDIMENSION)jround_up((long)compptr->height_in_blocks, + (long)compptr->v_samp_factor), + (JDIMENSION)access_rows); + } + diff->pub.consume_data = consume_data; + diff->pub._decompress_data = output_data; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + diff->pub.consume_data = dummy_consume_data; + diff->pub._decompress_data = decompress_data; + diff->whole_image[0] = NULL; /* flag for no virtual arrays */ + } +} + +#endif /* D_LOSSLESS_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jdhuff.c b/src/spice2x/external/libjpeg-turbo/jdhuff.c new file mode 100644 index 0000000..cd8c084 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdhuff.c @@ -0,0 +1,836 @@ +/* + * jdhuff.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2009-2011, 2016, 2018-2019, 2022, D. R. Commander. + * Copyright (C) 2018, Matthias Räncker. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains Huffman entropy decoding routines. + * + * Much of the complexity here has to do with supporting input suspension. + * If the data source module demands suspension, we want to be able to back + * up to the start of the current MCU. To do this, we copy state variables + * into local working storage, and update them back to the permanent + * storage only upon successful completion of an MCU. + * + * NOTE: All referenced figures are from + * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdhuff.h" /* Declarations shared with jd*huff.c */ +#include "jpegapicomp.h" +#include "jstdhuff.c" + + +/* + * Expanded entropy decoder object for Huffman decoding. + * + * The savable_state subrecord contains fields that change within an MCU, + * but must not be updated permanently until we complete the MCU. + */ + +typedef struct { + int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */ +} savable_state; + +typedef struct { + struct jpeg_entropy_decoder pub; /* public fields */ + + /* These fields are loaded into local variables at start of each MCU. + * In case of suspension, we exit WITHOUT updating them. + */ + bitread_perm_state bitstate; /* Bit buffer at start of MCU */ + savable_state saved; /* Other state at start of MCU */ + + /* These fields are NOT loaded into local working state. */ + unsigned int restarts_to_go; /* MCUs left in this restart interval */ + + /* Pointers to derived tables (these workspaces have image lifespan) */ + d_derived_tbl *dc_derived_tbls[NUM_HUFF_TBLS]; + d_derived_tbl *ac_derived_tbls[NUM_HUFF_TBLS]; + + /* Precalculated info set up by start_pass for use in decode_mcu: */ + + /* Pointers to derived tables to be used for each block within an MCU */ + d_derived_tbl *dc_cur_tbls[D_MAX_BLOCKS_IN_MCU]; + d_derived_tbl *ac_cur_tbls[D_MAX_BLOCKS_IN_MCU]; + /* Whether we care about the DC and AC coefficient values for each block */ + boolean dc_needed[D_MAX_BLOCKS_IN_MCU]; + boolean ac_needed[D_MAX_BLOCKS_IN_MCU]; +} huff_entropy_decoder; + +typedef huff_entropy_decoder *huff_entropy_ptr; + + +/* + * Initialize for a Huffman-compressed scan. + */ + +METHODDEF(void) +start_pass_huff_decoder(j_decompress_ptr cinfo) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + int ci, blkn, dctbl, actbl; + d_derived_tbl **pdtbl; + jpeg_component_info *compptr; + + /* Check that the scan parameters Ss, Se, Ah/Al are OK for sequential JPEG. + * This ought to be an error condition, but we make it a warning because + * there are some baseline files out there with all zeroes in these bytes. + */ + if (cinfo->Ss != 0 || cinfo->Se != DCTSIZE2 - 1 || + cinfo->Ah != 0 || cinfo->Al != 0) + WARNMS(cinfo, JWRN_NOT_SEQUENTIAL); + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + dctbl = compptr->dc_tbl_no; + actbl = compptr->ac_tbl_no; + /* Compute derived values for Huffman tables */ + /* We may do this more than once for a table, but it's not expensive */ + pdtbl = (d_derived_tbl **)(entropy->dc_derived_tbls) + dctbl; + jpeg_make_d_derived_tbl(cinfo, TRUE, dctbl, pdtbl); + pdtbl = (d_derived_tbl **)(entropy->ac_derived_tbls) + actbl; + jpeg_make_d_derived_tbl(cinfo, FALSE, actbl, pdtbl); + /* Initialize DC predictions to 0 */ + entropy->saved.last_dc_val[ci] = 0; + } + + /* Precalculate decoding info for each block in an MCU of this scan */ + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + /* Precalculate which table to use for each block */ + entropy->dc_cur_tbls[blkn] = entropy->dc_derived_tbls[compptr->dc_tbl_no]; + entropy->ac_cur_tbls[blkn] = entropy->ac_derived_tbls[compptr->ac_tbl_no]; + /* Decide whether we really care about the coefficient values */ + if (compptr->component_needed) { + entropy->dc_needed[blkn] = TRUE; + /* we don't need the ACs if producing a 1/8th-size image */ + entropy->ac_needed[blkn] = (compptr->_DCT_scaled_size > 1); + } else { + entropy->dc_needed[blkn] = entropy->ac_needed[blkn] = FALSE; + } + } + + /* Initialize bitread state variables */ + entropy->bitstate.bits_left = 0; + entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */ + entropy->pub.insufficient_data = FALSE; + + /* Initialize restart counter */ + entropy->restarts_to_go = cinfo->restart_interval; +} + + +/* + * Compute the derived values for a Huffman table. + * This routine also performs some validation checks on the table. + * + * Note this is also used by jdphuff.c and jdlhuff.c. + */ + +GLOBAL(void) +jpeg_make_d_derived_tbl(j_decompress_ptr cinfo, boolean isDC, int tblno, + d_derived_tbl **pdtbl) +{ + JHUFF_TBL *htbl; + d_derived_tbl *dtbl; + int p, i, l, si, numsymbols; + int lookbits, ctr; + char huffsize[257]; + unsigned int huffcode[257]; + unsigned int code; + + /* Note that huffsize[] and huffcode[] are filled in code-length order, + * paralleling the order of the symbols themselves in htbl->huffval[]. + */ + + /* Find the input Huffman table */ + if (tblno < 0 || tblno >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno); + htbl = + isDC ? cinfo->dc_huff_tbl_ptrs[tblno] : cinfo->ac_huff_tbl_ptrs[tblno]; + if (htbl == NULL) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, tblno); + + /* Allocate a workspace if we haven't already done so. */ + if (*pdtbl == NULL) + *pdtbl = (d_derived_tbl *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(d_derived_tbl)); + dtbl = *pdtbl; + dtbl->pub = htbl; /* fill in back link */ + + /* Figure C.1: make table of Huffman code length for each symbol */ + + p = 0; + for (l = 1; l <= 16; l++) { + i = (int)htbl->bits[l]; + if (i < 0 || p + i > 256) /* protect against table overrun */ + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + while (i--) + huffsize[p++] = (char)l; + } + huffsize[p] = 0; + numsymbols = p; + + /* Figure C.2: generate the codes themselves */ + /* We also validate that the counts represent a legal Huffman code tree. */ + + code = 0; + si = huffsize[0]; + p = 0; + while (huffsize[p]) { + while (((int)huffsize[p]) == si) { + huffcode[p++] = code; + code++; + } + /* code is now 1 more than the last code used for codelength si; but + * it must still fit in si bits, since no code is allowed to be all ones. + */ + if (((JLONG)code) >= (((JLONG)1) << si)) + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + code <<= 1; + si++; + } + + /* Figure F.15: generate decoding tables for bit-sequential decoding */ + + p = 0; + for (l = 1; l <= 16; l++) { + if (htbl->bits[l]) { + /* valoffset[l] = huffval[] index of 1st symbol of code length l, + * minus the minimum code of length l + */ + dtbl->valoffset[l] = (JLONG)p - (JLONG)huffcode[p]; + p += htbl->bits[l]; + dtbl->maxcode[l] = huffcode[p - 1]; /* maximum code of length l */ + } else { + dtbl->maxcode[l] = -1; /* -1 if no codes of this length */ + } + } + dtbl->valoffset[17] = 0; + dtbl->maxcode[17] = 0xFFFFFL; /* ensures jpeg_huff_decode terminates */ + + /* Compute lookahead tables to speed up decoding. + * First we set all the table entries to 0, indicating "too long"; + * then we iterate through the Huffman codes that are short enough and + * fill in all the entries that correspond to bit sequences starting + * with that code. + */ + + for (i = 0; i < (1 << HUFF_LOOKAHEAD); i++) + dtbl->lookup[i] = (HUFF_LOOKAHEAD + 1) << HUFF_LOOKAHEAD; + + p = 0; + for (l = 1; l <= HUFF_LOOKAHEAD; l++) { + for (i = 1; i <= (int)htbl->bits[l]; i++, p++) { + /* l = current code's length, p = its index in huffcode[] & huffval[]. */ + /* Generate left-justified code followed by all possible bit sequences */ + lookbits = huffcode[p] << (HUFF_LOOKAHEAD - l); + for (ctr = 1 << (HUFF_LOOKAHEAD - l); ctr > 0; ctr--) { + dtbl->lookup[lookbits] = (l << HUFF_LOOKAHEAD) | htbl->huffval[p]; + lookbits++; + } + } + } + + /* Validate symbols as being reasonable. + * For AC tables, we make no check, but accept all byte values 0..255. + * For DC tables, we require the symbols to be in range 0..15 in lossy mode + * and 0..16 in lossless mode. (Tighter bounds could be applied depending on + * the data depth and mode, but this is sufficient to ensure safe decoding.) + */ + if (isDC) { + for (i = 0; i < numsymbols; i++) { + int sym = htbl->huffval[i]; + if (sym < 0 || sym > (cinfo->master->lossless ? 16 : 15)) + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + } + } +} + + +/* + * Out-of-line code for bit fetching (shared with jdphuff.c and jdlhuff.c). + * See jdhuff.h for info about usage. + * Note: current values of get_buffer and bits_left are passed as parameters, + * but are returned in the corresponding fields of the state struct. + * + * On most machines MIN_GET_BITS should be 25 to allow the full 32-bit width + * of get_buffer to be used. (On machines with wider words, an even larger + * buffer could be used.) However, on some machines 32-bit shifts are + * quite slow and take time proportional to the number of places shifted. + * (This is true with most PC compilers, for instance.) In this case it may + * be a win to set MIN_GET_BITS to the minimum value of 15. This reduces the + * average shift distance at the cost of more calls to jpeg_fill_bit_buffer. + */ + +#ifdef SLOW_SHIFT_32 +#define MIN_GET_BITS 15 /* minimum allowable value */ +#else +#define MIN_GET_BITS (BIT_BUF_SIZE - 7) +#endif + + +GLOBAL(boolean) +jpeg_fill_bit_buffer(bitread_working_state *state, + register bit_buf_type get_buffer, register int bits_left, + int nbits) +/* Load up the bit buffer to a depth of at least nbits */ +{ + /* Copy heavily used state fields into locals (hopefully registers) */ + register const JOCTET *next_input_byte = state->next_input_byte; + register size_t bytes_in_buffer = state->bytes_in_buffer; + j_decompress_ptr cinfo = state->cinfo; + + /* Attempt to load at least MIN_GET_BITS bits into get_buffer. */ + /* (It is assumed that no request will be for more than that many bits.) */ + /* We fail to do so only if we hit a marker or are forced to suspend. */ + + if (cinfo->unread_marker == 0) { /* cannot advance past a marker */ + while (bits_left < MIN_GET_BITS) { + register int c; + + /* Attempt to read a byte */ + if (bytes_in_buffer == 0) { + if (!(*cinfo->src->fill_input_buffer) (cinfo)) + return FALSE; + next_input_byte = cinfo->src->next_input_byte; + bytes_in_buffer = cinfo->src->bytes_in_buffer; + } + bytes_in_buffer--; + c = *next_input_byte++; + + /* If it's 0xFF, check and discard stuffed zero byte */ + if (c == 0xFF) { + /* Loop here to discard any padding FF's on terminating marker, + * so that we can save a valid unread_marker value. NOTE: we will + * accept multiple FF's followed by a 0 as meaning a single FF data + * byte. This data pattern is not valid according to the standard. + */ + do { + if (bytes_in_buffer == 0) { + if (!(*cinfo->src->fill_input_buffer) (cinfo)) + return FALSE; + next_input_byte = cinfo->src->next_input_byte; + bytes_in_buffer = cinfo->src->bytes_in_buffer; + } + bytes_in_buffer--; + c = *next_input_byte++; + } while (c == 0xFF); + + if (c == 0) { + /* Found FF/00, which represents an FF data byte */ + c = 0xFF; + } else { + /* Oops, it's actually a marker indicating end of compressed data. + * Save the marker code for later use. + * Fine point: it might appear that we should save the marker into + * bitread working state, not straight into permanent state. But + * once we have hit a marker, we cannot need to suspend within the + * current MCU, because we will read no more bytes from the data + * source. So it is OK to update permanent state right away. + */ + cinfo->unread_marker = c; + /* See if we need to insert some fake zero bits. */ + goto no_more_bytes; + } + } + + /* OK, load c into get_buffer */ + get_buffer = (get_buffer << 8) | c; + bits_left += 8; + } /* end while */ + } else { +no_more_bytes: + /* We get here if we've read the marker that terminates the compressed + * data segment. There should be enough bits in the buffer register + * to satisfy the request; if so, no problem. + */ + if (nbits > bits_left) { + /* Uh-oh. Report corrupted data to user and stuff zeroes into + * the data stream, so that we can produce some kind of image. + * We use a nonvolatile flag to ensure that only one warning message + * appears per data segment. + */ + if (!cinfo->entropy->insufficient_data) { + WARNMS(cinfo, JWRN_HIT_MARKER); + cinfo->entropy->insufficient_data = TRUE; + } + /* Fill the buffer with zero bits */ + get_buffer <<= MIN_GET_BITS - bits_left; + bits_left = MIN_GET_BITS; + } + } + + /* Unload the local registers */ + state->next_input_byte = next_input_byte; + state->bytes_in_buffer = bytes_in_buffer; + state->get_buffer = get_buffer; + state->bits_left = bits_left; + + return TRUE; +} + + +/* Macro version of the above, which performs much better but does not + handle markers. We have to hand off any blocks with markers to the + slower routines. */ + +#define GET_BYTE { \ + register int c0, c1; \ + c0 = *buffer++; \ + c1 = *buffer; \ + /* Pre-execute most common case */ \ + get_buffer = (get_buffer << 8) | c0; \ + bits_left += 8; \ + if (c0 == 0xFF) { \ + /* Pre-execute case of FF/00, which represents an FF data byte */ \ + buffer++; \ + if (c1 != 0) { \ + /* Oops, it's actually a marker indicating end of compressed data. */ \ + cinfo->unread_marker = c1; \ + /* Back out pre-execution and fill the buffer with zero bits */ \ + buffer -= 2; \ + get_buffer &= ~0xFF; \ + } \ + } \ +} + +#if SIZEOF_SIZE_T == 8 || defined(_WIN64) || (defined(__x86_64__) && defined(__ILP32__)) + +/* Pre-fetch 48 bytes, because the holding register is 64-bit */ +#define FILL_BIT_BUFFER_FAST \ + if (bits_left <= 16) { \ + GET_BYTE GET_BYTE GET_BYTE GET_BYTE GET_BYTE GET_BYTE \ + } + +#else + +/* Pre-fetch 16 bytes, because the holding register is 32-bit */ +#define FILL_BIT_BUFFER_FAST \ + if (bits_left <= 16) { \ + GET_BYTE GET_BYTE \ + } + +#endif + + +/* + * Out-of-line code for Huffman code decoding. + * See jdhuff.h for info about usage. + */ + +GLOBAL(int) +jpeg_huff_decode(bitread_working_state *state, + register bit_buf_type get_buffer, register int bits_left, + d_derived_tbl *htbl, int min_bits) +{ + register int l = min_bits; + register JLONG code; + + /* HUFF_DECODE has determined that the code is at least min_bits */ + /* bits long, so fetch that many bits in one swoop. */ + + CHECK_BIT_BUFFER(*state, l, return -1); + code = GET_BITS(l); + + /* Collect the rest of the Huffman code one bit at a time. */ + /* This is per Figure F.16. */ + + while (code > htbl->maxcode[l]) { + code <<= 1; + CHECK_BIT_BUFFER(*state, 1, return -1); + code |= GET_BITS(1); + l++; + } + + /* Unload the local registers */ + state->get_buffer = get_buffer; + state->bits_left = bits_left; + + /* With garbage input we may reach the sentinel value l = 17. */ + + if (l > 16) { + WARNMS(state->cinfo, JWRN_HUFF_BAD_CODE); + return 0; /* fake a zero as the safest result */ + } + + return htbl->pub->huffval[(int)(code + htbl->valoffset[l])]; +} + + +/* + * Figure F.12: extend sign bit. + * On some machines, a shift and add will be faster than a table lookup. + */ + +#define AVOID_TABLES +#ifdef AVOID_TABLES + +#define NEG_1 ((unsigned int)-1) +#define HUFF_EXTEND(x, s) \ + ((x) + ((((x) - (1 << ((s) - 1))) >> 31) & (((NEG_1) << (s)) + 1))) + +#else + +#define HUFF_EXTEND(x, s) \ + ((x) < extend_test[s] ? (x) + extend_offset[s] : (x)) + +static const int extend_test[16] = { /* entry n is 2**(n-1) */ + 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, + 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 +}; + +static const int extend_offset[16] = { /* entry n is (-1 << n) + 1 */ + 0, ((-1) << 1) + 1, ((-1) << 2) + 1, ((-1) << 3) + 1, ((-1) << 4) + 1, + ((-1) << 5) + 1, ((-1) << 6) + 1, ((-1) << 7) + 1, ((-1) << 8) + 1, + ((-1) << 9) + 1, ((-1) << 10) + 1, ((-1) << 11) + 1, ((-1) << 12) + 1, + ((-1) << 13) + 1, ((-1) << 14) + 1, ((-1) << 15) + 1 +}; + +#endif /* AVOID_TABLES */ + + +/* + * Check for a restart marker & resynchronize decoder. + * Returns FALSE if must suspend. + */ + +LOCAL(boolean) +process_restart(j_decompress_ptr cinfo) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + int ci; + + /* Throw away any unused bits remaining in bit buffer; */ + /* include any full bytes in next_marker's count of discarded bytes */ + cinfo->marker->discarded_bytes += entropy->bitstate.bits_left / 8; + entropy->bitstate.bits_left = 0; + + /* Advance past the RSTn marker */ + if (!(*cinfo->marker->read_restart_marker) (cinfo)) + return FALSE; + + /* Re-initialize DC predictions to 0 */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) + entropy->saved.last_dc_val[ci] = 0; + + /* Reset restart counter */ + entropy->restarts_to_go = cinfo->restart_interval; + + /* Reset out-of-data flag, unless read_restart_marker left us smack up + * against a marker. In that case we will end up treating the next data + * segment as empty, and we can avoid producing bogus output pixels by + * leaving the flag set. + */ + if (cinfo->unread_marker == 0) + entropy->pub.insufficient_data = FALSE; + + return TRUE; +} + + +#if defined(__has_feature) +#if __has_feature(undefined_behavior_sanitizer) +__attribute__((no_sanitize("signed-integer-overflow"), + no_sanitize("unsigned-integer-overflow"))) +#endif +#endif +LOCAL(boolean) +decode_mcu_slow(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + BITREAD_STATE_VARS; + int blkn; + savable_state state; + /* Outer loop handles each block in the MCU */ + + /* Load up working state */ + BITREAD_LOAD_STATE(cinfo, entropy->bitstate); + state = entropy->saved; + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + JBLOCKROW block = MCU_data ? MCU_data[blkn] : NULL; + d_derived_tbl *dctbl = entropy->dc_cur_tbls[blkn]; + d_derived_tbl *actbl = entropy->ac_cur_tbls[blkn]; + register int s, k, r; + + /* Decode a single block's worth of coefficients */ + + /* Section F.2.2.1: decode the DC coefficient difference */ + HUFF_DECODE(s, br_state, dctbl, return FALSE, label1); + if (s) { + CHECK_BIT_BUFFER(br_state, s, return FALSE); + r = GET_BITS(s); + s = HUFF_EXTEND(r, s); + } + + if (entropy->dc_needed[blkn]) { + /* Convert DC difference to actual value, update last_dc_val */ + int ci = cinfo->MCU_membership[blkn]; + /* Certain malformed JPEG images produce repeated DC coefficient + * differences of 2047 or -2047, which causes state.last_dc_val[ci] to + * grow until it overflows or underflows a 32-bit signed integer. This + * behavior is, to the best of our understanding, innocuous, and it is + * unclear how to work around it without potentially affecting + * performance. Thus, we (hopefully temporarily) suppress UBSan integer + * overflow errors for this function and decode_mcu_fast(). + */ + s += state.last_dc_val[ci]; + state.last_dc_val[ci] = s; + if (block) { + /* Output the DC coefficient (assumes jpeg_natural_order[0] = 0) */ + (*block)[0] = (JCOEF)s; + } + } + + if (entropy->ac_needed[blkn] && block) { + + /* Section F.2.2.2: decode the AC coefficients */ + /* Since zeroes are skipped, output area must be cleared beforehand */ + for (k = 1; k < DCTSIZE2; k++) { + HUFF_DECODE(s, br_state, actbl, return FALSE, label2); + + r = s >> 4; + s &= 15; + + if (s) { + k += r; + CHECK_BIT_BUFFER(br_state, s, return FALSE); + r = GET_BITS(s); + s = HUFF_EXTEND(r, s); + /* Output coefficient in natural (dezigzagged) order. + * Note: the extra entries in jpeg_natural_order[] will save us + * if k >= DCTSIZE2, which could happen if the data is corrupted. + */ + (*block)[jpeg_natural_order[k]] = (JCOEF)s; + } else { + if (r != 15) + break; + k += 15; + } + } + + } else { + + /* Section F.2.2.2: decode the AC coefficients */ + /* In this path we just discard the values */ + for (k = 1; k < DCTSIZE2; k++) { + HUFF_DECODE(s, br_state, actbl, return FALSE, label3); + + r = s >> 4; + s &= 15; + + if (s) { + k += r; + CHECK_BIT_BUFFER(br_state, s, return FALSE); + DROP_BITS(s); + } else { + if (r != 15) + break; + k += 15; + } + } + } + } + + /* Completed MCU, so update state */ + BITREAD_SAVE_STATE(cinfo, entropy->bitstate); + entropy->saved = state; + return TRUE; +} + + +#if defined(__has_feature) +#if __has_feature(undefined_behavior_sanitizer) +__attribute__((no_sanitize("signed-integer-overflow"), + no_sanitize("unsigned-integer-overflow"))) +#endif +#endif +LOCAL(boolean) +decode_mcu_fast(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + BITREAD_STATE_VARS; + JOCTET *buffer; + int blkn; + savable_state state; + /* Outer loop handles each block in the MCU */ + + /* Load up working state */ + BITREAD_LOAD_STATE(cinfo, entropy->bitstate); + buffer = (JOCTET *)br_state.next_input_byte; + state = entropy->saved; + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + JBLOCKROW block = MCU_data ? MCU_data[blkn] : NULL; + d_derived_tbl *dctbl = entropy->dc_cur_tbls[blkn]; + d_derived_tbl *actbl = entropy->ac_cur_tbls[blkn]; + register int s, k, r, l; + + HUFF_DECODE_FAST(s, l, dctbl); + if (s) { + FILL_BIT_BUFFER_FAST + r = GET_BITS(s); + s = HUFF_EXTEND(r, s); + } + + if (entropy->dc_needed[blkn]) { + int ci = cinfo->MCU_membership[blkn]; + /* Refer to the comment in decode_mcu_slow() regarding the supression of + * a UBSan integer overflow error in this line of code. + */ + s += state.last_dc_val[ci]; + state.last_dc_val[ci] = s; + if (block) + (*block)[0] = (JCOEF)s; + } + + if (entropy->ac_needed[blkn] && block) { + + for (k = 1; k < DCTSIZE2; k++) { + HUFF_DECODE_FAST(s, l, actbl); + r = s >> 4; + s &= 15; + + if (s) { + k += r; + FILL_BIT_BUFFER_FAST + r = GET_BITS(s); + s = HUFF_EXTEND(r, s); + (*block)[jpeg_natural_order[k]] = (JCOEF)s; + } else { + if (r != 15) break; + k += 15; + } + } + + } else { + + for (k = 1; k < DCTSIZE2; k++) { + HUFF_DECODE_FAST(s, l, actbl); + r = s >> 4; + s &= 15; + + if (s) { + k += r; + FILL_BIT_BUFFER_FAST + DROP_BITS(s); + } else { + if (r != 15) break; + k += 15; + } + } + } + } + + if (cinfo->unread_marker != 0) { + cinfo->unread_marker = 0; + return FALSE; + } + + br_state.bytes_in_buffer -= (buffer - br_state.next_input_byte); + br_state.next_input_byte = buffer; + BITREAD_SAVE_STATE(cinfo, entropy->bitstate); + entropy->saved = state; + return TRUE; +} + + +/* + * Decode and return one MCU's worth of Huffman-compressed coefficients. + * The coefficients are reordered from zigzag order into natural array order, + * but are not dequantized. + * + * The i'th block of the MCU is stored into the block pointed to by + * MCU_data[i]. WE ASSUME THIS AREA HAS BEEN ZEROED BY THE CALLER. + * (Wholesale zeroing is usually a little faster than retail...) + * + * Returns FALSE if data source requested suspension. In that case no + * changes have been made to permanent state. (Exception: some output + * coefficients may already have been assigned. This is harmless for + * this module, since we'll just re-assign them on the next call.) + */ + +#define BUFSIZE (DCTSIZE2 * 8) + +METHODDEF(boolean) +decode_mcu(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + huff_entropy_ptr entropy = (huff_entropy_ptr)cinfo->entropy; + int usefast = 1; + + /* Process restart marker if needed; may have to suspend */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + if (!process_restart(cinfo)) + return FALSE; + usefast = 0; + } + + if (cinfo->src->bytes_in_buffer < BUFSIZE * (size_t)cinfo->blocks_in_MCU || + cinfo->unread_marker != 0) + usefast = 0; + + /* If we've run out of data, just leave the MCU set to zeroes. + * This way, we return uniform gray for the remainder of the segment. + */ + if (!entropy->pub.insufficient_data) { + + if (usefast) { + if (!decode_mcu_fast(cinfo, MCU_data)) goto use_slow; + } else { +use_slow: + if (!decode_mcu_slow(cinfo, MCU_data)) return FALSE; + } + + } + + /* Account for restart interval (no-op if not using restarts) */ + if (cinfo->restart_interval) + entropy->restarts_to_go--; + + return TRUE; +} + + +/* + * Module initialization routine for Huffman entropy decoding. + */ + +GLOBAL(void) +jinit_huff_decoder(j_decompress_ptr cinfo) +{ + huff_entropy_ptr entropy; + int i; + + /* Motion JPEG frames typically do not include the Huffman tables if they + are the default tables. Thus, if the tables are not set by the time + the Huffman decoder is initialized (usually within the body of + jpeg_start_decompress()), we set them to default values. */ + std_huff_tables((j_common_ptr)cinfo); + + entropy = (huff_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(huff_entropy_decoder)); + cinfo->entropy = (struct jpeg_entropy_decoder *)entropy; + entropy->pub.start_pass = start_pass_huff_decoder; + entropy->pub.decode_mcu = decode_mcu; + + /* Mark tables unallocated */ + for (i = 0; i < NUM_HUFF_TBLS; i++) { + entropy->dc_derived_tbls[i] = entropy->ac_derived_tbls[i] = NULL; + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jdhuff.h b/src/spice2x/external/libjpeg-turbo/jdhuff.h new file mode 100644 index 0000000..3eee002 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdhuff.h @@ -0,0 +1,250 @@ +/* + * jdhuff.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2010-2011, 2015-2016, 2021, D. R. Commander. + * Copyright (C) 2018, Matthias Räncker. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains declarations for Huffman entropy decoding routines + * that are shared between the sequential decoder (jdhuff.c), the progressive + * decoder (jdphuff.c), and the lossless decoder (jdlhuff.c). No other modules + * need to see these. + */ + +#include "jconfigint.h" + + +/* Derived data constructed for each Huffman table */ + +#define HUFF_LOOKAHEAD 8 /* # of bits of lookahead */ + +typedef struct { + /* Basic tables: (element [0] of each array is unused) */ + JLONG maxcode[18]; /* largest code of length k (-1 if none) */ + /* (maxcode[17] is a sentinel to ensure jpeg_huff_decode terminates) */ + JLONG valoffset[18]; /* huffval[] offset for codes of length k */ + /* valoffset[k] = huffval[] index of 1st symbol of code length k, less + * the smallest code of length k; so given a code of length k, the + * corresponding symbol is huffval[code + valoffset[k]] + */ + + /* Link to public Huffman table (needed only in jpeg_huff_decode) */ + JHUFF_TBL *pub; + + /* Lookahead table: indexed by the next HUFF_LOOKAHEAD bits of + * the input data stream. If the next Huffman code is no more + * than HUFF_LOOKAHEAD bits long, we can obtain its length and + * the corresponding symbol directly from this tables. + * + * The lower 8 bits of each table entry contain the number of + * bits in the corresponding Huffman code, or HUFF_LOOKAHEAD + 1 + * if too long. The next 8 bits of each entry contain the + * symbol. + */ + int lookup[1 << HUFF_LOOKAHEAD]; +} d_derived_tbl; + +/* Expand a Huffman table definition into the derived format */ +EXTERN(void) jpeg_make_d_derived_tbl(j_decompress_ptr cinfo, boolean isDC, + int tblno, d_derived_tbl **pdtbl); + + +/* + * Fetching the next N bits from the input stream is a time-critical operation + * for the Huffman decoders. We implement it with a combination of inline + * macros and out-of-line subroutines. Note that N (the number of bits + * demanded at one time) never exceeds 15 for JPEG use. + * + * We read source bytes into get_buffer and dole out bits as needed. + * If get_buffer already contains enough bits, they are fetched in-line + * by the macros CHECK_BIT_BUFFER and GET_BITS. When there aren't enough + * bits, jpeg_fill_bit_buffer is called; it will attempt to fill get_buffer + * as full as possible (not just to the number of bits needed; this + * prefetching reduces the overhead cost of calling jpeg_fill_bit_buffer). + * Note that jpeg_fill_bit_buffer may return FALSE to indicate suspension. + * On TRUE return, jpeg_fill_bit_buffer guarantees that get_buffer contains + * at least the requested number of bits --- dummy zeroes are inserted if + * necessary. + */ + +#if !defined(_WIN32) && !defined(SIZEOF_SIZE_T) +#error Cannot determine word size +#endif + +#if SIZEOF_SIZE_T == 8 || defined(_WIN64) + +typedef size_t bit_buf_type; /* type of bit-extraction buffer */ +#define BIT_BUF_SIZE 64 /* size of buffer in bits */ + +#elif defined(__x86_64__) && defined(__ILP32__) + +typedef unsigned long long bit_buf_type; /* type of bit-extraction buffer */ +#define BIT_BUF_SIZE 64 /* size of buffer in bits */ + +#else + +typedef unsigned long bit_buf_type; /* type of bit-extraction buffer */ +#define BIT_BUF_SIZE 32 /* size of buffer in bits */ + +#endif + +/* If long is > 32 bits on your machine, and shifting/masking longs is + * reasonably fast, making bit_buf_type be long and setting BIT_BUF_SIZE + * appropriately should be a win. Unfortunately we can't define the size + * with something like #define BIT_BUF_SIZE (sizeof(bit_buf_type)*8) + * because not all machines measure sizeof in 8-bit bytes. + */ + +typedef struct { /* Bitreading state saved across MCUs */ + bit_buf_type get_buffer; /* current bit-extraction buffer */ + int bits_left; /* # of unused bits in it */ +} bitread_perm_state; + +typedef struct { /* Bitreading working state within an MCU */ + /* Current data source location */ + /* We need a copy, rather than munging the original, in case of suspension */ + const JOCTET *next_input_byte; /* => next byte to read from source */ + size_t bytes_in_buffer; /* # of bytes remaining in source buffer */ + /* Bit input buffer --- note these values are kept in register variables, + * not in this struct, inside the inner loops. + */ + bit_buf_type get_buffer; /* current bit-extraction buffer */ + int bits_left; /* # of unused bits in it */ + /* Pointer needed by jpeg_fill_bit_buffer. */ + j_decompress_ptr cinfo; /* back link to decompress master record */ +} bitread_working_state; + +/* Macros to declare and load/save bitread local variables. */ +#define BITREAD_STATE_VARS \ + register bit_buf_type get_buffer; \ + register int bits_left; \ + bitread_working_state br_state + +#define BITREAD_LOAD_STATE(cinfop, permstate) \ + br_state.cinfo = cinfop; \ + br_state.next_input_byte = cinfop->src->next_input_byte; \ + br_state.bytes_in_buffer = cinfop->src->bytes_in_buffer; \ + get_buffer = permstate.get_buffer; \ + bits_left = permstate.bits_left; + +#define BITREAD_SAVE_STATE(cinfop, permstate) \ + cinfop->src->next_input_byte = br_state.next_input_byte; \ + cinfop->src->bytes_in_buffer = br_state.bytes_in_buffer; \ + permstate.get_buffer = get_buffer; \ + permstate.bits_left = bits_left + +/* + * These macros provide the in-line portion of bit fetching. + * Use CHECK_BIT_BUFFER to ensure there are N bits in get_buffer + * before using GET_BITS, PEEK_BITS, or DROP_BITS. + * The variables get_buffer and bits_left are assumed to be locals, + * but the state struct might not be (jpeg_huff_decode needs this). + * CHECK_BIT_BUFFER(state, n, action); + * Ensure there are N bits in get_buffer; if suspend, take action. + * val = GET_BITS(n); + * Fetch next N bits. + * val = PEEK_BITS(n); + * Fetch next N bits without removing them from the buffer. + * DROP_BITS(n); + * Discard next N bits. + * The value N should be a simple variable, not an expression, because it + * is evaluated multiple times. + */ + +#define CHECK_BIT_BUFFER(state, nbits, action) { \ + if (bits_left < (nbits)) { \ + if (!jpeg_fill_bit_buffer(&(state), get_buffer, bits_left, nbits)) \ + { action; } \ + get_buffer = (state).get_buffer; bits_left = (state).bits_left; \ + } \ +} + +#define GET_BITS(nbits) \ + (((int)(get_buffer >> (bits_left -= (nbits)))) & ((1 << (nbits)) - 1)) + +#define PEEK_BITS(nbits) \ + (((int)(get_buffer >> (bits_left - (nbits)))) & ((1 << (nbits)) - 1)) + +#define DROP_BITS(nbits) \ + (bits_left -= (nbits)) + +/* Load up the bit buffer to a depth of at least nbits */ +EXTERN(boolean) jpeg_fill_bit_buffer(bitread_working_state *state, + register bit_buf_type get_buffer, + register int bits_left, int nbits); + + +/* + * Code for extracting next Huffman-coded symbol from input bit stream. + * Again, this is time-critical and we make the main paths be macros. + * + * We use a lookahead table to process codes of up to HUFF_LOOKAHEAD bits + * without looping. Usually, more than 95% of the Huffman codes will be 8 + * or fewer bits long. The few overlength codes are handled with a loop, + * which need not be inline code. + * + * Notes about the HUFF_DECODE macro: + * 1. Near the end of the data segment, we may fail to get enough bits + * for a lookahead. In that case, we do it the hard way. + * 2. If the lookahead table contains no entry, the next code must be + * more than HUFF_LOOKAHEAD bits long. + * 3. jpeg_huff_decode returns -1 if forced to suspend. + */ + +#define HUFF_DECODE(result, state, htbl, failaction, slowlabel) { \ + register int nb, look; \ + if (bits_left < HUFF_LOOKAHEAD) { \ + if (!jpeg_fill_bit_buffer(&state, get_buffer, bits_left, 0)) \ + { failaction; } \ + get_buffer = state.get_buffer; bits_left = state.bits_left; \ + if (bits_left < HUFF_LOOKAHEAD) { \ + nb = 1; goto slowlabel; \ + } \ + } \ + look = PEEK_BITS(HUFF_LOOKAHEAD); \ + if ((nb = (htbl->lookup[look] >> HUFF_LOOKAHEAD)) <= HUFF_LOOKAHEAD) { \ + DROP_BITS(nb); \ + result = htbl->lookup[look] & ((1 << HUFF_LOOKAHEAD) - 1); \ + } else { \ +slowlabel: \ + if ((result = \ + jpeg_huff_decode(&state, get_buffer, bits_left, htbl, nb)) < 0) \ + { failaction; } \ + get_buffer = state.get_buffer; bits_left = state.bits_left; \ + } \ +} + +#define HUFF_DECODE_FAST(s, nb, htbl) \ + FILL_BIT_BUFFER_FAST; \ + s = PEEK_BITS(HUFF_LOOKAHEAD); \ + s = htbl->lookup[s]; \ + nb = s >> HUFF_LOOKAHEAD; \ + /* Pre-execute the common case of nb <= HUFF_LOOKAHEAD */ \ + DROP_BITS(nb); \ + s = s & ((1 << HUFF_LOOKAHEAD) - 1); \ + if (nb > HUFF_LOOKAHEAD) { \ + /* Equivalent of jpeg_huff_decode() */ \ + /* Don't use GET_BITS() here because we don't want to modify bits_left */ \ + s = (get_buffer >> bits_left) & ((1 << (nb)) - 1); \ + while (s > htbl->maxcode[nb]) { \ + s <<= 1; \ + s |= GET_BITS(1); \ + nb++; \ + } \ + if (nb > 16) \ + s = 0; \ + else \ + s = htbl->pub->huffval[(int)(s + htbl->valoffset[nb]) & 0xFF]; \ + } + +/* Out-of-line case for Huffman code fetching */ +EXTERN(int) jpeg_huff_decode(bitread_working_state *state, + register bit_buf_type get_buffer, + register int bits_left, d_derived_tbl *htbl, + int min_bits); diff --git a/src/spice2x/external/libjpeg-turbo/jdicc.c b/src/spice2x/external/libjpeg-turbo/jdicc.c new file mode 100644 index 0000000..50aa9a9 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdicc.c @@ -0,0 +1,167 @@ +/* + * jdicc.c + * + * Copyright (C) 1997-1998, Thomas G. Lane, Todd Newman. + * Copyright (C) 2017, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file provides code to read International Color Consortium (ICC) device + * profiles embedded in JFIF JPEG image files. The ICC has defined a standard + * for including such data in JPEG "APP2" markers. The code given here does + * not know anything about the internal structure of the ICC profile data; it + * just knows how to get the profile data from a JPEG file while reading it. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jerror.h" + + +#define ICC_MARKER (JPEG_APP0 + 2) /* JPEG marker code for ICC */ +#define ICC_OVERHEAD_LEN 14 /* size of non-profile data in APP2 */ + + +/* + * Handy subroutine to test whether a saved marker is an ICC profile marker. + */ + +LOCAL(boolean) +marker_is_icc(jpeg_saved_marker_ptr marker) +{ + return + marker->marker == ICC_MARKER && + marker->data_length >= ICC_OVERHEAD_LEN && + /* verify the identifying string */ + marker->data[0] == 0x49 && + marker->data[1] == 0x43 && + marker->data[2] == 0x43 && + marker->data[3] == 0x5F && + marker->data[4] == 0x50 && + marker->data[5] == 0x52 && + marker->data[6] == 0x4F && + marker->data[7] == 0x46 && + marker->data[8] == 0x49 && + marker->data[9] == 0x4C && + marker->data[10] == 0x45 && + marker->data[11] == 0x0; +} + + +/* + * See if there was an ICC profile in the JPEG file being read; if so, + * reassemble and return the profile data. + * + * TRUE is returned if an ICC profile was found, FALSE if not. If TRUE is + * returned, *icc_data_ptr is set to point to the returned data, and + * *icc_data_len is set to its length. + * + * IMPORTANT: the data at *icc_data_ptr is allocated with malloc() and must be + * freed by the caller with free() when the caller no longer needs it. + * (Alternatively, we could write this routine to use the IJG library's memory + * allocator, so that the data would be freed implicitly when + * jpeg_finish_decompress() is called. But it seems likely that many + * applications will prefer to have the data stick around after decompression + * finishes.) + */ + +GLOBAL(boolean) +jpeg_read_icc_profile(j_decompress_ptr cinfo, JOCTET **icc_data_ptr, + unsigned int *icc_data_len) +{ + jpeg_saved_marker_ptr marker; + int num_markers = 0; + int seq_no; + JOCTET *icc_data; + unsigned int total_length; +#define MAX_SEQ_NO 255 /* sufficient since marker numbers are bytes */ + char marker_present[MAX_SEQ_NO + 1]; /* 1 if marker found */ + unsigned int data_length[MAX_SEQ_NO + 1]; /* size of profile data in marker */ + unsigned int data_offset[MAX_SEQ_NO + 1]; /* offset for data in marker */ + + if (icc_data_ptr == NULL || icc_data_len == NULL) + ERREXIT(cinfo, JERR_BUFFER_SIZE); + if (cinfo->global_state < DSTATE_READY) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + *icc_data_ptr = NULL; /* avoid confusion if FALSE return */ + *icc_data_len = 0; + + /* This first pass over the saved markers discovers whether there are + * any ICC markers and verifies the consistency of the marker numbering. + */ + + for (seq_no = 1; seq_no <= MAX_SEQ_NO; seq_no++) + marker_present[seq_no] = 0; + + for (marker = cinfo->marker_list; marker != NULL; marker = marker->next) { + if (marker_is_icc(marker)) { + if (num_markers == 0) + num_markers = marker->data[13]; + else if (num_markers != marker->data[13]) { + WARNMS(cinfo, JWRN_BOGUS_ICC); /* inconsistent num_markers fields */ + return FALSE; + } + seq_no = marker->data[12]; + if (seq_no <= 0 || seq_no > num_markers) { + WARNMS(cinfo, JWRN_BOGUS_ICC); /* bogus sequence number */ + return FALSE; + } + if (marker_present[seq_no]) { + WARNMS(cinfo, JWRN_BOGUS_ICC); /* duplicate sequence numbers */ + return FALSE; + } + marker_present[seq_no] = 1; + data_length[seq_no] = marker->data_length - ICC_OVERHEAD_LEN; + } + } + + if (num_markers == 0) + return FALSE; + + /* Check for missing markers, count total space needed, + * compute offset of each marker's part of the data. + */ + + total_length = 0; + for (seq_no = 1; seq_no <= num_markers; seq_no++) { + if (marker_present[seq_no] == 0) { + WARNMS(cinfo, JWRN_BOGUS_ICC); /* missing sequence number */ + return FALSE; + } + data_offset[seq_no] = total_length; + total_length += data_length[seq_no]; + } + + if (total_length == 0) { + WARNMS(cinfo, JWRN_BOGUS_ICC); /* found only empty markers? */ + return FALSE; + } + + /* Allocate space for assembled data */ + icc_data = (JOCTET *)malloc(total_length * sizeof(JOCTET)); + if (icc_data == NULL) + ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 11); /* oops, out of memory */ + + /* and fill it in */ + for (marker = cinfo->marker_list; marker != NULL; marker = marker->next) { + if (marker_is_icc(marker)) { + JOCTET FAR *src_ptr; + JOCTET *dst_ptr; + unsigned int length; + seq_no = marker->data[12]; + dst_ptr = icc_data + data_offset[seq_no]; + src_ptr = marker->data + ICC_OVERHEAD_LEN; + length = data_length[seq_no]; + while (length--) { + *dst_ptr++ = *src_ptr++; + } + } + } + + *icc_data_ptr = icc_data; + *icc_data_len = total_length; + + return TRUE; +} diff --git a/src/spice2x/external/libjpeg-turbo/jdinput.c b/src/spice2x/external/libjpeg-turbo/jdinput.c new file mode 100644 index 0000000..136bef5 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdinput.c @@ -0,0 +1,419 @@ +/* + * jdinput.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2010, 2016, 2018, 2022, D. R. Commander. + * Copyright (C) 2015, Google, Inc. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains input control logic for the JPEG decompressor. + * These routines are concerned with controlling the decompressor's input + * processing (marker reading and coefficient/difference decoding). + * The actual input reading is done in jdmarker.c, jdhuff.c, jdphuff.c, + * and jdlhuff.c. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jpegapicomp.h" + + +/* Private state */ + +typedef struct { + struct jpeg_input_controller pub; /* public fields */ + + boolean inheaders; /* TRUE until first SOS is reached */ +} my_input_controller; + +typedef my_input_controller *my_inputctl_ptr; + + +/* Forward declarations */ +METHODDEF(int) consume_markers(j_decompress_ptr cinfo); + + +/* + * Routines to calculate various quantities related to the size of the image. + */ + +LOCAL(void) +initial_setup(j_decompress_ptr cinfo) +/* Called once, when first SOS marker is reached */ +{ + int ci; + jpeg_component_info *compptr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + /* Make sure image isn't bigger than I can handle */ + if ((long)cinfo->image_height > (long)JPEG_MAX_DIMENSION || + (long)cinfo->image_width > (long)JPEG_MAX_DIMENSION) + ERREXIT1(cinfo, JERR_IMAGE_TOO_BIG, (unsigned int)JPEG_MAX_DIMENSION); + + /* For now, precision must match compiled-in value... */ +#ifdef D_LOSSLESS_SUPPORTED + if (cinfo->data_precision != 8 && cinfo->data_precision != 12 && + cinfo->data_precision != 16) +#else + if (cinfo->data_precision != 8 && cinfo->data_precision != 12) +#endif + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + /* Check that number of components won't exceed internal array sizes */ + if (cinfo->num_components > MAX_COMPONENTS) + ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->num_components, + MAX_COMPONENTS); + + /* Compute maximum sampling factors; check factor validity */ + cinfo->max_h_samp_factor = 1; + cinfo->max_v_samp_factor = 1; + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + if (compptr->h_samp_factor <= 0 || + compptr->h_samp_factor > MAX_SAMP_FACTOR || + compptr->v_samp_factor <= 0 || + compptr->v_samp_factor > MAX_SAMP_FACTOR) + ERREXIT(cinfo, JERR_BAD_SAMPLING); + cinfo->max_h_samp_factor = MAX(cinfo->max_h_samp_factor, + compptr->h_samp_factor); + cinfo->max_v_samp_factor = MAX(cinfo->max_v_samp_factor, + compptr->v_samp_factor); + } + +#if JPEG_LIB_VERSION >= 80 + cinfo->block_size = data_unit; + cinfo->natural_order = jpeg_natural_order; + cinfo->lim_Se = DCTSIZE2 - 1; +#endif + + /* We initialize DCT_scaled_size and min_DCT_scaled_size to DCTSIZE in lossy + * mode. In the full decompressor, this will be overridden by jdmaster.c; + * but in the transcoder, jdmaster.c is not used, so we must do it here. + */ +#if JPEG_LIB_VERSION >= 70 + cinfo->min_DCT_h_scaled_size = cinfo->min_DCT_v_scaled_size = data_unit; +#else + cinfo->min_DCT_scaled_size = data_unit; +#endif + + /* Compute dimensions of components */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { +#if JPEG_LIB_VERSION >= 70 + compptr->DCT_h_scaled_size = compptr->DCT_v_scaled_size = data_unit; +#else + compptr->DCT_scaled_size = data_unit; +#endif + /* Size in data units */ + compptr->width_in_blocks = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * (long)compptr->h_samp_factor, + (long)(cinfo->max_h_samp_factor * data_unit)); + compptr->height_in_blocks = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * (long)compptr->v_samp_factor, + (long)(cinfo->max_v_samp_factor * data_unit)); + /* Set the first and last MCU columns to decompress from multi-scan images. + * By default, decompress all of the MCU columns. + */ + cinfo->master->first_MCU_col[ci] = 0; + cinfo->master->last_MCU_col[ci] = compptr->width_in_blocks - 1; + /* downsampled_width and downsampled_height will also be overridden by + * jdmaster.c if we are doing full decompression. The transcoder library + * doesn't use these values, but the calling application might. + */ + /* Size in samples */ + compptr->downsampled_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * (long)compptr->h_samp_factor, + (long)cinfo->max_h_samp_factor); + compptr->downsampled_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * (long)compptr->v_samp_factor, + (long)cinfo->max_v_samp_factor); + /* Mark component needed, until color conversion says otherwise */ + compptr->component_needed = TRUE; + /* Mark no quantization table yet saved for component */ + compptr->quant_table = NULL; + } + + /* Compute number of fully interleaved MCU rows. */ + cinfo->total_iMCU_rows = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height, + (long)(cinfo->max_v_samp_factor * data_unit)); + + /* Decide whether file contains multiple scans */ + if (cinfo->comps_in_scan < cinfo->num_components || cinfo->progressive_mode) + cinfo->inputctl->has_multiple_scans = TRUE; + else + cinfo->inputctl->has_multiple_scans = FALSE; +} + + +LOCAL(void) +per_scan_setup(j_decompress_ptr cinfo) +/* Do computations that are needed before processing a JPEG scan */ +/* cinfo->comps_in_scan and cinfo->cur_comp_info[] were set from SOS marker */ +{ + int ci, mcublks, tmp; + jpeg_component_info *compptr; + int data_unit = cinfo->master->lossless ? 1 : DCTSIZE; + + if (cinfo->comps_in_scan == 1) { + + /* Noninterleaved (single-component) scan */ + compptr = cinfo->cur_comp_info[0]; + + /* Overall image size in MCUs */ + cinfo->MCUs_per_row = compptr->width_in_blocks; + cinfo->MCU_rows_in_scan = compptr->height_in_blocks; + + /* For noninterleaved scan, always one data unit per MCU */ + compptr->MCU_width = 1; + compptr->MCU_height = 1; + compptr->MCU_blocks = 1; + compptr->MCU_sample_width = compptr->_DCT_scaled_size; + compptr->last_col_width = 1; + /* For noninterleaved scans, it is convenient to define last_row_height + * as the number of data unit rows present in the last iMCU row. + */ + tmp = (int)(compptr->height_in_blocks % compptr->v_samp_factor); + if (tmp == 0) tmp = compptr->v_samp_factor; + compptr->last_row_height = tmp; + + /* Prepare array describing MCU composition */ + cinfo->blocks_in_MCU = 1; + cinfo->MCU_membership[0] = 0; + + } else { + + /* Interleaved (multi-component) scan */ + if (cinfo->comps_in_scan <= 0 || cinfo->comps_in_scan > MAX_COMPS_IN_SCAN) + ERREXIT2(cinfo, JERR_COMPONENT_COUNT, cinfo->comps_in_scan, + MAX_COMPS_IN_SCAN); + + /* Overall image size in MCUs */ + cinfo->MCUs_per_row = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width, + (long)(cinfo->max_h_samp_factor * data_unit)); + cinfo->MCU_rows_in_scan = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height, + (long)(cinfo->max_v_samp_factor * data_unit)); + + cinfo->blocks_in_MCU = 0; + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* Sampling factors give # of data units of component in each MCU */ + compptr->MCU_width = compptr->h_samp_factor; + compptr->MCU_height = compptr->v_samp_factor; + compptr->MCU_blocks = compptr->MCU_width * compptr->MCU_height; + compptr->MCU_sample_width = compptr->MCU_width * + compptr->_DCT_scaled_size; + /* Figure number of non-dummy data units in last MCU column & row */ + tmp = (int)(compptr->width_in_blocks % compptr->MCU_width); + if (tmp == 0) tmp = compptr->MCU_width; + compptr->last_col_width = tmp; + tmp = (int)(compptr->height_in_blocks % compptr->MCU_height); + if (tmp == 0) tmp = compptr->MCU_height; + compptr->last_row_height = tmp; + /* Prepare array describing MCU composition */ + mcublks = compptr->MCU_blocks; + if (cinfo->blocks_in_MCU + mcublks > D_MAX_BLOCKS_IN_MCU) + ERREXIT(cinfo, JERR_BAD_MCU_SIZE); + while (mcublks-- > 0) { + cinfo->MCU_membership[cinfo->blocks_in_MCU++] = ci; + } + } + + } +} + + +/* + * Save away a copy of the Q-table referenced by each component present + * in the current scan, unless already saved during a prior scan. + * + * In a multiple-scan JPEG file, the encoder could assign different components + * the same Q-table slot number, but change table definitions between scans + * so that each component uses a different Q-table. (The IJG encoder is not + * currently capable of doing this, but other encoders might.) Since we want + * to be able to dequantize all the components at the end of the file, this + * means that we have to save away the table actually used for each component. + * We do this by copying the table at the start of the first scan containing + * the component. + * Rec. ITU-T T.81 | ISO/IEC 10918-1 prohibits the encoder from changing the + * contents of a Q-table slot between scans of a component using that slot. If + * the encoder does so anyway, this decoder will simply use the Q-table values + * that were current at the start of the first scan for the component. + * + * The decompressor output side looks only at the saved quant tables, + * not at the current Q-table slots. + */ + +LOCAL(void) +latch_quant_tables(j_decompress_ptr cinfo) +{ + int ci, qtblno; + jpeg_component_info *compptr; + JQUANT_TBL *qtbl; + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* No work if we already saved Q-table for this component */ + if (compptr->quant_table != NULL) + continue; + /* Make sure specified quantization table is present */ + qtblno = compptr->quant_tbl_no; + if (qtblno < 0 || qtblno >= NUM_QUANT_TBLS || + cinfo->quant_tbl_ptrs[qtblno] == NULL) + ERREXIT1(cinfo, JERR_NO_QUANT_TABLE, qtblno); + /* OK, save away the quantization table */ + qtbl = (JQUANT_TBL *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(JQUANT_TBL)); + memcpy(qtbl, cinfo->quant_tbl_ptrs[qtblno], sizeof(JQUANT_TBL)); + compptr->quant_table = qtbl; + } +} + + +/* + * Initialize the input modules to read a scan of compressed data. + * The first call to this is done by jdmaster.c after initializing + * the entire decompressor (during jpeg_start_decompress). + * Subsequent calls come from consume_markers, below. + */ + +METHODDEF(void) +start_input_pass(j_decompress_ptr cinfo) +{ + per_scan_setup(cinfo); + if (!cinfo->master->lossless) + latch_quant_tables(cinfo); + (*cinfo->entropy->start_pass) (cinfo); + (*cinfo->coef->start_input_pass) (cinfo); + cinfo->inputctl->consume_input = cinfo->coef->consume_data; +} + + +/* + * Finish up after inputting a compressed-data scan. + * This is called by the coefficient or difference controller after it's read + * all the expected data of the scan. + */ + +METHODDEF(void) +finish_input_pass(j_decompress_ptr cinfo) +{ + cinfo->inputctl->consume_input = consume_markers; +} + + +/* + * Read JPEG markers before, between, or after compressed-data scans. + * Change state as necessary when a new scan is reached. + * Return value is JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI. + * + * The consume_input method pointer points either here or to the + * coefficient or difference controller's consume_data routine, depending on + * whether we are reading a compressed data segment or inter-segment markers. + */ + +METHODDEF(int) +consume_markers(j_decompress_ptr cinfo) +{ + my_inputctl_ptr inputctl = (my_inputctl_ptr)cinfo->inputctl; + int val; + + if (inputctl->pub.eoi_reached) /* After hitting EOI, read no further */ + return JPEG_REACHED_EOI; + + val = (*cinfo->marker->read_markers) (cinfo); + + switch (val) { + case JPEG_REACHED_SOS: /* Found SOS */ + if (inputctl->inheaders) { /* 1st SOS */ + initial_setup(cinfo); + inputctl->inheaders = FALSE; + /* Note: start_input_pass must be called by jdmaster.c + * before any more input can be consumed. jdapimin.c is + * responsible for enforcing this sequencing. + */ + } else { /* 2nd or later SOS marker */ + if (!inputctl->pub.has_multiple_scans) + ERREXIT(cinfo, JERR_EOI_EXPECTED); /* Oops, I wasn't expecting this! */ + start_input_pass(cinfo); + } + break; + case JPEG_REACHED_EOI: /* Found EOI */ + inputctl->pub.eoi_reached = TRUE; + if (inputctl->inheaders) { /* Tables-only datastream, apparently */ + if (cinfo->marker->saw_SOF) + ERREXIT(cinfo, JERR_SOF_NO_SOS); + } else { + /* Prevent infinite loop in coef ctlr's decompress_data routine + * if user set output_scan_number larger than number of scans. + */ + if (cinfo->output_scan_number > cinfo->input_scan_number) + cinfo->output_scan_number = cinfo->input_scan_number; + } + break; + case JPEG_SUSPENDED: + break; + } + + return val; +} + + +/* + * Reset state to begin a fresh datastream. + */ + +METHODDEF(void) +reset_input_controller(j_decompress_ptr cinfo) +{ + my_inputctl_ptr inputctl = (my_inputctl_ptr)cinfo->inputctl; + + inputctl->pub.consume_input = consume_markers; + inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */ + inputctl->pub.eoi_reached = FALSE; + inputctl->inheaders = TRUE; + /* Reset other modules */ + (*cinfo->err->reset_error_mgr) ((j_common_ptr)cinfo); + (*cinfo->marker->reset_marker_reader) (cinfo); + /* Reset progression state -- would be cleaner if entropy decoder did this */ + cinfo->coef_bits = NULL; +} + + +/* + * Initialize the input controller module. + * This is called only once, when the decompression object is created. + */ + +GLOBAL(void) +jinit_input_controller(j_decompress_ptr cinfo) +{ + my_inputctl_ptr inputctl; + + /* Create subobject in permanent pool */ + inputctl = (my_inputctl_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(my_input_controller)); + cinfo->inputctl = (struct jpeg_input_controller *)inputctl; + /* Initialize method pointers */ + inputctl->pub.consume_input = consume_markers; + inputctl->pub.reset_input_controller = reset_input_controller; + inputctl->pub.start_input_pass = start_input_pass; + inputctl->pub.finish_input_pass = finish_input_pass; + /* Initialize state: can't use reset_input_controller since we don't + * want to try to reset other modules yet. + */ + inputctl->pub.has_multiple_scans = FALSE; /* "unknown" would be better */ + inputctl->pub.eoi_reached = FALSE; + inputctl->inheaders = TRUE; +} diff --git a/src/spice2x/external/libjpeg-turbo/jdlhuff.c b/src/spice2x/external/libjpeg-turbo/jdlhuff.c new file mode 100644 index 0000000..9964830 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdlhuff.c @@ -0,0 +1,302 @@ +/* + * jdlhuff.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains Huffman entropy decoding routines for lossless JPEG. + * + * Much of the complexity here has to do with supporting input suspension. + * If the data source module demands suspension, we want to be able to back + * up to the start of the current MCU. To do this, we copy state variables + * into local working storage, and update them back to the permanent + * storage only upon successful completion of an MCU. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jlossls.h" /* Private declarations for lossless codec */ +#include "jdhuff.h" /* Declarations shared with jd*huff.c */ + + +#ifdef D_LOSSLESS_SUPPORTED + +typedef struct { + int ci, yoffset, MCU_width; +} lhd_output_ptr_info; + +/* + * Expanded entropy decoder object for Huffman decoding in lossless mode. + */ + +typedef struct { + struct jpeg_entropy_decoder pub; /* public fields */ + + /* These fields are loaded into local variables at start of each MCU. + * In case of suspension, we exit WITHOUT updating them. + */ + bitread_perm_state bitstate; /* Bit buffer at start of MCU */ + + /* Pointers to derived tables (these workspaces have image lifespan) */ + d_derived_tbl *derived_tbls[NUM_HUFF_TBLS]; + + /* Precalculated info set up by start_pass for use in decode_mcus: */ + + /* Pointers to derived tables to be used for each data unit within an MCU */ + d_derived_tbl *cur_tbls[D_MAX_BLOCKS_IN_MCU]; + + /* Pointers to the proper output difference row for each group of data units + * within an MCU. For each component, there are Vi groups of Hi data units. + */ + JDIFFROW output_ptr[D_MAX_BLOCKS_IN_MCU]; + + /* Number of output pointers in use for the current MCU. This is the sum + * of all Vi in the MCU. + */ + int num_output_ptrs; + + /* Information used for positioning the output pointers within the output + * difference rows. + */ + lhd_output_ptr_info output_ptr_info[D_MAX_BLOCKS_IN_MCU]; + + /* Index of the proper output pointer for each data unit within an MCU */ + int output_ptr_index[D_MAX_BLOCKS_IN_MCU]; + +} lhuff_entropy_decoder; + +typedef lhuff_entropy_decoder *lhuff_entropy_ptr; + + +/* + * Initialize for a Huffman-compressed scan. + */ + +METHODDEF(void) +start_pass_lhuff_decoder(j_decompress_ptr cinfo) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + int ci, dctbl, sampn, ptrn, yoffset, xoffset; + jpeg_component_info *compptr; + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + dctbl = compptr->dc_tbl_no; + /* Make sure requested tables are present */ + if (dctbl < 0 || dctbl >= NUM_HUFF_TBLS || + cinfo->dc_huff_tbl_ptrs[dctbl] == NULL) + ERREXIT1(cinfo, JERR_NO_HUFF_TABLE, dctbl); + /* Compute derived values for Huffman tables */ + /* We may do this more than once for a table, but it's not expensive */ + jpeg_make_d_derived_tbl(cinfo, TRUE, dctbl, + &entropy->derived_tbls[dctbl]); + } + + /* Precalculate decoding info for each sample in an MCU of this scan */ + for (sampn = 0, ptrn = 0; sampn < cinfo->blocks_in_MCU;) { + compptr = cinfo->cur_comp_info[cinfo->MCU_membership[sampn]]; + ci = compptr->component_index; + for (yoffset = 0; yoffset < compptr->MCU_height; yoffset++, ptrn++) { + /* Precalculate the setup info for each output pointer */ + entropy->output_ptr_info[ptrn].ci = ci; + entropy->output_ptr_info[ptrn].yoffset = yoffset; + entropy->output_ptr_info[ptrn].MCU_width = compptr->MCU_width; + for (xoffset = 0; xoffset < compptr->MCU_width; xoffset++, sampn++) { + /* Precalculate the output pointer index for each sample */ + entropy->output_ptr_index[sampn] = ptrn; + /* Precalculate which table to use for each sample */ + entropy->cur_tbls[sampn] = entropy->derived_tbls[compptr->dc_tbl_no]; + } + } + } + entropy->num_output_ptrs = ptrn; + + /* Initialize bitread state variables */ + entropy->bitstate.bits_left = 0; + entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */ + entropy->pub.insufficient_data = FALSE; +} + + +/* + * Figure F.12: extend sign bit. + * On some machines, a shift and add will be faster than a table lookup. + */ + +#define AVOID_TABLES +#ifdef AVOID_TABLES + +#define NEG_1 ((unsigned int)-1) +#define HUFF_EXTEND(x, s) \ + ((x) + ((((x) - (1 << ((s) - 1))) >> 31) & (((NEG_1) << (s)) + 1))) + +#else + +#define HUFF_EXTEND(x, s) \ + ((x) < extend_test[s] ? (x) + extend_offset[s] : (x)) + +static const int extend_test[16] = { /* entry n is 2**(n-1) */ + 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, + 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 +}; + +static const int extend_offset[16] = { /* entry n is (-1 << n) + 1 */ + 0, ((-1) << 1) + 1, ((-1) << 2) + 1, ((-1) << 3) + 1, ((-1) << 4) + 1, + ((-1) << 5) + 1, ((-1) << 6) + 1, ((-1) << 7) + 1, ((-1) << 8) + 1, + ((-1) << 9) + 1, ((-1) << 10) + 1, ((-1) << 11) + 1, ((-1) << 12) + 1, + ((-1) << 13) + 1, ((-1) << 14) + 1, ((-1) << 15) + 1 +}; + +#endif /* AVOID_TABLES */ + + +/* + * Check for a restart marker & resynchronize decoder. + * Returns FALSE if must suspend. + */ + +LOCAL(boolean) +process_restart(j_decompress_ptr cinfo) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + + /* Throw away any unused bits remaining in bit buffer; */ + /* include any full bytes in next_marker's count of discarded bytes */ + cinfo->marker->discarded_bytes += entropy->bitstate.bits_left / 8; + entropy->bitstate.bits_left = 0; + + /* Advance past the RSTn marker */ + if (!(*cinfo->marker->read_restart_marker) (cinfo)) + return FALSE; + + /* Reset out-of-data flag, unless read_restart_marker left us smack up + * against a marker. In that case we will end up treating the next data + * segment as empty, and we can avoid producing bogus output pixels by + * leaving the flag set. + */ + if (cinfo->unread_marker == 0) + entropy->pub.insufficient_data = FALSE; + + return TRUE; +} + + +/* + * Decode and return nMCU MCUs' worth of Huffman-compressed differences. + * Each MCU is also disassembled and placed accordingly in diff_buf. + * + * MCU_col_num specifies the column of the first MCU being requested within + * the MCU row. This tells us where to position the output row pointers in + * diff_buf. + * + * Returns the number of MCUs decoded. This may be less than nMCU MCUs if + * data source requested suspension. In that case no changes have been made + * to permanent state. (Exception: some output differences may already have + * been assigned. This is harmless for this module, since we'll just + * re-assign them on the next call.) + */ + +METHODDEF(JDIMENSION) +decode_mcus(j_decompress_ptr cinfo, JDIFFIMAGE diff_buf, + JDIMENSION MCU_row_num, JDIMENSION MCU_col_num, JDIMENSION nMCU) +{ + lhuff_entropy_ptr entropy = (lhuff_entropy_ptr)cinfo->entropy; + int sampn, ci, yoffset, MCU_width, ptrn; + JDIMENSION mcu_num; + BITREAD_STATE_VARS; + + /* Set output pointer locations based on MCU_col_num */ + for (ptrn = 0; ptrn < entropy->num_output_ptrs; ptrn++) { + ci = entropy->output_ptr_info[ptrn].ci; + yoffset = entropy->output_ptr_info[ptrn].yoffset; + MCU_width = entropy->output_ptr_info[ptrn].MCU_width; + entropy->output_ptr[ptrn] = + diff_buf[ci][MCU_row_num + yoffset] + (MCU_col_num * MCU_width); + } + + /* + * If we've run out of data, zero out the buffers and return. + * By resetting the undifferencer, the output samples will be CENTERJSAMPLE. + * + * NB: We should find a way to do this without interacting with the + * undifferencer module directly. + */ + if (entropy->pub.insufficient_data) { + for (ptrn = 0; ptrn < entropy->num_output_ptrs; ptrn++) + jzero_far((void FAR *)entropy->output_ptr[ptrn], + nMCU * entropy->output_ptr_info[ptrn].MCU_width * + sizeof(JDIFF)); + + (*cinfo->idct->start_pass) (cinfo); + + } else { + + /* Load up working state */ + BITREAD_LOAD_STATE(cinfo, entropy->bitstate); + + /* Outer loop handles the number of MCUs requested */ + + for (mcu_num = 0; mcu_num < nMCU; mcu_num++) { + + /* Inner loop handles the samples in the MCU */ + for (sampn = 0; sampn < cinfo->blocks_in_MCU; sampn++) { + d_derived_tbl *dctbl = entropy->cur_tbls[sampn]; + register int s, r; + + /* Section H.2.2: decode the sample difference */ + HUFF_DECODE(s, br_state, dctbl, return mcu_num, label1); + if (s) { + if (s == 16) /* special case: always output 32768 */ + s = 32768; + else { /* normal case: fetch subsequent bits */ + CHECK_BIT_BUFFER(br_state, s, return mcu_num); + r = GET_BITS(s); + s = HUFF_EXTEND(r, s); + } + } + + /* Output the sample difference */ + *entropy->output_ptr[entropy->output_ptr_index[sampn]]++ = (JDIFF)s; + } + + /* Completed MCU, so update state */ + BITREAD_SAVE_STATE(cinfo, entropy->bitstate); + } + } + + return nMCU; +} + + +/* + * Module initialization routine for lossless mode Huffman entropy decoding. + */ + +GLOBAL(void) +jinit_lhuff_decoder(j_decompress_ptr cinfo) +{ + lhuff_entropy_ptr entropy; + int i; + + entropy = (lhuff_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(lhuff_entropy_decoder)); + cinfo->entropy = (struct jpeg_entropy_decoder *)entropy; + entropy->pub.start_pass = start_pass_lhuff_decoder; + entropy->pub.decode_mcus = decode_mcus; + entropy->pub.process_restart = process_restart; + + /* Mark tables unallocated */ + for (i = 0; i < NUM_HUFF_TBLS; i++) { + entropy->derived_tbls[i] = NULL; + } +} + +#endif /* D_LOSSLESS_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jdlossls.c b/src/spice2x/external/libjpeg-turbo/jdlossls.c new file mode 100644 index 0000000..4d15e6b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdlossls.c @@ -0,0 +1,289 @@ +/* + * jdlossls.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1998, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains prediction, sample undifferencing, point transform, and + * sample scaling routines for the lossless JPEG decompressor. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jlossls.h" + +#ifdef D_LOSSLESS_SUPPORTED + + +/**************** Sample undifferencing (reconstruction) *****************/ + +/* + * In order to avoid a performance penalty for checking which predictor is + * being used and which row is being processed for each call of the + * undifferencer, and to promote optimization, we have separate undifferencing + * functions for each predictor selection value. + * + * We are able to avoid duplicating source code by implementing the predictors + * and undifferencers as macros. Each of the undifferencing functions is + * simply a wrapper around an UNDIFFERENCE macro with the appropriate PREDICTOR + * macro passed as an argument. + */ + +/* Predictor for the first column of the first row: 2^(P-Pt-1) */ +#define INITIAL_PREDICTORx (1 << (cinfo->data_precision - cinfo->Al - 1)) + +/* Predictor for the first column of the remaining rows: Rb */ +#define INITIAL_PREDICTOR2 prev_row[0] + + +/* + * 1-Dimensional undifferencer routine. + * + * This macro implements the 1-D horizontal predictor (1). INITIAL_PREDICTOR + * is used as the special case predictor for the first column, which must be + * either INITIAL_PREDICTOR2 or INITIAL_PREDICTORx. The remaining samples + * use PREDICTOR1. + * + * The reconstructed sample is supposed to be calculated modulo 2^16, so we + * logically AND the result with 0xFFFF. + */ + +#define UNDIFFERENCE_1D(INITIAL_PREDICTOR) \ + int Ra; \ + \ + Ra = (*diff_buf++ + INITIAL_PREDICTOR) & 0xFFFF; \ + *undiff_buf++ = Ra; \ + \ + while (--width) { \ + Ra = (*diff_buf++ + PREDICTOR1) & 0xFFFF; \ + *undiff_buf++ = Ra; \ + } + + +/* + * 2-Dimensional undifferencer routine. + * + * This macro implements the 2-D horizontal predictors (#2-7). PREDICTOR2 is + * used as the special case predictor for the first column. The remaining + * samples use PREDICTOR, which is a function of Ra, Rb, and Rc. + * + * Because prev_row and output_buf may point to the same storage area (in an + * interleaved image with Vi=1, for example), we must take care to buffer Rb/Rc + * before writing the current reconstructed sample value into output_buf. + * + * The reconstructed sample is supposed to be calculated modulo 2^16, so we + * logically AND the result with 0xFFFF. + */ + +#define UNDIFFERENCE_2D(PREDICTOR) \ + int Ra, Rb, Rc; \ + \ + Rb = *prev_row++; \ + Ra = (*diff_buf++ + PREDICTOR2) & 0xFFFF; \ + *undiff_buf++ = Ra; \ + \ + while (--width) { \ + Rc = Rb; \ + Rb = *prev_row++; \ + Ra = (*diff_buf++ + PREDICTOR) & 0xFFFF; \ + *undiff_buf++ = Ra; \ + } + + +/* + * Undifferencers for the second and subsequent rows in a scan or restart + * interval. The first sample in the row is undifferenced using the vertical + * predictor (2). The rest of the samples are undifferenced using the + * predictor specified in the scan header. + */ + +METHODDEF(void) +jpeg_undifference1(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + UNDIFFERENCE_1D(INITIAL_PREDICTOR2); +} + +METHODDEF(void) +jpeg_undifference2(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + UNDIFFERENCE_2D(PREDICTOR2); + (void)(Rc); +} + +METHODDEF(void) +jpeg_undifference3(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + UNDIFFERENCE_2D(PREDICTOR3); +} + +METHODDEF(void) +jpeg_undifference4(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + UNDIFFERENCE_2D(PREDICTOR4); +} + +METHODDEF(void) +jpeg_undifference5(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + UNDIFFERENCE_2D(PREDICTOR5); +} + +METHODDEF(void) +jpeg_undifference6(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + UNDIFFERENCE_2D(PREDICTOR6); +} + +METHODDEF(void) +jpeg_undifference7(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + UNDIFFERENCE_2D(PREDICTOR7); + (void)(Rc); +} + + +/* + * Undifferencer for the first row in a scan or restart interval. The first + * sample in the row is undifferenced using the special predictor constant + * x=2^(P-Pt-1). The rest of the samples are undifferenced using the + * 1-D horizontal predictor (1). + */ + +METHODDEF(void) +jpeg_undifference_first_row(j_decompress_ptr cinfo, int comp_index, + JDIFFROW diff_buf, JDIFFROW prev_row, + JDIFFROW undiff_buf, JDIMENSION width) +{ + lossless_decomp_ptr losslessd = (lossless_decomp_ptr)cinfo->idct; + + UNDIFFERENCE_1D(INITIAL_PREDICTORx); + + /* + * Now that we have undifferenced the first row, we want to use the + * undifferencer that corresponds to the predictor specified in the + * scan header. + */ + switch (cinfo->Ss) { + case 1: + losslessd->predict_undifference[comp_index] = jpeg_undifference1; + break; + case 2: + losslessd->predict_undifference[comp_index] = jpeg_undifference2; + break; + case 3: + losslessd->predict_undifference[comp_index] = jpeg_undifference3; + break; + case 4: + losslessd->predict_undifference[comp_index] = jpeg_undifference4; + break; + case 5: + losslessd->predict_undifference[comp_index] = jpeg_undifference5; + break; + case 6: + losslessd->predict_undifference[comp_index] = jpeg_undifference6; + break; + case 7: + losslessd->predict_undifference[comp_index] = jpeg_undifference7; + break; + } +} + + +/*********************** Sample upscaling by 2^Pt ************************/ + +METHODDEF(void) +simple_upscale(j_decompress_ptr cinfo, + JDIFFROW diff_buf, _JSAMPROW output_buf, JDIMENSION width) +{ + do { + *output_buf++ = (_JSAMPLE)(*diff_buf++ << cinfo->Al); + } while (--width); +} + +METHODDEF(void) +noscale(j_decompress_ptr cinfo, + JDIFFROW diff_buf, _JSAMPROW output_buf, JDIMENSION width) +{ + do { + *output_buf++ = (_JSAMPLE)(*diff_buf++); + } while (--width); +} + + +/* + * Initialize for an input processing pass. + */ + +METHODDEF(void) +start_pass_lossless(j_decompress_ptr cinfo) +{ + lossless_decomp_ptr losslessd = (lossless_decomp_ptr)cinfo->idct; + int ci; + + /* Check that the scan parameters Ss, Se, Ah, Al are OK for lossless JPEG. + * + * Ss is the predictor selection value (psv). Legal values for sequential + * lossless JPEG are: 1 <= psv <= 7. + * + * Se and Ah are not used and should be zero. + * + * Al specifies the point transform (Pt). + * Legal values are: 0 <= Pt <= (data precision - 1). + */ + if (cinfo->Ss < 1 || cinfo->Ss > 7 || + cinfo->Se != 0 || cinfo->Ah != 0 || + cinfo->Al < 0 || cinfo->Al >= cinfo->data_precision) + ERREXIT4(cinfo, JERR_BAD_PROGRESSION, + cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al); + + /* Set undifference functions to first row function */ + for (ci = 0; ci < cinfo->num_components; ci++) + losslessd->predict_undifference[ci] = jpeg_undifference_first_row; + + /* Set scaler function based on Pt */ + if (cinfo->Al) + losslessd->scaler_scale = simple_upscale; + else + losslessd->scaler_scale = noscale; +} + + +/* + * Initialize the lossless decompressor. + */ + +GLOBAL(void) +_jinit_lossless_decompressor(j_decompress_ptr cinfo) +{ + lossless_decomp_ptr losslessd; + + /* Create subobject in permanent pool */ + losslessd = (lossless_decomp_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(jpeg_lossless_decompressor)); + cinfo->idct = (struct jpeg_inverse_dct *)losslessd; + losslessd->pub.start_pass = start_pass_lossless; +} + +#endif /* D_LOSSLESS_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jdmainct.c b/src/spice2x/external/libjpeg-turbo/jdmainct.c new file mode 100644 index 0000000..c672b4b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmainct.c @@ -0,0 +1,469 @@ +/* + * jdmainct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2010, 2016, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the main buffer controller for decompression. + * The main buffer lies between the JPEG decompressor proper and the + * post-processor; it holds downsampled data in the JPEG colorspace. + * + * Note that this code is bypassed in raw-data mode, since the application + * supplies the equivalent of the main buffer in that case. + */ + +#include "jinclude.h" +#include "jdmainct.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) + +/* + * In the current system design, the main buffer need never be a full-image + * buffer; any full-height buffers will be found inside the coefficient, + * difference, or postprocessing controllers. Nonetheless, the main controller + * is not trivial. Its responsibility is to provide context rows for + * upsampling/rescaling, and doing this in an efficient fashion is a bit + * tricky. + * + * Postprocessor input data is counted in "row groups". A row group + * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size) + * sample rows of each component. (We require DCT_scaled_size values to be + * chosen such that these numbers are integers. In practice DCT_scaled_size + * values will likely be powers of two, so we actually have the stronger + * condition that DCT_scaled_size / min_DCT_scaled_size is an integer.) + * Upsampling will typically produce max_v_samp_factor pixel rows from each + * row group (times any additional scale factor that the upsampler is + * applying). + * + * The coefficient or difference controller will deliver data to us one iMCU + * row at a time; each iMCU row contains v_samp_factor * DCT_scaled_size sample + * rows, or exactly min_DCT_scaled_size row groups. (This amount of data + * corresponds to one row of MCUs when the image is fully interleaved.) Note + * that the number of sample rows varies across components, but the number of + * row groups does not. Some garbage sample rows may be included in the last + * iMCU row at the bottom of the image. + * + * Depending on the vertical scaling algorithm used, the upsampler may need + * access to the sample row(s) above and below its current input row group. + * The upsampler is required to set need_context_rows TRUE at global selection + * time if so. When need_context_rows is FALSE, this controller can simply + * obtain one iMCU row at a time from the coefficient or difference controller + * and dole it out as row groups to the postprocessor. + * + * When need_context_rows is TRUE, this controller guarantees that the buffer + * passed to postprocessing contains at least one row group's worth of samples + * above and below the row group(s) being processed. Note that the context + * rows "above" the first passed row group appear at negative row offsets in + * the passed buffer. At the top and bottom of the image, the required + * context rows are manufactured by duplicating the first or last real sample + * row; this avoids having special cases in the upsampling inner loops. + * + * The amount of context is fixed at one row group just because that's a + * convenient number for this controller to work with. The existing + * upsamplers really only need one sample row of context. An upsampler + * supporting arbitrary output rescaling might wish for more than one row + * group of context when shrinking the image; tough, we don't handle that. + * (This is justified by the assumption that downsizing will be handled mostly + * by adjusting the DCT_scaled_size values, so that the actual scale factor at + * the upsample step needn't be much less than one.) + * + * To provide the desired context, we have to retain the last two row groups + * of one iMCU row while reading in the next iMCU row. (The last row group + * can't be processed until we have another row group for its below-context, + * and so we have to save the next-to-last group too for its above-context.) + * We could do this most simply by copying data around in our buffer, but + * that'd be very slow. We can avoid copying any data by creating a rather + * strange pointer structure. Here's how it works. We allocate a workspace + * consisting of M+2 row groups (where M = min_DCT_scaled_size is the number + * of row groups per iMCU row). We create two sets of redundant pointers to + * the workspace. Labeling the physical row groups 0 to M+1, the synthesized + * pointer lists look like this: + * M+1 M-1 + * master pointer --> 0 master pointer --> 0 + * 1 1 + * ... ... + * M-3 M-3 + * M-2 M + * M-1 M+1 + * M M-2 + * M+1 M-1 + * 0 0 + * We read alternate iMCU rows using each master pointer; thus the last two + * row groups of the previous iMCU row remain un-overwritten in the workspace. + * The pointer lists are set up so that the required context rows appear to + * be adjacent to the proper places when we pass the pointer lists to the + * upsampler. + * + * The above pictures describe the normal state of the pointer lists. + * At top and bottom of the image, we diddle the pointer lists to duplicate + * the first or last sample row as necessary (this is cheaper than copying + * sample rows around). + * + * This scheme breaks down if M < 2, ie, min_DCT_scaled_size is 1. In that + * situation each iMCU row provides only one row group so the buffering logic + * must be different (eg, we must read two iMCU rows before we can emit the + * first row group). For now, we simply do not support providing context + * rows when min_DCT_scaled_size is 1. That combination seems unlikely to + * be worth providing --- if someone wants a 1/8th-size preview, they probably + * want it quick and dirty, so a context-free upsampler is sufficient. + */ + + +/* Forward declarations */ +METHODDEF(void) process_data_simple_main(j_decompress_ptr cinfo, + _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +METHODDEF(void) process_data_context_main(j_decompress_ptr cinfo, + _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +#ifdef QUANT_2PASS_SUPPORTED +METHODDEF(void) process_data_crank_post(j_decompress_ptr cinfo, + _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +#endif + + +LOCAL(void) +alloc_funny_pointers(j_decompress_ptr cinfo) +/* Allocate space for the funny pointer lists. + * This is done only once, not once per pass. + */ +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + int ci, rgroup; + int M = cinfo->_min_DCT_scaled_size; + jpeg_component_info *compptr; + _JSAMPARRAY xbuf; + + /* Get top-level space for component array pointers. + * We alloc both arrays with one call to save a few cycles. + */ + main_ptr->xbuffer[0] = (_JSAMPIMAGE) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + cinfo->num_components * 2 * + sizeof(_JSAMPARRAY)); + main_ptr->xbuffer[1] = main_ptr->xbuffer[0] + cinfo->num_components; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / + cinfo->_min_DCT_scaled_size; /* height of a row group of component */ + /* Get space for pointer lists --- M+4 row groups in each list. + * We alloc both pointer lists with one call to save a few cycles. + */ + xbuf = (_JSAMPARRAY) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + 2 * (rgroup * (M + 4)) * sizeof(_JSAMPROW)); + xbuf += rgroup; /* want one row group at negative offsets */ + main_ptr->xbuffer[0][ci] = xbuf; + xbuf += rgroup * (M + 4); + main_ptr->xbuffer[1][ci] = xbuf; + } +} + + +LOCAL(void) +make_funny_pointers(j_decompress_ptr cinfo) +/* Create the funny pointer lists discussed in the comments above. + * The actual workspace is already allocated (in main_ptr->buffer), + * and the space for the pointer lists is allocated too. + * This routine just fills in the curiously ordered lists. + * This will be repeated at the beginning of each pass. + */ +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + int ci, i, rgroup; + int M = cinfo->_min_DCT_scaled_size; + jpeg_component_info *compptr; + _JSAMPARRAY buf, xbuf0, xbuf1; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / + cinfo->_min_DCT_scaled_size; /* height of a row group of component */ + xbuf0 = main_ptr->xbuffer[0][ci]; + xbuf1 = main_ptr->xbuffer[1][ci]; + /* First copy the workspace pointers as-is */ + buf = main_ptr->buffer[ci]; + for (i = 0; i < rgroup * (M + 2); i++) { + xbuf0[i] = xbuf1[i] = buf[i]; + } + /* In the second list, put the last four row groups in swapped order */ + for (i = 0; i < rgroup * 2; i++) { + xbuf1[rgroup * (M - 2) + i] = buf[rgroup * M + i]; + xbuf1[rgroup * M + i] = buf[rgroup * (M - 2) + i]; + } + /* The wraparound pointers at top and bottom will be filled later + * (see set_wraparound_pointers, below). Initially we want the "above" + * pointers to duplicate the first actual data line. This only needs + * to happen in xbuffer[0]. + */ + for (i = 0; i < rgroup; i++) { + xbuf0[i - rgroup] = xbuf0[0]; + } + } +} + + +LOCAL(void) +set_bottom_pointers(j_decompress_ptr cinfo) +/* Change the pointer lists to duplicate the last sample row at the bottom + * of the image. whichptr indicates which xbuffer holds the final iMCU row. + * Also sets rowgroups_avail to indicate number of nondummy row groups in row. + */ +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + int ci, i, rgroup, iMCUheight, rows_left; + jpeg_component_info *compptr; + _JSAMPARRAY xbuf; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Count sample rows in one iMCU row and in one row group */ + iMCUheight = compptr->v_samp_factor * compptr->_DCT_scaled_size; + rgroup = iMCUheight / cinfo->_min_DCT_scaled_size; + /* Count nondummy sample rows remaining for this component */ + rows_left = (int)(compptr->downsampled_height % (JDIMENSION)iMCUheight); + if (rows_left == 0) rows_left = iMCUheight; + /* Count nondummy row groups. Should get same answer for each component, + * so we need only do it once. + */ + if (ci == 0) { + main_ptr->rowgroups_avail = (JDIMENSION)((rows_left - 1) / rgroup + 1); + } + /* Duplicate the last real sample row rgroup*2 times; this pads out the + * last partial rowgroup and ensures at least one full rowgroup of context. + */ + xbuf = main_ptr->xbuffer[main_ptr->whichptr][ci]; + for (i = 0; i < rgroup * 2; i++) { + xbuf[rows_left + i] = xbuf[rows_left - 1]; + } + } +} + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_main(j_decompress_ptr cinfo, J_BUF_MODE pass_mode) +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + + switch (pass_mode) { + case JBUF_PASS_THRU: + if (cinfo->upsample->need_context_rows) { + main_ptr->pub._process_data = process_data_context_main; + make_funny_pointers(cinfo); /* Create the xbuffer[] lists */ + main_ptr->whichptr = 0; /* Read first iMCU row into xbuffer[0] */ + main_ptr->context_state = CTX_PREPARE_FOR_IMCU; + main_ptr->iMCU_row_ctr = 0; + } else { + /* Simple case with no context needed */ + main_ptr->pub._process_data = process_data_simple_main; + } + main_ptr->buffer_full = FALSE; /* Mark buffer empty */ + main_ptr->rowgroup_ctr = 0; + break; +#ifdef QUANT_2PASS_SUPPORTED + case JBUF_CRANK_DEST: + /* For last pass of 2-pass quantization, just crank the postprocessor */ + main_ptr->pub._process_data = process_data_crank_post; + break; +#endif + default: + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + break; + } +} + + +/* + * Process some data. + * This handles the simple case where no context is required. + */ + +METHODDEF(void) +process_data_simple_main(j_decompress_ptr cinfo, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + JDIMENSION rowgroups_avail; + + /* Read input data if we haven't filled the main buffer yet */ + if (!main_ptr->buffer_full) { + if (!(*cinfo->coef->_decompress_data) (cinfo, main_ptr->buffer)) + return; /* suspension forced, can do nothing more */ + main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */ + } + + /* There are always min_DCT_scaled_size row groups in an iMCU row. */ + rowgroups_avail = (JDIMENSION)cinfo->_min_DCT_scaled_size; + /* Note: at the bottom of the image, we may pass extra garbage row groups + * to the postprocessor. The postprocessor has to check for bottom + * of image anyway (at row resolution), so no point in us doing it too. + */ + + /* Feed the postprocessor */ + (*cinfo->post->_post_process_data) (cinfo, main_ptr->buffer, + &main_ptr->rowgroup_ctr, rowgroups_avail, + output_buf, out_row_ctr, out_rows_avail); + + /* Has postprocessor consumed all the data yet? If so, mark buffer empty */ + if (main_ptr->rowgroup_ctr >= rowgroups_avail) { + main_ptr->buffer_full = FALSE; + main_ptr->rowgroup_ctr = 0; + } +} + + +/* + * Process some data. + * This handles the case where context rows must be provided. + */ + +METHODDEF(void) +process_data_context_main(j_decompress_ptr cinfo, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + + /* Read input data if we haven't filled the main buffer yet */ + if (!main_ptr->buffer_full) { + if (!(*cinfo->coef->_decompress_data) (cinfo, + main_ptr->xbuffer[main_ptr->whichptr])) + return; /* suspension forced, can do nothing more */ + main_ptr->buffer_full = TRUE; /* OK, we have an iMCU row to work with */ + main_ptr->iMCU_row_ctr++; /* count rows received */ + } + + /* Postprocessor typically will not swallow all the input data it is handed + * in one call (due to filling the output buffer first). Must be prepared + * to exit and restart. This switch lets us keep track of how far we got. + * Note that each case falls through to the next on successful completion. + */ + switch (main_ptr->context_state) { + case CTX_POSTPONED_ROW: + /* Call postprocessor using previously set pointers for postponed row */ + (*cinfo->post->_post_process_data) (cinfo, + main_ptr->xbuffer[main_ptr->whichptr], + &main_ptr->rowgroup_ctr, + main_ptr->rowgroups_avail, output_buf, + out_row_ctr, out_rows_avail); + if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail) + return; /* Need to suspend */ + main_ptr->context_state = CTX_PREPARE_FOR_IMCU; + if (*out_row_ctr >= out_rows_avail) + return; /* Postprocessor exactly filled output buf */ + FALLTHROUGH /*FALLTHROUGH*/ + case CTX_PREPARE_FOR_IMCU: + /* Prepare to process first M-1 row groups of this iMCU row */ + main_ptr->rowgroup_ctr = 0; + main_ptr->rowgroups_avail = (JDIMENSION)(cinfo->_min_DCT_scaled_size - 1); + /* Check for bottom of image: if so, tweak pointers to "duplicate" + * the last sample row, and adjust rowgroups_avail to ignore padding rows. + */ + if (main_ptr->iMCU_row_ctr == cinfo->total_iMCU_rows) + set_bottom_pointers(cinfo); + main_ptr->context_state = CTX_PROCESS_IMCU; + FALLTHROUGH /*FALLTHROUGH*/ + case CTX_PROCESS_IMCU: + /* Call postprocessor using previously set pointers */ + (*cinfo->post->_post_process_data) (cinfo, + main_ptr->xbuffer[main_ptr->whichptr], + &main_ptr->rowgroup_ctr, + main_ptr->rowgroups_avail, output_buf, + out_row_ctr, out_rows_avail); + if (main_ptr->rowgroup_ctr < main_ptr->rowgroups_avail) + return; /* Need to suspend */ + /* After the first iMCU, change wraparound pointers to normal state */ + if (main_ptr->iMCU_row_ctr == 1) + set_wraparound_pointers(cinfo); + /* Prepare to load new iMCU row using other xbuffer list */ + main_ptr->whichptr ^= 1; /* 0=>1 or 1=>0 */ + main_ptr->buffer_full = FALSE; + /* Still need to process last row group of this iMCU row, */ + /* which is saved at index M+1 of the other xbuffer */ + main_ptr->rowgroup_ctr = (JDIMENSION)(cinfo->_min_DCT_scaled_size + 1); + main_ptr->rowgroups_avail = (JDIMENSION)(cinfo->_min_DCT_scaled_size + 2); + main_ptr->context_state = CTX_POSTPONED_ROW; + } +} + + +/* + * Process some data. + * Final pass of two-pass quantization: just call the postprocessor. + * Source data will be the postprocessor controller's internal buffer. + */ + +#ifdef QUANT_2PASS_SUPPORTED + +METHODDEF(void) +process_data_crank_post(j_decompress_ptr cinfo, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +{ + (*cinfo->post->_post_process_data) (cinfo, (_JSAMPIMAGE)NULL, + (JDIMENSION *)NULL, (JDIMENSION)0, + output_buf, out_row_ctr, out_rows_avail); +} + +#endif /* QUANT_2PASS_SUPPORTED */ + + +/* + * Initialize main buffer controller. + */ + +GLOBAL(void) +_jinit_d_main_controller(j_decompress_ptr cinfo, boolean need_full_buffer) +{ + my_main_ptr main_ptr; + int ci, rgroup, ngroups; + jpeg_component_info *compptr; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + main_ptr = (my_main_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_main_controller)); + cinfo->main = (struct jpeg_d_main_controller *)main_ptr; + main_ptr->pub.start_pass = start_pass_main; + + if (need_full_buffer) /* shouldn't happen */ + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + + /* Allocate the workspace. + * ngroups is the number of row groups we need. + */ + if (cinfo->upsample->need_context_rows) { + if (cinfo->_min_DCT_scaled_size < 2) /* unsupported, see comments above */ + ERREXIT(cinfo, JERR_NOTIMPL); + alloc_funny_pointers(cinfo); /* Alloc space for xbuffer[] lists */ + ngroups = cinfo->_min_DCT_scaled_size + 2; + } else { + ngroups = cinfo->_min_DCT_scaled_size; + } + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / + cinfo->_min_DCT_scaled_size; /* height of a row group of component */ + main_ptr->buffer[ci] = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + compptr->width_in_blocks * compptr->_DCT_scaled_size, + (JDIMENSION)(rgroup * ngroups)); + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jdmainct.h b/src/spice2x/external/libjpeg-turbo/jdmainct.h new file mode 100644 index 0000000..914ad11 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmainct.h @@ -0,0 +1,78 @@ +/* + * jdmainct.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + */ + +#define JPEG_INTERNALS +#include "jpeglib.h" +#include "jpegapicomp.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_d_main_controller pub; /* public fields */ + + /* Pointer to allocated workspace (M or M+2 row groups). */ + _JSAMPARRAY buffer[MAX_COMPONENTS]; + + boolean buffer_full; /* Have we gotten an iMCU row from decoder? */ + JDIMENSION rowgroup_ctr; /* counts row groups output to postprocessor */ + + /* Remaining fields are only used in the context case. */ + + /* These are the master pointers to the funny-order pointer lists. */ + _JSAMPIMAGE xbuffer[2]; /* pointers to weird pointer lists */ + + int whichptr; /* indicates which pointer set is now in use */ + int context_state; /* process_data state machine status */ + JDIMENSION rowgroups_avail; /* row groups available to postprocessor */ + JDIMENSION iMCU_row_ctr; /* counts iMCU rows to detect image top/bot */ +} my_main_controller; + +typedef my_main_controller *my_main_ptr; + + +/* context_state values: */ +#define CTX_PREPARE_FOR_IMCU 0 /* need to prepare for MCU row */ +#define CTX_PROCESS_IMCU 1 /* feeding iMCU to postprocessor */ +#define CTX_POSTPONED_ROW 2 /* feeding postponed row group */ + + +LOCAL(void) +set_wraparound_pointers(j_decompress_ptr cinfo) +/* Set up the "wraparound" pointers at top and bottom of the pointer lists. + * This changes the pointer list state from top-of-image to the normal state. + */ +{ + my_main_ptr main_ptr = (my_main_ptr)cinfo->main; + int ci, i, rgroup; + int M = cinfo->_min_DCT_scaled_size; + jpeg_component_info *compptr; + _JSAMPARRAY xbuf0, xbuf1; + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + rgroup = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / + cinfo->_min_DCT_scaled_size; /* height of a row group of component */ + xbuf0 = main_ptr->xbuffer[0][ci]; + xbuf1 = main_ptr->xbuffer[1][ci]; + for (i = 0; i < rgroup; i++) { + xbuf0[i - rgroup] = xbuf0[rgroup * (M + 1) + i]; + xbuf1[i - rgroup] = xbuf1[rgroup * (M + 1) + i]; + xbuf0[rgroup * (M + 2) + i] = xbuf0[i]; + xbuf1[rgroup * (M + 2) + i] = xbuf1[i]; + } + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jdmarker.c b/src/spice2x/external/libjpeg-turbo/jdmarker.c new file mode 100644 index 0000000..f918ee4 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmarker.c @@ -0,0 +1,1384 @@ +/* + * jdmarker.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1998, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2012, 2015, 2022, 2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains routines to decode JPEG datastream markers. + * Most of the complexity arises from our desire to support input + * suspension: if not all of the data for a marker is available, + * we must exit back to the application. On resumption, we reprocess + * the marker. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" + + +typedef enum { /* JPEG marker codes */ + M_SOF0 = 0xc0, + M_SOF1 = 0xc1, + M_SOF2 = 0xc2, + M_SOF3 = 0xc3, + + M_SOF5 = 0xc5, + M_SOF6 = 0xc6, + M_SOF7 = 0xc7, + + M_JPG = 0xc8, + M_SOF9 = 0xc9, + M_SOF10 = 0xca, + M_SOF11 = 0xcb, + + M_SOF13 = 0xcd, + M_SOF14 = 0xce, + M_SOF15 = 0xcf, + + M_DHT = 0xc4, + + M_DAC = 0xcc, + + M_RST0 = 0xd0, + M_RST1 = 0xd1, + M_RST2 = 0xd2, + M_RST3 = 0xd3, + M_RST4 = 0xd4, + M_RST5 = 0xd5, + M_RST6 = 0xd6, + M_RST7 = 0xd7, + + M_SOI = 0xd8, + M_EOI = 0xd9, + M_SOS = 0xda, + M_DQT = 0xdb, + M_DNL = 0xdc, + M_DRI = 0xdd, + M_DHP = 0xde, + M_EXP = 0xdf, + + M_APP0 = 0xe0, + M_APP1 = 0xe1, + M_APP2 = 0xe2, + M_APP3 = 0xe3, + M_APP4 = 0xe4, + M_APP5 = 0xe5, + M_APP6 = 0xe6, + M_APP7 = 0xe7, + M_APP8 = 0xe8, + M_APP9 = 0xe9, + M_APP10 = 0xea, + M_APP11 = 0xeb, + M_APP12 = 0xec, + M_APP13 = 0xed, + M_APP14 = 0xee, + M_APP15 = 0xef, + + M_JPG0 = 0xf0, + M_JPG13 = 0xfd, + M_COM = 0xfe, + + M_TEM = 0x01, + + M_ERROR = 0x100 +} JPEG_MARKER; + + +/* Private state */ + +typedef struct { + struct jpeg_marker_reader pub; /* public fields */ + + /* Application-overridable marker processing methods */ + jpeg_marker_parser_method process_COM; + jpeg_marker_parser_method process_APPn[16]; + + /* Limit on marker data length to save for each marker type */ + unsigned int length_limit_COM; + unsigned int length_limit_APPn[16]; + + /* Status of COM/APPn marker saving */ + jpeg_saved_marker_ptr cur_marker; /* NULL if not processing a marker */ + unsigned int bytes_read; /* data bytes read so far in marker */ + /* Note: cur_marker is not linked into marker_list until it's all read. */ +} my_marker_reader; + +typedef my_marker_reader *my_marker_ptr; + + +/* + * Macros for fetching data from the data source module. + * + * At all times, cinfo->src->next_input_byte and ->bytes_in_buffer reflect + * the current restart point; we update them only when we have reached a + * suitable place to restart if a suspension occurs. + */ + +/* Declare and initialize local copies of input pointer/count */ +#define INPUT_VARS(cinfo) \ + struct jpeg_source_mgr *datasrc = (cinfo)->src; \ + const JOCTET *next_input_byte = datasrc->next_input_byte; \ + size_t bytes_in_buffer = datasrc->bytes_in_buffer + +/* Unload the local copies --- do this only at a restart boundary */ +#define INPUT_SYNC(cinfo) \ + ( datasrc->next_input_byte = next_input_byte, \ + datasrc->bytes_in_buffer = bytes_in_buffer ) + +/* Reload the local copies --- used only in MAKE_BYTE_AVAIL */ +#define INPUT_RELOAD(cinfo) \ + ( next_input_byte = datasrc->next_input_byte, \ + bytes_in_buffer = datasrc->bytes_in_buffer ) + +/* Internal macro for INPUT_BYTE and INPUT_2BYTES: make a byte available. + * Note we do *not* do INPUT_SYNC before calling fill_input_buffer, + * but we must reload the local copies after a successful fill. + */ +#define MAKE_BYTE_AVAIL(cinfo, action) \ + if (bytes_in_buffer == 0) { \ + if (!(*datasrc->fill_input_buffer) (cinfo)) \ + { action; } \ + INPUT_RELOAD(cinfo); \ + } + +/* Read a byte into variable V. + * If must suspend, take the specified action (typically "return FALSE"). + */ +#define INPUT_BYTE(cinfo, V, action) \ + MAKESTMT( MAKE_BYTE_AVAIL(cinfo, action); \ + bytes_in_buffer--; \ + V = *next_input_byte++; ) + +/* As above, but read two bytes interpreted as an unsigned 16-bit integer. + * V should be declared unsigned int or perhaps JLONG. + */ +#define INPUT_2BYTES(cinfo, V, action) \ + MAKESTMT( MAKE_BYTE_AVAIL(cinfo, action); \ + bytes_in_buffer--; \ + V = ((unsigned int)(*next_input_byte++)) << 8; \ + MAKE_BYTE_AVAIL(cinfo, action); \ + bytes_in_buffer--; \ + V += *next_input_byte++; ) + + +/* + * Routines to process JPEG markers. + * + * Entry condition: JPEG marker itself has been read and its code saved + * in cinfo->unread_marker; input restart point is just after the marker. + * + * Exit: if return TRUE, have read and processed any parameters, and have + * updated the restart point to point after the parameters. + * If return FALSE, was forced to suspend before reaching end of + * marker parameters; restart point has not been moved. Same routine + * will be called again after application supplies more input data. + * + * This approach to suspension assumes that all of a marker's parameters + * can fit into a single input bufferload. This should hold for "normal" + * markers. Some COM/APPn markers might have large parameter segments + * that might not fit. If we are simply dropping such a marker, we use + * skip_input_data to get past it, and thereby put the problem on the + * source manager's shoulders. If we are saving the marker's contents + * into memory, we use a slightly different convention: when forced to + * suspend, the marker processor updates the restart point to the end of + * what it's consumed (ie, the end of the buffer) before returning FALSE. + * On resumption, cinfo->unread_marker still contains the marker code, + * but the data source will point to the next chunk of marker data. + * The marker processor must retain internal state to deal with this. + * + * Note that we don't bother to avoid duplicate trace messages if a + * suspension occurs within marker parameters. Other side effects + * require more care. + */ + + +LOCAL(boolean) +get_soi(j_decompress_ptr cinfo) +/* Process an SOI marker */ +{ + int i; + + TRACEMS(cinfo, 1, JTRC_SOI); + + if (cinfo->marker->saw_SOI) + ERREXIT(cinfo, JERR_SOI_DUPLICATE); + + /* Reset all parameters that are defined to be reset by SOI */ + + for (i = 0; i < NUM_ARITH_TBLS; i++) { + cinfo->arith_dc_L[i] = 0; + cinfo->arith_dc_U[i] = 1; + cinfo->arith_ac_K[i] = 5; + } + cinfo->restart_interval = 0; + + /* Set initial assumptions for colorspace etc */ + + cinfo->jpeg_color_space = JCS_UNKNOWN; + cinfo->CCIR601_sampling = FALSE; /* Assume non-CCIR sampling??? */ + + cinfo->saw_JFIF_marker = FALSE; + cinfo->JFIF_major_version = 1; /* set default JFIF APP0 values */ + cinfo->JFIF_minor_version = 1; + cinfo->density_unit = 0; + cinfo->X_density = 1; + cinfo->Y_density = 1; + cinfo->saw_Adobe_marker = FALSE; + cinfo->Adobe_transform = 0; + + cinfo->marker->saw_SOI = TRUE; + + return TRUE; +} + + +LOCAL(boolean) +get_sof(j_decompress_ptr cinfo, boolean is_prog, boolean is_lossless, + boolean is_arith) +/* Process a SOFn marker */ +{ + JLONG length; + int c, ci; + jpeg_component_info *compptr; + INPUT_VARS(cinfo); + + if (cinfo->marker->saw_SOF) + ERREXIT(cinfo, JERR_SOF_DUPLICATE); + + cinfo->progressive_mode = is_prog; + cinfo->master->lossless = is_lossless; + cinfo->arith_code = is_arith; + + INPUT_2BYTES(cinfo, length, return FALSE); + + INPUT_BYTE(cinfo, cinfo->data_precision, return FALSE); + INPUT_2BYTES(cinfo, cinfo->image_height, return FALSE); + INPUT_2BYTES(cinfo, cinfo->image_width, return FALSE); + INPUT_BYTE(cinfo, cinfo->num_components, return FALSE); + + length -= 8; + + TRACEMS4(cinfo, 1, JTRC_SOF, cinfo->unread_marker, + (int)cinfo->image_width, (int)cinfo->image_height, + cinfo->num_components); + + /* We don't support files in which the image height is initially specified */ + /* as 0 and is later redefined by DNL. As long as we have to check that, */ + /* might as well have a general sanity check. */ + if (cinfo->image_height <= 0 || cinfo->image_width <= 0 || + cinfo->num_components <= 0) + ERREXIT(cinfo, JERR_EMPTY_IMAGE); + + if (length != (cinfo->num_components * 3)) + ERREXIT(cinfo, JERR_BAD_LENGTH); + + if (cinfo->comp_info == NULL) /* do only once, even if suspend */ + cinfo->comp_info = (jpeg_component_info *)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + cinfo->num_components * sizeof(jpeg_component_info)); + + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + compptr->component_index = ci; + INPUT_BYTE(cinfo, compptr->component_id, return FALSE); + INPUT_BYTE(cinfo, c, return FALSE); + compptr->h_samp_factor = (c >> 4) & 15; + compptr->v_samp_factor = (c ) & 15; + INPUT_BYTE(cinfo, compptr->quant_tbl_no, return FALSE); + + TRACEMS4(cinfo, 1, JTRC_SOF_COMPONENT, + compptr->component_id, compptr->h_samp_factor, + compptr->v_samp_factor, compptr->quant_tbl_no); + } + + cinfo->marker->saw_SOF = TRUE; + + INPUT_SYNC(cinfo); + return TRUE; +} + + +LOCAL(boolean) +get_sos(j_decompress_ptr cinfo) +/* Process a SOS marker */ +{ + JLONG length; + int i, ci, n, c, cc, pi; + jpeg_component_info *compptr; + INPUT_VARS(cinfo); + + if (!cinfo->marker->saw_SOF) + ERREXIT(cinfo, JERR_SOS_NO_SOF); + + INPUT_2BYTES(cinfo, length, return FALSE); + + INPUT_BYTE(cinfo, n, return FALSE); /* Number of components */ + + TRACEMS1(cinfo, 1, JTRC_SOS, n); + + if (length != (n * 2 + 6) || n < 1 || n > MAX_COMPS_IN_SCAN) + ERREXIT(cinfo, JERR_BAD_LENGTH); + + cinfo->comps_in_scan = n; + + /* Collect the component-spec parameters */ + + for (i = 0; i < MAX_COMPS_IN_SCAN; i++) + cinfo->cur_comp_info[i] = NULL; + + for (i = 0; i < n; i++) { + INPUT_BYTE(cinfo, cc, return FALSE); + INPUT_BYTE(cinfo, c, return FALSE); + + for (ci = 0, compptr = cinfo->comp_info; + ci < cinfo->num_components && ci < MAX_COMPS_IN_SCAN; + ci++, compptr++) { + if (cc == compptr->component_id && !cinfo->cur_comp_info[ci]) + goto id_found; + } + + ERREXIT1(cinfo, JERR_BAD_COMPONENT_ID, cc); + +id_found: + + cinfo->cur_comp_info[i] = compptr; + compptr->dc_tbl_no = (c >> 4) & 15; + compptr->ac_tbl_no = (c ) & 15; + + TRACEMS3(cinfo, 1, JTRC_SOS_COMPONENT, cc, + compptr->dc_tbl_no, compptr->ac_tbl_no); + + /* This CSi (cc) should differ from the previous CSi */ + for (pi = 0; pi < i; pi++) { + if (cinfo->cur_comp_info[pi] == compptr) { + ERREXIT1(cinfo, JERR_BAD_COMPONENT_ID, cc); + } + } + } + + /* Collect the additional scan parameters Ss, Se, Ah/Al. */ + INPUT_BYTE(cinfo, c, return FALSE); + cinfo->Ss = c; + INPUT_BYTE(cinfo, c, return FALSE); + cinfo->Se = c; + INPUT_BYTE(cinfo, c, return FALSE); + cinfo->Ah = (c >> 4) & 15; + cinfo->Al = (c ) & 15; + + TRACEMS4(cinfo, 1, JTRC_SOS_PARAMS, cinfo->Ss, cinfo->Se, + cinfo->Ah, cinfo->Al); + + /* Prepare to scan data & restart markers */ + cinfo->marker->next_restart_num = 0; + + /* Count another SOS marker */ + cinfo->input_scan_number++; + + INPUT_SYNC(cinfo); + return TRUE; +} + + +#ifdef D_ARITH_CODING_SUPPORTED + +LOCAL(boolean) +get_dac(j_decompress_ptr cinfo) +/* Process a DAC marker */ +{ + JLONG length; + int index, val; + INPUT_VARS(cinfo); + + INPUT_2BYTES(cinfo, length, return FALSE); + length -= 2; + + while (length > 0) { + INPUT_BYTE(cinfo, index, return FALSE); + INPUT_BYTE(cinfo, val, return FALSE); + + length -= 2; + + TRACEMS2(cinfo, 1, JTRC_DAC, index, val); + + if (index < 0 || index >= (2 * NUM_ARITH_TBLS)) + ERREXIT1(cinfo, JERR_DAC_INDEX, index); + + if (index >= NUM_ARITH_TBLS) { /* define AC table */ + cinfo->arith_ac_K[index - NUM_ARITH_TBLS] = (UINT8)val; + } else { /* define DC table */ + cinfo->arith_dc_L[index] = (UINT8)(val & 0x0F); + cinfo->arith_dc_U[index] = (UINT8)(val >> 4); + if (cinfo->arith_dc_L[index] > cinfo->arith_dc_U[index]) + ERREXIT1(cinfo, JERR_DAC_VALUE, val); + } + } + + if (length != 0) + ERREXIT(cinfo, JERR_BAD_LENGTH); + + INPUT_SYNC(cinfo); + return TRUE; +} + +#else /* !D_ARITH_CODING_SUPPORTED */ + +#define get_dac(cinfo) skip_variable(cinfo) + +#endif /* D_ARITH_CODING_SUPPORTED */ + + +LOCAL(boolean) +get_dht(j_decompress_ptr cinfo) +/* Process a DHT marker */ +{ + JLONG length; + UINT8 bits[17]; + UINT8 huffval[256]; + int i, index, count; + JHUFF_TBL **htblptr; + INPUT_VARS(cinfo); + + INPUT_2BYTES(cinfo, length, return FALSE); + length -= 2; + + while (length > 16) { + INPUT_BYTE(cinfo, index, return FALSE); + + TRACEMS1(cinfo, 1, JTRC_DHT, index); + + bits[0] = 0; + count = 0; + for (i = 1; i <= 16; i++) { + INPUT_BYTE(cinfo, bits[i], return FALSE); + count += bits[i]; + } + + length -= 1 + 16; + + TRACEMS8(cinfo, 2, JTRC_HUFFBITS, + bits[1], bits[2], bits[3], bits[4], + bits[5], bits[6], bits[7], bits[8]); + TRACEMS8(cinfo, 2, JTRC_HUFFBITS, + bits[9], bits[10], bits[11], bits[12], + bits[13], bits[14], bits[15], bits[16]); + + /* Here we just do minimal validation of the counts to avoid walking + * off the end of our table space. jdhuff.c will check more carefully. + */ + if (count > 256 || ((JLONG)count) > length) + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + + for (i = 0; i < count; i++) + INPUT_BYTE(cinfo, huffval[i], return FALSE); + + memset(&huffval[count], 0, (256 - count) * sizeof(UINT8)); + + length -= count; + + if (index & 0x10) { /* AC table definition */ + index -= 0x10; + if (index < 0 || index >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_DHT_INDEX, index); + htblptr = &cinfo->ac_huff_tbl_ptrs[index]; + } else { /* DC table definition */ + if (index < 0 || index >= NUM_HUFF_TBLS) + ERREXIT1(cinfo, JERR_DHT_INDEX, index); + htblptr = &cinfo->dc_huff_tbl_ptrs[index]; + } + + if (*htblptr == NULL) + *htblptr = jpeg_alloc_huff_table((j_common_ptr)cinfo); + + memcpy((*htblptr)->bits, bits, sizeof((*htblptr)->bits)); + memcpy((*htblptr)->huffval, huffval, sizeof((*htblptr)->huffval)); + } + + if (length != 0) + ERREXIT(cinfo, JERR_BAD_LENGTH); + + INPUT_SYNC(cinfo); + return TRUE; +} + + +LOCAL(boolean) +get_dqt(j_decompress_ptr cinfo) +/* Process a DQT marker */ +{ + JLONG length; + int n, i, prec; + unsigned int tmp; + JQUANT_TBL *quant_ptr; + INPUT_VARS(cinfo); + + INPUT_2BYTES(cinfo, length, return FALSE); + length -= 2; + + while (length > 0) { + INPUT_BYTE(cinfo, n, return FALSE); + prec = n >> 4; + n &= 0x0F; + + TRACEMS2(cinfo, 1, JTRC_DQT, n, prec); + + if (n >= NUM_QUANT_TBLS) + ERREXIT1(cinfo, JERR_DQT_INDEX, n); + + if (cinfo->quant_tbl_ptrs[n] == NULL) + cinfo->quant_tbl_ptrs[n] = jpeg_alloc_quant_table((j_common_ptr)cinfo); + quant_ptr = cinfo->quant_tbl_ptrs[n]; + + for (i = 0; i < DCTSIZE2; i++) { + if (prec) + INPUT_2BYTES(cinfo, tmp, return FALSE); + else + INPUT_BYTE(cinfo, tmp, return FALSE); + /* We convert the zigzag-order table to natural array order. */ + quant_ptr->quantval[jpeg_natural_order[i]] = (UINT16)tmp; + } + + if (cinfo->err->trace_level >= 2) { + for (i = 0; i < DCTSIZE2; i += 8) { + TRACEMS8(cinfo, 2, JTRC_QUANTVALS, + quant_ptr->quantval[i], quant_ptr->quantval[i + 1], + quant_ptr->quantval[i + 2], quant_ptr->quantval[i + 3], + quant_ptr->quantval[i + 4], quant_ptr->quantval[i + 5], + quant_ptr->quantval[i + 6], quant_ptr->quantval[i + 7]); + } + } + + length -= DCTSIZE2 + 1; + if (prec) length -= DCTSIZE2; + } + + if (length != 0) + ERREXIT(cinfo, JERR_BAD_LENGTH); + + INPUT_SYNC(cinfo); + return TRUE; +} + + +LOCAL(boolean) +get_dri(j_decompress_ptr cinfo) +/* Process a DRI marker */ +{ + JLONG length; + unsigned int tmp; + INPUT_VARS(cinfo); + + INPUT_2BYTES(cinfo, length, return FALSE); + + if (length != 4) + ERREXIT(cinfo, JERR_BAD_LENGTH); + + INPUT_2BYTES(cinfo, tmp, return FALSE); + + TRACEMS1(cinfo, 1, JTRC_DRI, tmp); + + cinfo->restart_interval = tmp; + + INPUT_SYNC(cinfo); + return TRUE; +} + + +/* + * Routines for processing APPn and COM markers. + * These are either saved in memory or discarded, per application request. + * APP0 and APP14 are specially checked to see if they are + * JFIF and Adobe markers, respectively. + */ + +#define APP0_DATA_LEN 14 /* Length of interesting data in APP0 */ +#define APP14_DATA_LEN 12 /* Length of interesting data in APP14 */ +#define APPN_DATA_LEN 14 /* Must be the largest of the above!! */ + + +LOCAL(void) +examine_app0(j_decompress_ptr cinfo, JOCTET *data, unsigned int datalen, + JLONG remaining) +/* Examine first few bytes from an APP0. + * Take appropriate action if it is a JFIF marker. + * datalen is # of bytes at data[], remaining is length of rest of marker data. + */ +{ + JLONG totallen = (JLONG)datalen + remaining; + + if (datalen >= APP0_DATA_LEN && + data[0] == 0x4A && + data[1] == 0x46 && + data[2] == 0x49 && + data[3] == 0x46 && + data[4] == 0) { + /* Found JFIF APP0 marker: save info */ + cinfo->saw_JFIF_marker = TRUE; + cinfo->JFIF_major_version = data[5]; + cinfo->JFIF_minor_version = data[6]; + cinfo->density_unit = data[7]; + cinfo->X_density = (data[8] << 8) + data[9]; + cinfo->Y_density = (data[10] << 8) + data[11]; + /* Check version. + * Major version must be 1, anything else signals an incompatible change. + * (We used to treat this as an error, but now it's a nonfatal warning, + * because some bozo at Hijaak couldn't read the spec.) + * Minor version should be 0..2, but process anyway if newer. + */ + if (cinfo->JFIF_major_version != 1) + WARNMS2(cinfo, JWRN_JFIF_MAJOR, + cinfo->JFIF_major_version, cinfo->JFIF_minor_version); + /* Generate trace messages */ + TRACEMS5(cinfo, 1, JTRC_JFIF, + cinfo->JFIF_major_version, cinfo->JFIF_minor_version, + cinfo->X_density, cinfo->Y_density, cinfo->density_unit); + /* Validate thumbnail dimensions and issue appropriate messages */ + if (data[12] | data[13]) + TRACEMS2(cinfo, 1, JTRC_JFIF_THUMBNAIL, data[12], data[13]); + totallen -= APP0_DATA_LEN; + if (totallen != ((JLONG)data[12] * (JLONG)data[13] * (JLONG)3)) + TRACEMS1(cinfo, 1, JTRC_JFIF_BADTHUMBNAILSIZE, (int)totallen); + } else if (datalen >= 6 && + data[0] == 0x4A && + data[1] == 0x46 && + data[2] == 0x58 && + data[3] == 0x58 && + data[4] == 0) { + /* Found JFIF "JFXX" extension APP0 marker */ + /* The library doesn't actually do anything with these, + * but we try to produce a helpful trace message. + */ + switch (data[5]) { + case 0x10: + TRACEMS1(cinfo, 1, JTRC_THUMB_JPEG, (int)totallen); + break; + case 0x11: + TRACEMS1(cinfo, 1, JTRC_THUMB_PALETTE, (int)totallen); + break; + case 0x13: + TRACEMS1(cinfo, 1, JTRC_THUMB_RGB, (int)totallen); + break; + default: + TRACEMS2(cinfo, 1, JTRC_JFIF_EXTENSION, data[5], (int)totallen); + break; + } + } else { + /* Start of APP0 does not match "JFIF" or "JFXX", or too short */ + TRACEMS1(cinfo, 1, JTRC_APP0, (int)totallen); + } +} + + +LOCAL(void) +examine_app14(j_decompress_ptr cinfo, JOCTET *data, unsigned int datalen, + JLONG remaining) +/* Examine first few bytes from an APP14. + * Take appropriate action if it is an Adobe marker. + * datalen is # of bytes at data[], remaining is length of rest of marker data. + */ +{ + unsigned int version, flags0, flags1, transform; + + if (datalen >= APP14_DATA_LEN && + data[0] == 0x41 && + data[1] == 0x64 && + data[2] == 0x6F && + data[3] == 0x62 && + data[4] == 0x65) { + /* Found Adobe APP14 marker */ + version = (data[5] << 8) + data[6]; + flags0 = (data[7] << 8) + data[8]; + flags1 = (data[9] << 8) + data[10]; + transform = data[11]; + TRACEMS4(cinfo, 1, JTRC_ADOBE, version, flags0, flags1, transform); + cinfo->saw_Adobe_marker = TRUE; + cinfo->Adobe_transform = (UINT8)transform; + } else { + /* Start of APP14 does not match "Adobe", or too short */ + TRACEMS1(cinfo, 1, JTRC_APP14, (int)(datalen + remaining)); + } +} + + +METHODDEF(boolean) +get_interesting_appn(j_decompress_ptr cinfo) +/* Process an APP0 or APP14 marker without saving it */ +{ + JLONG length; + JOCTET b[APPN_DATA_LEN]; + unsigned int i, numtoread; + INPUT_VARS(cinfo); + + INPUT_2BYTES(cinfo, length, return FALSE); + length -= 2; + + /* get the interesting part of the marker data */ + if (length >= APPN_DATA_LEN) + numtoread = APPN_DATA_LEN; + else if (length > 0) + numtoread = (unsigned int)length; + else + numtoread = 0; + for (i = 0; i < numtoread; i++) + INPUT_BYTE(cinfo, b[i], return FALSE); + length -= numtoread; + + /* process it */ + switch (cinfo->unread_marker) { + case M_APP0: + examine_app0(cinfo, (JOCTET *)b, numtoread, length); + break; + case M_APP14: + examine_app14(cinfo, (JOCTET *)b, numtoread, length); + break; + default: + /* can't get here unless jpeg_save_markers chooses wrong processor */ + ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, cinfo->unread_marker); + break; + } + + /* skip any remaining data -- could be lots */ + INPUT_SYNC(cinfo); + if (length > 0) + (*cinfo->src->skip_input_data) (cinfo, (long)length); + + return TRUE; +} + + +#ifdef SAVE_MARKERS_SUPPORTED + +METHODDEF(boolean) +save_marker(j_decompress_ptr cinfo) +/* Save an APPn or COM marker into the marker list */ +{ + my_marker_ptr marker = (my_marker_ptr)cinfo->marker; + jpeg_saved_marker_ptr cur_marker = marker->cur_marker; + unsigned int bytes_read, data_length; + JOCTET *data; + JLONG length = 0; + INPUT_VARS(cinfo); + + if (cur_marker == NULL) { + /* begin reading a marker */ + INPUT_2BYTES(cinfo, length, return FALSE); + length -= 2; + if (length >= 0) { /* watch out for bogus length word */ + /* figure out how much we want to save */ + unsigned int limit; + if (cinfo->unread_marker == (int)M_COM) + limit = marker->length_limit_COM; + else + limit = marker->length_limit_APPn[cinfo->unread_marker - (int)M_APP0]; + if ((unsigned int)length < limit) + limit = (unsigned int)length; + /* allocate and initialize the marker item */ + cur_marker = (jpeg_saved_marker_ptr) + (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(struct jpeg_marker_struct) + limit); + cur_marker->next = NULL; + cur_marker->marker = (UINT8)cinfo->unread_marker; + cur_marker->original_length = (unsigned int)length; + cur_marker->data_length = limit; + /* data area is just beyond the jpeg_marker_struct */ + data = cur_marker->data = (JOCTET *)(cur_marker + 1); + marker->cur_marker = cur_marker; + marker->bytes_read = 0; + bytes_read = 0; + data_length = limit; + } else { + /* deal with bogus length word */ + bytes_read = data_length = 0; + data = NULL; + } + } else { + /* resume reading a marker */ + bytes_read = marker->bytes_read; + data_length = cur_marker->data_length; + data = cur_marker->data + bytes_read; + } + + while (bytes_read < data_length) { + INPUT_SYNC(cinfo); /* move the restart point to here */ + marker->bytes_read = bytes_read; + /* If there's not at least one byte in buffer, suspend */ + MAKE_BYTE_AVAIL(cinfo, return FALSE); + /* Copy bytes with reasonable rapidity */ + while (bytes_read < data_length && bytes_in_buffer > 0) { + *data++ = *next_input_byte++; + bytes_in_buffer--; + bytes_read++; + } + } + + /* Done reading what we want to read */ + if (cur_marker != NULL) { /* will be NULL if bogus length word */ + /* Add new marker to end of list */ + if (cinfo->marker_list == NULL || cinfo->master->marker_list_end == NULL) { + cinfo->marker_list = cinfo->master->marker_list_end = cur_marker; + } else { + cinfo->master->marker_list_end->next = cur_marker; + cinfo->master->marker_list_end = cur_marker; + } + /* Reset pointer & calc remaining data length */ + data = cur_marker->data; + length = cur_marker->original_length - data_length; + } + /* Reset to initial state for next marker */ + marker->cur_marker = NULL; + + /* Process the marker if interesting; else just make a generic trace msg */ + switch (cinfo->unread_marker) { + case M_APP0: + examine_app0(cinfo, data, data_length, length); + break; + case M_APP14: + examine_app14(cinfo, data, data_length, length); + break; + default: + TRACEMS2(cinfo, 1, JTRC_MISC_MARKER, cinfo->unread_marker, + (int)(data_length + length)); + break; + } + + /* skip any remaining data -- could be lots */ + INPUT_SYNC(cinfo); /* do before skip_input_data */ + if (length > 0) + (*cinfo->src->skip_input_data) (cinfo, (long)length); + + return TRUE; +} + +#endif /* SAVE_MARKERS_SUPPORTED */ + + +METHODDEF(boolean) +skip_variable(j_decompress_ptr cinfo) +/* Skip over an unknown or uninteresting variable-length marker */ +{ + JLONG length; + INPUT_VARS(cinfo); + + INPUT_2BYTES(cinfo, length, return FALSE); + length -= 2; + + TRACEMS2(cinfo, 1, JTRC_MISC_MARKER, cinfo->unread_marker, (int)length); + + INPUT_SYNC(cinfo); /* do before skip_input_data */ + if (length > 0) + (*cinfo->src->skip_input_data) (cinfo, (long)length); + + return TRUE; +} + + +/* + * Find the next JPEG marker, save it in cinfo->unread_marker. + * Returns FALSE if had to suspend before reaching a marker; + * in that case cinfo->unread_marker is unchanged. + * + * Note that the result might not be a valid marker code, + * but it will never be 0 or FF. + */ + +LOCAL(boolean) +next_marker(j_decompress_ptr cinfo) +{ + int c; + INPUT_VARS(cinfo); + + for (;;) { + INPUT_BYTE(cinfo, c, return FALSE); + /* Skip any non-FF bytes. + * This may look a bit inefficient, but it will not occur in a valid file. + * We sync after each discarded byte so that a suspending data source + * can discard the byte from its buffer. + */ + while (c != 0xFF) { + cinfo->marker->discarded_bytes++; + INPUT_SYNC(cinfo); + INPUT_BYTE(cinfo, c, return FALSE); + } + /* This loop swallows any duplicate FF bytes. Extra FFs are legal as + * pad bytes, so don't count them in discarded_bytes. We assume there + * will not be so many consecutive FF bytes as to overflow a suspending + * data source's input buffer. + */ + do { + INPUT_BYTE(cinfo, c, return FALSE); + } while (c == 0xFF); + if (c != 0) + break; /* found a valid marker, exit loop */ + /* Reach here if we found a stuffed-zero data sequence (FF/00). + * Discard it and loop back to try again. + */ + cinfo->marker->discarded_bytes += 2; + INPUT_SYNC(cinfo); + } + + if (cinfo->marker->discarded_bytes != 0) { + WARNMS2(cinfo, JWRN_EXTRANEOUS_DATA, cinfo->marker->discarded_bytes, c); + cinfo->marker->discarded_bytes = 0; + } + + cinfo->unread_marker = c; + + INPUT_SYNC(cinfo); + return TRUE; +} + + +LOCAL(boolean) +first_marker(j_decompress_ptr cinfo) +/* Like next_marker, but used to obtain the initial SOI marker. */ +/* For this marker, we do not allow preceding garbage or fill; otherwise, + * we might well scan an entire input file before realizing it ain't JPEG. + * If an application wants to process non-JFIF files, it must seek to the + * SOI before calling the JPEG library. + */ +{ + int c, c2; + INPUT_VARS(cinfo); + + INPUT_BYTE(cinfo, c, return FALSE); + INPUT_BYTE(cinfo, c2, return FALSE); + if (c != 0xFF || c2 != (int)M_SOI) + ERREXIT2(cinfo, JERR_NO_SOI, c, c2); + + cinfo->unread_marker = c2; + + INPUT_SYNC(cinfo); + return TRUE; +} + + +/* + * Read markers until SOS or EOI. + * + * Returns same codes as are defined for jpeg_consume_input: + * JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI. + */ + +METHODDEF(int) +read_markers(j_decompress_ptr cinfo) +{ + /* Outer loop repeats once for each marker. */ + for (;;) { + /* Collect the marker proper, unless we already did. */ + /* NB: first_marker() enforces the requirement that SOI appear first. */ + if (cinfo->unread_marker == 0) { + if (!cinfo->marker->saw_SOI) { + if (!first_marker(cinfo)) + return JPEG_SUSPENDED; + } else { + if (!next_marker(cinfo)) + return JPEG_SUSPENDED; + } + } + /* At this point cinfo->unread_marker contains the marker code and the + * input point is just past the marker proper, but before any parameters. + * A suspension will cause us to return with this state still true. + */ + switch (cinfo->unread_marker) { + case M_SOI: + if (!get_soi(cinfo)) + return JPEG_SUSPENDED; + break; + + case M_SOF0: /* Baseline */ + case M_SOF1: /* Extended sequential, Huffman */ + if (!get_sof(cinfo, FALSE, FALSE, FALSE)) + return JPEG_SUSPENDED; + break; + + case M_SOF2: /* Progressive, Huffman */ + if (!get_sof(cinfo, TRUE, FALSE, FALSE)) + return JPEG_SUSPENDED; + break; + + case M_SOF3: /* Lossless, Huffman */ + if (!get_sof(cinfo, FALSE, TRUE, FALSE)) + return JPEG_SUSPENDED; + break; + + case M_SOF9: /* Extended sequential, arithmetic */ + if (!get_sof(cinfo, FALSE, FALSE, TRUE)) + return JPEG_SUSPENDED; + break; + + case M_SOF10: /* Progressive, arithmetic */ + if (!get_sof(cinfo, TRUE, FALSE, TRUE)) + return JPEG_SUSPENDED; + break; + + case M_SOF11: /* Lossless, arithmetic */ + if (!get_sof(cinfo, FALSE, TRUE, TRUE)) + return JPEG_SUSPENDED; + break; + + /* Currently unsupported SOFn types */ + case M_SOF5: /* Differential sequential, Huffman */ + case M_SOF6: /* Differential progressive, Huffman */ + case M_SOF7: /* Differential lossless, Huffman */ + case M_JPG: /* Reserved for JPEG extensions */ + case M_SOF13: /* Differential sequential, arithmetic */ + case M_SOF14: /* Differential progressive, arithmetic */ + case M_SOF15: /* Differential lossless, arithmetic */ + ERREXIT1(cinfo, JERR_SOF_UNSUPPORTED, cinfo->unread_marker); + break; + + case M_SOS: + if (!get_sos(cinfo)) + return JPEG_SUSPENDED; + cinfo->unread_marker = 0; /* processed the marker */ + return JPEG_REACHED_SOS; + + case M_EOI: + TRACEMS(cinfo, 1, JTRC_EOI); + cinfo->unread_marker = 0; /* processed the marker */ + return JPEG_REACHED_EOI; + + case M_DAC: + if (!get_dac(cinfo)) + return JPEG_SUSPENDED; + break; + + case M_DHT: + if (!get_dht(cinfo)) + return JPEG_SUSPENDED; + break; + + case M_DQT: + if (!get_dqt(cinfo)) + return JPEG_SUSPENDED; + break; + + case M_DRI: + if (!get_dri(cinfo)) + return JPEG_SUSPENDED; + break; + + case M_APP0: + case M_APP1: + case M_APP2: + case M_APP3: + case M_APP4: + case M_APP5: + case M_APP6: + case M_APP7: + case M_APP8: + case M_APP9: + case M_APP10: + case M_APP11: + case M_APP12: + case M_APP13: + case M_APP14: + case M_APP15: + if (!(*((my_marker_ptr)cinfo->marker)->process_APPn[ + cinfo->unread_marker - (int)M_APP0]) (cinfo)) + return JPEG_SUSPENDED; + break; + + case M_COM: + if (!(*((my_marker_ptr)cinfo->marker)->process_COM) (cinfo)) + return JPEG_SUSPENDED; + break; + + case M_RST0: /* these are all parameterless */ + case M_RST1: + case M_RST2: + case M_RST3: + case M_RST4: + case M_RST5: + case M_RST6: + case M_RST7: + case M_TEM: + TRACEMS1(cinfo, 1, JTRC_PARMLESS_MARKER, cinfo->unread_marker); + break; + + case M_DNL: /* Ignore DNL ... perhaps the wrong thing */ + if (!skip_variable(cinfo)) + return JPEG_SUSPENDED; + break; + + default: /* must be DHP, EXP, JPGn, or RESn */ + /* For now, we treat the reserved markers as fatal errors since they are + * likely to be used to signal incompatible JPEG Part 3 extensions. + * Once the JPEG 3 version-number marker is well defined, this code + * ought to change! + */ + ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, cinfo->unread_marker); + break; + } + /* Successfully processed marker, so reset state variable */ + cinfo->unread_marker = 0; + } /* end loop */ +} + + +/* + * Read a restart marker, which is expected to appear next in the datastream; + * if the marker is not there, take appropriate recovery action. + * Returns FALSE if suspension is required. + * + * This is called by the entropy decoder after it has read an appropriate + * number of MCUs. cinfo->unread_marker may be nonzero if the entropy decoder + * has already read a marker from the data source. Under normal conditions + * cinfo->unread_marker will be reset to 0 before returning; if not reset, + * it holds a marker which the decoder will be unable to read past. + */ + +METHODDEF(boolean) +read_restart_marker(j_decompress_ptr cinfo) +{ + /* Obtain a marker unless we already did. */ + /* Note that next_marker will complain if it skips any data. */ + if (cinfo->unread_marker == 0) { + if (!next_marker(cinfo)) + return FALSE; + } + + if (cinfo->unread_marker == + ((int)M_RST0 + cinfo->marker->next_restart_num)) { + /* Normal case --- swallow the marker and let entropy decoder continue */ + TRACEMS1(cinfo, 3, JTRC_RST, cinfo->marker->next_restart_num); + cinfo->unread_marker = 0; + } else { + /* Uh-oh, the restart markers have been messed up. */ + /* Let the data source manager determine how to resync. */ + if (!(*cinfo->src->resync_to_restart) (cinfo, + cinfo->marker->next_restart_num)) + return FALSE; + } + + /* Update next-restart state */ + cinfo->marker->next_restart_num = (cinfo->marker->next_restart_num + 1) & 7; + + return TRUE; +} + + +/* + * This is the default resync_to_restart method for data source managers + * to use if they don't have any better approach. Some data source managers + * may be able to back up, or may have additional knowledge about the data + * which permits a more intelligent recovery strategy; such managers would + * presumably supply their own resync method. + * + * read_restart_marker calls resync_to_restart if it finds a marker other than + * the restart marker it was expecting. (This code is *not* used unless + * a nonzero restart interval has been declared.) cinfo->unread_marker is + * the marker code actually found (might be anything, except 0 or FF). + * The desired restart marker number (0..7) is passed as a parameter. + * This routine is supposed to apply whatever error recovery strategy seems + * appropriate in order to position the input stream to the next data segment. + * Note that cinfo->unread_marker is treated as a marker appearing before + * the current data-source input point; usually it should be reset to zero + * before returning. + * Returns FALSE if suspension is required. + * + * This implementation is substantially constrained by wanting to treat the + * input as a data stream; this means we can't back up. Therefore, we have + * only the following actions to work with: + * 1. Simply discard the marker and let the entropy decoder resume at next + * byte of file. + * 2. Read forward until we find another marker, discarding intervening + * data. (In theory we could look ahead within the current bufferload, + * without having to discard data if we don't find the desired marker. + * This idea is not implemented here, in part because it makes behavior + * dependent on buffer size and chance buffer-boundary positions.) + * 3. Leave the marker unread (by failing to zero cinfo->unread_marker). + * This will cause the entropy decoder to process an empty data segment, + * inserting dummy zeroes, and then we will reprocess the marker. + * + * #2 is appropriate if we think the desired marker lies ahead, while #3 is + * appropriate if the found marker is a future restart marker (indicating + * that we have missed the desired restart marker, probably because it got + * corrupted). + * We apply #2 or #3 if the found marker is a restart marker no more than + * two counts behind or ahead of the expected one. We also apply #2 if the + * found marker is not a legal JPEG marker code (it's certainly bogus data). + * If the found marker is a restart marker more than 2 counts away, we do #1 + * (too much risk that the marker is erroneous; with luck we will be able to + * resync at some future point). + * For any valid non-restart JPEG marker, we apply #3. This keeps us from + * overrunning the end of a scan. An implementation limited to single-scan + * files might find it better to apply #2 for markers other than EOI, since + * any other marker would have to be bogus data in that case. + */ + +GLOBAL(boolean) +jpeg_resync_to_restart(j_decompress_ptr cinfo, int desired) +{ + int marker = cinfo->unread_marker; + int action = 1; + + /* Always put up a warning. */ + WARNMS2(cinfo, JWRN_MUST_RESYNC, marker, desired); + + /* Outer loop handles repeated decision after scanning forward. */ + for (;;) { + if (marker < (int)M_SOF0) + action = 2; /* invalid marker */ + else if (marker < (int)M_RST0 || marker > (int)M_RST7) + action = 3; /* valid non-restart marker */ + else { + if (marker == ((int)M_RST0 + ((desired + 1) & 7)) || + marker == ((int)M_RST0 + ((desired + 2) & 7))) + action = 3; /* one of the next two expected restarts */ + else if (marker == ((int)M_RST0 + ((desired - 1) & 7)) || + marker == ((int)M_RST0 + ((desired - 2) & 7))) + action = 2; /* a prior restart, so advance */ + else + action = 1; /* desired restart or too far away */ + } + TRACEMS2(cinfo, 4, JTRC_RECOVERY_ACTION, marker, action); + switch (action) { + case 1: + /* Discard marker and let entropy decoder resume processing. */ + cinfo->unread_marker = 0; + return TRUE; + case 2: + /* Scan to the next marker, and repeat the decision loop. */ + if (!next_marker(cinfo)) + return FALSE; + marker = cinfo->unread_marker; + break; + case 3: + /* Return without advancing past this marker. */ + /* Entropy decoder will be forced to process an empty segment. */ + return TRUE; + } + } /* end loop */ +} + + +/* + * Reset marker processing state to begin a fresh datastream. + */ + +METHODDEF(void) +reset_marker_reader(j_decompress_ptr cinfo) +{ + my_marker_ptr marker = (my_marker_ptr)cinfo->marker; + + cinfo->comp_info = NULL; /* until allocated by get_sof */ + cinfo->input_scan_number = 0; /* no SOS seen yet */ + cinfo->unread_marker = 0; /* no pending marker */ + marker->pub.saw_SOI = FALSE; /* set internal state too */ + marker->pub.saw_SOF = FALSE; + marker->pub.discarded_bytes = 0; + marker->cur_marker = NULL; +} + + +/* + * Initialize the marker reader module. + * This is called only once, when the decompression object is created. + */ + +GLOBAL(void) +jinit_marker_reader(j_decompress_ptr cinfo) +{ + my_marker_ptr marker; + int i; + + /* Create subobject in permanent pool */ + marker = (my_marker_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_PERMANENT, + sizeof(my_marker_reader)); + cinfo->marker = (struct jpeg_marker_reader *)marker; + /* Initialize public method pointers */ + marker->pub.reset_marker_reader = reset_marker_reader; + marker->pub.read_markers = read_markers; + marker->pub.read_restart_marker = read_restart_marker; + /* Initialize COM/APPn processing. + * By default, we examine and then discard APP0 and APP14, + * but simply discard COM and all other APPn. + */ + marker->process_COM = skip_variable; + marker->length_limit_COM = 0; + for (i = 0; i < 16; i++) { + marker->process_APPn[i] = skip_variable; + marker->length_limit_APPn[i] = 0; + } + marker->process_APPn[0] = get_interesting_appn; + marker->process_APPn[14] = get_interesting_appn; + /* Reset marker processing state */ + reset_marker_reader(cinfo); +} + + +/* + * Control saving of COM and APPn markers into marker_list. + */ + +#ifdef SAVE_MARKERS_SUPPORTED + +GLOBAL(void) +jpeg_save_markers(j_decompress_ptr cinfo, int marker_code, + unsigned int length_limit) +{ + my_marker_ptr marker = (my_marker_ptr)cinfo->marker; + long maxlength; + jpeg_marker_parser_method processor; + + /* Length limit mustn't be larger than what we can allocate + * (should only be a concern in a 16-bit environment). + */ + maxlength = cinfo->mem->max_alloc_chunk - sizeof(struct jpeg_marker_struct); + if (((long)length_limit) > maxlength) + length_limit = (unsigned int)maxlength; + + /* Choose processor routine to use. + * APP0/APP14 have special requirements. + */ + if (length_limit) { + processor = save_marker; + /* If saving APP0/APP14, save at least enough for our internal use. */ + if (marker_code == (int)M_APP0 && length_limit < APP0_DATA_LEN) + length_limit = APP0_DATA_LEN; + else if (marker_code == (int)M_APP14 && length_limit < APP14_DATA_LEN) + length_limit = APP14_DATA_LEN; + } else { + processor = skip_variable; + /* If discarding APP0/APP14, use our regular on-the-fly processor. */ + if (marker_code == (int)M_APP0 || marker_code == (int)M_APP14) + processor = get_interesting_appn; + } + + if (marker_code == (int)M_COM) { + marker->process_COM = processor; + marker->length_limit_COM = length_limit; + } else if (marker_code >= (int)M_APP0 && marker_code <= (int)M_APP15) { + marker->process_APPn[marker_code - (int)M_APP0] = processor; + marker->length_limit_APPn[marker_code - (int)M_APP0] = length_limit; + } else + ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, marker_code); +} + +#endif /* SAVE_MARKERS_SUPPORTED */ + + +/* + * Install a special processing method for COM or APPn markers. + */ + +GLOBAL(void) +jpeg_set_marker_processor(j_decompress_ptr cinfo, int marker_code, + jpeg_marker_parser_method routine) +{ + my_marker_ptr marker = (my_marker_ptr)cinfo->marker; + + if (marker_code == (int)M_COM) + marker->process_COM = routine; + else if (marker_code >= (int)M_APP0 && marker_code <= (int)M_APP15) + marker->process_APPn[marker_code - (int)M_APP0] = routine; + else + ERREXIT1(cinfo, JERR_UNKNOWN_MARKER, marker_code); +} diff --git a/src/spice2x/external/libjpeg-turbo/jdmaster.c b/src/spice2x/external/libjpeg-turbo/jdmaster.c new file mode 100644 index 0000000..80a4842 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmaster.c @@ -0,0 +1,891 @@ +/* + * jdmaster.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Modified 2002-2009 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2009-2011, 2016, 2019, 2022-2023, D. R. Commander. + * Copyright (C) 2013, Linaro Limited. + * Copyright (C) 2015, Google, Inc. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains master control logic for the JPEG decompressor. + * These routines are concerned with selecting the modules to be executed + * and with determining the number of passes and the work to be done in each + * pass. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jpegapicomp.h" +#include "jdmaster.h" + + +/* + * Determine whether merged upsample/color conversion should be used. + * CRUCIAL: this must match the actual capabilities of jdmerge.c! + */ + +LOCAL(boolean) +use_merged_upsample(j_decompress_ptr cinfo) +{ +#ifdef UPSAMPLE_MERGING_SUPPORTED + /* Colorspace conversion is not supported with lossless JPEG images */ + if (cinfo->master->lossless) + return FALSE; + /* Merging is the equivalent of plain box-filter upsampling */ + if (cinfo->do_fancy_upsampling || cinfo->CCIR601_sampling) + return FALSE; + /* jdmerge.c only supports YCC=>RGB and YCC=>RGB565 color conversion */ + if (cinfo->jpeg_color_space != JCS_YCbCr || cinfo->num_components != 3 || + (cinfo->out_color_space != JCS_RGB && + cinfo->out_color_space != JCS_RGB565 && + cinfo->out_color_space != JCS_EXT_RGB && + cinfo->out_color_space != JCS_EXT_RGBX && + cinfo->out_color_space != JCS_EXT_BGR && + cinfo->out_color_space != JCS_EXT_BGRX && + cinfo->out_color_space != JCS_EXT_XBGR && + cinfo->out_color_space != JCS_EXT_XRGB && + cinfo->out_color_space != JCS_EXT_RGBA && + cinfo->out_color_space != JCS_EXT_BGRA && + cinfo->out_color_space != JCS_EXT_ABGR && + cinfo->out_color_space != JCS_EXT_ARGB)) + return FALSE; + if ((cinfo->out_color_space == JCS_RGB565 && + cinfo->out_color_components != 3) || + (cinfo->out_color_space != JCS_RGB565 && + cinfo->out_color_components != rgb_pixelsize[cinfo->out_color_space])) + return FALSE; + /* and it only handles 2h1v or 2h2v sampling ratios */ + if (cinfo->comp_info[0].h_samp_factor != 2 || + cinfo->comp_info[1].h_samp_factor != 1 || + cinfo->comp_info[2].h_samp_factor != 1 || + cinfo->comp_info[0].v_samp_factor > 2 || + cinfo->comp_info[1].v_samp_factor != 1 || + cinfo->comp_info[2].v_samp_factor != 1) + return FALSE; + /* furthermore, it doesn't work if we've scaled the IDCTs differently */ + if (cinfo->comp_info[0]._DCT_scaled_size != cinfo->_min_DCT_scaled_size || + cinfo->comp_info[1]._DCT_scaled_size != cinfo->_min_DCT_scaled_size || + cinfo->comp_info[2]._DCT_scaled_size != cinfo->_min_DCT_scaled_size) + return FALSE; + /* ??? also need to test for upsample-time rescaling, when & if supported */ + return TRUE; /* by golly, it'll work... */ +#else + return FALSE; +#endif +} + + +/* + * Compute output image dimensions and related values. + * NOTE: this is exported for possible use by application. + * Hence it mustn't do anything that can't be done twice. + */ + +#if JPEG_LIB_VERSION >= 80 +GLOBAL(void) +#else +LOCAL(void) +#endif +jpeg_core_output_dimensions(j_decompress_ptr cinfo) +/* Do computations that are needed before master selection phase. + * This function is used for transcoding and full decompression. + */ +{ +#ifdef IDCT_SCALING_SUPPORTED + int ci; + jpeg_component_info *compptr; + + if (!cinfo->master->lossless) { + /* Compute actual output image dimensions and DCT scaling choices. */ + if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom) { + /* Provide 1/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 1; + cinfo->_min_DCT_v_scaled_size = 1; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 2) { + /* Provide 2/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 2L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 2L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 2; + cinfo->_min_DCT_v_scaled_size = 2; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 3) { + /* Provide 3/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 3L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 3L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 3; + cinfo->_min_DCT_v_scaled_size = 3; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 4) { + /* Provide 4/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 4L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 4L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 4; + cinfo->_min_DCT_v_scaled_size = 4; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 5) { + /* Provide 5/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 5L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 5L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 5; + cinfo->_min_DCT_v_scaled_size = 5; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 6) { + /* Provide 6/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 6L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 6L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 6; + cinfo->_min_DCT_v_scaled_size = 6; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 7) { + /* Provide 7/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 7L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 7L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 7; + cinfo->_min_DCT_v_scaled_size = 7; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 8) { + /* Provide 8/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 8L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 8L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 8; + cinfo->_min_DCT_v_scaled_size = 8; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 9) { + /* Provide 9/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 9L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 9L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 9; + cinfo->_min_DCT_v_scaled_size = 9; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 10) { + /* Provide 10/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 10L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 10L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 10; + cinfo->_min_DCT_v_scaled_size = 10; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 11) { + /* Provide 11/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 11L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 11L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 11; + cinfo->_min_DCT_v_scaled_size = 11; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 12) { + /* Provide 12/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 12L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 12L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 12; + cinfo->_min_DCT_v_scaled_size = 12; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 13) { + /* Provide 13/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 13L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 13L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 13; + cinfo->_min_DCT_v_scaled_size = 13; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 14) { + /* Provide 14/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 14L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 14L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 14; + cinfo->_min_DCT_v_scaled_size = 14; + } else if (cinfo->scale_num * DCTSIZE <= cinfo->scale_denom * 15) { + /* Provide 15/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 15L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 15L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 15; + cinfo->_min_DCT_v_scaled_size = 15; + } else { + /* Provide 16/block_size scaling */ + cinfo->output_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * 16L, (long)DCTSIZE); + cinfo->output_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * 16L, (long)DCTSIZE); + cinfo->_min_DCT_h_scaled_size = 16; + cinfo->_min_DCT_v_scaled_size = 16; + } + + /* Recompute dimensions of components */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + compptr->_DCT_h_scaled_size = cinfo->_min_DCT_h_scaled_size; + compptr->_DCT_v_scaled_size = cinfo->_min_DCT_v_scaled_size; + } + } else +#endif /* !IDCT_SCALING_SUPPORTED */ + { + /* Hardwire it to "no scaling" */ + cinfo->output_width = cinfo->image_width; + cinfo->output_height = cinfo->image_height; + /* jdinput.c has already initialized DCT_scaled_size, + * and has computed unscaled downsampled_width and downsampled_height. + */ + } +} + + +/* + * Compute output image dimensions and related values. + * NOTE: this is exported for possible use by application. + * Hence it mustn't do anything that can't be done twice. + * Also note that it may be called before the master module is initialized! + */ + +GLOBAL(void) +jpeg_calc_output_dimensions(j_decompress_ptr cinfo) +/* Do computations that are needed before master selection phase */ +{ +#ifdef IDCT_SCALING_SUPPORTED + int ci; + jpeg_component_info *compptr; +#endif + + /* Prevent application from calling me at wrong times */ + if (cinfo->global_state != DSTATE_READY) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + /* Compute core output image dimensions and DCT scaling choices. */ + jpeg_core_output_dimensions(cinfo); + +#ifdef IDCT_SCALING_SUPPORTED + + if (!cinfo->master->lossless) { + /* In selecting the actual DCT scaling for each component, we try to + * scale up the chroma components via IDCT scaling rather than upsampling. + * This saves time if the upsampler gets to use 1:1 scaling. + * Note this code adapts subsampling ratios which are powers of 2. + */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + int ssize = cinfo->_min_DCT_scaled_size; + while (ssize < DCTSIZE && + ((cinfo->max_h_samp_factor * cinfo->_min_DCT_scaled_size) % + (compptr->h_samp_factor * ssize * 2) == 0) && + ((cinfo->max_v_samp_factor * cinfo->_min_DCT_scaled_size) % + (compptr->v_samp_factor * ssize * 2) == 0)) { + ssize = ssize * 2; + } +#if JPEG_LIB_VERSION >= 70 + compptr->DCT_h_scaled_size = compptr->DCT_v_scaled_size = ssize; +#else + compptr->DCT_scaled_size = ssize; +#endif + } + + /* Recompute downsampled dimensions of components; + * application needs to know these if using raw downsampled data. + */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Size in samples, after IDCT scaling */ + compptr->downsampled_width = (JDIMENSION) + jdiv_round_up((long)cinfo->image_width * + (long)(compptr->h_samp_factor * + compptr->_DCT_scaled_size), + (long)(cinfo->max_h_samp_factor * DCTSIZE)); + compptr->downsampled_height = (JDIMENSION) + jdiv_round_up((long)cinfo->image_height * + (long)(compptr->v_samp_factor * + compptr->_DCT_scaled_size), + (long)(cinfo->max_v_samp_factor * DCTSIZE)); + } + } else +#endif /* IDCT_SCALING_SUPPORTED */ + { + /* Hardwire it to "no scaling" */ + cinfo->output_width = cinfo->image_width; + cinfo->output_height = cinfo->image_height; + /* jdinput.c has already initialized DCT_scaled_size to DCTSIZE, + * and has computed unscaled downsampled_width and downsampled_height. + */ + } + + /* Report number of components in selected colorspace. */ + /* Probably this should be in the color conversion module... */ + switch (cinfo->out_color_space) { + case JCS_GRAYSCALE: + cinfo->out_color_components = 1; + break; + case JCS_RGB: + case JCS_EXT_RGB: + case JCS_EXT_RGBX: + case JCS_EXT_BGR: + case JCS_EXT_BGRX: + case JCS_EXT_XBGR: + case JCS_EXT_XRGB: + case JCS_EXT_RGBA: + case JCS_EXT_BGRA: + case JCS_EXT_ABGR: + case JCS_EXT_ARGB: + cinfo->out_color_components = rgb_pixelsize[cinfo->out_color_space]; + break; + case JCS_YCbCr: + case JCS_RGB565: + cinfo->out_color_components = 3; + break; + case JCS_CMYK: + case JCS_YCCK: + cinfo->out_color_components = 4; + break; + default: /* else must be same colorspace as in file */ + cinfo->out_color_components = cinfo->num_components; + break; + } + cinfo->output_components = (cinfo->quantize_colors ? 1 : + cinfo->out_color_components); + + /* See if upsampler will want to emit more than one row at a time */ + if (use_merged_upsample(cinfo)) + cinfo->rec_outbuf_height = cinfo->max_v_samp_factor; + else + cinfo->rec_outbuf_height = 1; +} + + +/* + * Several decompression processes need to range-limit values to the range + * 0..MAXJSAMPLE; the input value may fall somewhat outside this range + * due to noise introduced by quantization, roundoff error, etc. These + * processes are inner loops and need to be as fast as possible. On most + * machines, particularly CPUs with pipelines or instruction prefetch, + * a (subscript-check-less) C table lookup + * x = sample_range_limit[x]; + * is faster than explicit tests + * if (x < 0) x = 0; + * else if (x > MAXJSAMPLE) x = MAXJSAMPLE; + * These processes all use a common table prepared by the routine below. + * + * For most steps we can mathematically guarantee that the initial value + * of x is within MAXJSAMPLE+1 of the legal range, so a table running from + * -(MAXJSAMPLE+1) to 2*MAXJSAMPLE+1 is sufficient. But for the initial + * limiting step (just after the IDCT), a wildly out-of-range value is + * possible if the input data is corrupt. To avoid any chance of indexing + * off the end of memory and getting a bad-pointer trap, we perform the + * post-IDCT limiting thus: + * x = range_limit[x & MASK]; + * where MASK is 2 bits wider than legal sample data, ie 10 bits for 8-bit + * samples. Under normal circumstances this is more than enough range and + * a correct output will be generated; with bogus input data the mask will + * cause wraparound, and we will safely generate a bogus-but-in-range output. + * For the post-IDCT step, we want to convert the data from signed to unsigned + * representation by adding CENTERJSAMPLE at the same time that we limit it. + * So the post-IDCT limiting table ends up looking like this: + * CENTERJSAMPLE,CENTERJSAMPLE+1,...,MAXJSAMPLE, + * MAXJSAMPLE (repeat 2*(MAXJSAMPLE+1)-CENTERJSAMPLE times), + * 0 (repeat 2*(MAXJSAMPLE+1)-CENTERJSAMPLE times), + * 0,1,...,CENTERJSAMPLE-1 + * Negative inputs select values from the upper half of the table after + * masking. + * + * We can save some space by overlapping the start of the post-IDCT table + * with the simpler range limiting table. The post-IDCT table begins at + * sample_range_limit + CENTERJSAMPLE. + */ + +LOCAL(void) +prepare_range_limit_table(j_decompress_ptr cinfo) +/* Allocate and fill in the sample_range_limit table */ +{ + JSAMPLE *table; + J12SAMPLE *table12; +#ifdef D_LOSSLESS_SUPPORTED + J16SAMPLE *table16; +#endif + int i; + + if (cinfo->data_precision == 16) { +#ifdef D_LOSSLESS_SUPPORTED + table16 = (J16SAMPLE *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (5 * (MAXJ16SAMPLE + 1) + CENTERJ16SAMPLE) * + sizeof(J16SAMPLE)); + table16 += (MAXJ16SAMPLE + 1); /* allow negative subscripts of simple + table */ + cinfo->sample_range_limit = (JSAMPLE *)table16; + /* First segment of "simple" table: limit[x] = 0 for x < 0 */ + memset(table16 - (MAXJ16SAMPLE + 1), 0, + (MAXJ16SAMPLE + 1) * sizeof(J16SAMPLE)); + /* Main part of "simple" table: limit[x] = x */ + for (i = 0; i <= MAXJ16SAMPLE; i++) + table16[i] = (J16SAMPLE)i; + table16 += CENTERJ16SAMPLE; /* Point to where post-IDCT table starts */ + /* End of simple table, rest of first half of post-IDCT table */ + for (i = CENTERJ16SAMPLE; i < 2 * (MAXJ16SAMPLE + 1); i++) + table16[i] = MAXJ16SAMPLE; + /* Second half of post-IDCT table */ + memset(table16 + (2 * (MAXJ16SAMPLE + 1)), 0, + (2 * (MAXJ16SAMPLE + 1) - CENTERJ16SAMPLE) * sizeof(J16SAMPLE)); + memcpy(table16 + (4 * (MAXJ16SAMPLE + 1) - CENTERJ16SAMPLE), + cinfo->sample_range_limit, CENTERJ16SAMPLE * sizeof(J16SAMPLE)); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); +#endif + } else if (cinfo->data_precision == 12) { + table12 = (J12SAMPLE *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (5 * (MAXJ12SAMPLE + 1) + CENTERJ12SAMPLE) * + sizeof(J12SAMPLE)); + table12 += (MAXJ12SAMPLE + 1); /* allow negative subscripts of simple + table */ + cinfo->sample_range_limit = (JSAMPLE *)table12; + /* First segment of "simple" table: limit[x] = 0 for x < 0 */ + memset(table12 - (MAXJ12SAMPLE + 1), 0, + (MAXJ12SAMPLE + 1) * sizeof(J12SAMPLE)); + /* Main part of "simple" table: limit[x] = x */ + for (i = 0; i <= MAXJ12SAMPLE; i++) + table12[i] = (J12SAMPLE)i; + table12 += CENTERJ12SAMPLE; /* Point to where post-IDCT table starts */ + /* End of simple table, rest of first half of post-IDCT table */ + for (i = CENTERJ12SAMPLE; i < 2 * (MAXJ12SAMPLE + 1); i++) + table12[i] = MAXJ12SAMPLE; + /* Second half of post-IDCT table */ + memset(table12 + (2 * (MAXJ12SAMPLE + 1)), 0, + (2 * (MAXJ12SAMPLE + 1) - CENTERJ12SAMPLE) * sizeof(J12SAMPLE)); + memcpy(table12 + (4 * (MAXJ12SAMPLE + 1) - CENTERJ12SAMPLE), + cinfo->sample_range_limit, CENTERJ12SAMPLE * sizeof(J12SAMPLE)); + } else { + table = (JSAMPLE *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (5 * (MAXJSAMPLE + 1) + CENTERJSAMPLE) * sizeof(JSAMPLE)); + table += (MAXJSAMPLE + 1); /* allow negative subscripts of simple table */ + cinfo->sample_range_limit = table; + /* First segment of "simple" table: limit[x] = 0 for x < 0 */ + memset(table - (MAXJSAMPLE + 1), 0, (MAXJSAMPLE + 1) * sizeof(JSAMPLE)); + /* Main part of "simple" table: limit[x] = x */ + for (i = 0; i <= MAXJSAMPLE; i++) + table[i] = (JSAMPLE)i; + table += CENTERJSAMPLE; /* Point to where post-IDCT table starts */ + /* End of simple table, rest of first half of post-IDCT table */ + for (i = CENTERJSAMPLE; i < 2 * (MAXJSAMPLE + 1); i++) + table[i] = MAXJSAMPLE; + /* Second half of post-IDCT table */ + memset(table + (2 * (MAXJSAMPLE + 1)), 0, + (2 * (MAXJSAMPLE + 1) - CENTERJSAMPLE) * sizeof(JSAMPLE)); + memcpy(table + (4 * (MAXJSAMPLE + 1) - CENTERJSAMPLE), + cinfo->sample_range_limit, CENTERJSAMPLE * sizeof(JSAMPLE)); + } +} + + +/* + * Master selection of decompression modules. + * This is done once at jpeg_start_decompress time. We determine + * which modules will be used and give them appropriate initialization calls. + * We also initialize the decompressor input side to begin consuming data. + * + * Since jpeg_read_header has finished, we know what is in the SOF + * and (first) SOS markers. We also have all the application parameter + * settings. + */ + +LOCAL(void) +master_selection(j_decompress_ptr cinfo) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + boolean use_c_buffer; + long samplesperrow; + JDIMENSION jd_samplesperrow; + + /* Disable IDCT scaling and raw (downsampled) data output in lossless mode. + * IDCT scaling is not useful in lossless mode, and it must be disabled in + * order to properly calculate the output dimensions. Raw data output isn't + * particularly useful without subsampling and has not been tested in + * lossless mode. + */ + if (cinfo->master->lossless) { + cinfo->raw_data_out = FALSE; + cinfo->scale_num = cinfo->scale_denom = 1; + } + + /* Initialize dimensions and other stuff */ + jpeg_calc_output_dimensions(cinfo); + prepare_range_limit_table(cinfo); + + /* Width of an output scanline must be representable as JDIMENSION. */ + samplesperrow = (long)cinfo->output_width * + (long)cinfo->out_color_components; + jd_samplesperrow = (JDIMENSION)samplesperrow; + if ((long)jd_samplesperrow != samplesperrow) + ERREXIT(cinfo, JERR_WIDTH_OVERFLOW); + + /* Initialize my private state */ + master->pass_number = 0; + master->using_merged_upsample = use_merged_upsample(cinfo); + + /* Color quantizer selection */ + master->quantizer_1pass = NULL; + master->quantizer_2pass = NULL; + /* No mode changes if not using buffered-image mode. */ + if (!cinfo->quantize_colors || !cinfo->buffered_image) { + cinfo->enable_1pass_quant = FALSE; + cinfo->enable_external_quant = FALSE; + cinfo->enable_2pass_quant = FALSE; + } + if (cinfo->quantize_colors) { + if (cinfo->raw_data_out) + ERREXIT(cinfo, JERR_NOTIMPL); + /* 2-pass quantizer only works in 3-component color space. */ + if (cinfo->out_color_components != 3 || + cinfo->out_color_space == JCS_RGB565) { + cinfo->enable_1pass_quant = TRUE; + cinfo->enable_external_quant = FALSE; + cinfo->enable_2pass_quant = FALSE; + cinfo->colormap = NULL; + } else if (cinfo->colormap != NULL) { + cinfo->enable_external_quant = TRUE; + } else if (cinfo->two_pass_quantize) { + cinfo->enable_2pass_quant = TRUE; + } else { + cinfo->enable_1pass_quant = TRUE; + } + + if (cinfo->enable_1pass_quant) { +#ifdef QUANT_1PASS_SUPPORTED + if (cinfo->data_precision == 16) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + else if (cinfo->data_precision == 12) + j12init_1pass_quantizer(cinfo); + else + jinit_1pass_quantizer(cinfo); + master->quantizer_1pass = cinfo->cquantize; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } + + /* We use the 2-pass code to map to external colormaps. */ + if (cinfo->enable_2pass_quant || cinfo->enable_external_quant) { +#ifdef QUANT_2PASS_SUPPORTED + if (cinfo->data_precision == 16) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + else if (cinfo->data_precision == 12) + j12init_2pass_quantizer(cinfo); + else + jinit_2pass_quantizer(cinfo); + master->quantizer_2pass = cinfo->cquantize; +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } + /* If both quantizers are initialized, the 2-pass one is left active; + * this is necessary for starting with quantization to an external map. + */ + } + + /* Post-processing: in particular, color conversion first */ + if (!cinfo->raw_data_out) { + if (master->using_merged_upsample) { +#ifdef UPSAMPLE_MERGING_SUPPORTED + if (cinfo->data_precision == 16) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + else if (cinfo->data_precision == 12) + j12init_merged_upsampler(cinfo); /* does color conversion too */ + else + jinit_merged_upsampler(cinfo); /* does color conversion too */ +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + if (cinfo->data_precision == 16) { +#ifdef D_LOSSLESS_SUPPORTED + j16init_color_deconverter(cinfo); + j16init_upsampler(cinfo); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); +#endif + } else if (cinfo->data_precision == 12) { + j12init_color_deconverter(cinfo); + j12init_upsampler(cinfo); + } else { + jinit_color_deconverter(cinfo); + jinit_upsampler(cinfo); + } + } + if (cinfo->data_precision == 16) +#ifdef D_LOSSLESS_SUPPORTED + j16init_d_post_controller(cinfo, cinfo->enable_2pass_quant); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); +#endif + else if (cinfo->data_precision == 12) + j12init_d_post_controller(cinfo, cinfo->enable_2pass_quant); + else + jinit_d_post_controller(cinfo, cinfo->enable_2pass_quant); + } + + if (cinfo->master->lossless) { +#ifdef D_LOSSLESS_SUPPORTED + /* Prediction, sample undifferencing, point transform, and sample size + * scaling + */ + if (cinfo->data_precision == 16) + j16init_lossless_decompressor(cinfo); + else if (cinfo->data_precision == 12) + j12init_lossless_decompressor(cinfo); + else + jinit_lossless_decompressor(cinfo); + /* Entropy decoding: either Huffman or arithmetic coding. */ + if (cinfo->arith_code) { + ERREXIT(cinfo, JERR_ARITH_NOTIMPL); + } else { + jinit_lhuff_decoder(cinfo); + } + + /* Initialize principal buffer controllers. */ + use_c_buffer = cinfo->inputctl->has_multiple_scans || + cinfo->buffered_image; + if (cinfo->data_precision == 16) + j16init_d_diff_controller(cinfo, use_c_buffer); + else if (cinfo->data_precision == 12) + j12init_d_diff_controller(cinfo, use_c_buffer); + else + jinit_d_diff_controller(cinfo, use_c_buffer); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else { + if (cinfo->data_precision == 16) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + /* Inverse DCT */ + if (cinfo->data_precision == 12) + j12init_inverse_dct(cinfo); + else + jinit_inverse_dct(cinfo); + /* Entropy decoding: either Huffman or arithmetic coding. */ + if (cinfo->arith_code) { +#ifdef D_ARITH_CODING_SUPPORTED + jinit_arith_decoder(cinfo); +#else + ERREXIT(cinfo, JERR_ARITH_NOTIMPL); +#endif + } else { + if (cinfo->progressive_mode) { +#ifdef D_PROGRESSIVE_SUPPORTED + jinit_phuff_decoder(cinfo); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else + jinit_huff_decoder(cinfo); + } + + /* Initialize principal buffer controllers. */ + use_c_buffer = cinfo->inputctl->has_multiple_scans || + cinfo->buffered_image; + if (cinfo->data_precision == 12) + j12init_d_coef_controller(cinfo, use_c_buffer); + else + jinit_d_coef_controller(cinfo, use_c_buffer); + } + + if (!cinfo->raw_data_out) { + if (cinfo->data_precision == 16) +#ifdef D_LOSSLESS_SUPPORTED + j16init_d_main_controller(cinfo, + FALSE /* never need full buffer here */); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); +#endif + else if (cinfo->data_precision == 12) + j12init_d_main_controller(cinfo, + FALSE /* never need full buffer here */); + else + jinit_d_main_controller(cinfo, FALSE /* never need full buffer here */); + } + + /* We can now tell the memory manager to allocate virtual arrays. */ + (*cinfo->mem->realize_virt_arrays) ((j_common_ptr)cinfo); + + /* Initialize input side of decompressor to consume first scan. */ + (*cinfo->inputctl->start_input_pass) (cinfo); + + /* Set the first and last iMCU columns to decompress from single-scan images. + * By default, decompress all of the iMCU columns. + */ + cinfo->master->first_iMCU_col = 0; + cinfo->master->last_iMCU_col = cinfo->MCUs_per_row - 1; + cinfo->master->last_good_iMCU_row = 0; + +#ifdef D_MULTISCAN_FILES_SUPPORTED + /* If jpeg_start_decompress will read the whole file, initialize + * progress monitoring appropriately. The input step is counted + * as one pass. + */ + if (cinfo->progress != NULL && !cinfo->buffered_image && + cinfo->inputctl->has_multiple_scans) { + int nscans; + /* Estimate number of scans to set pass_limit. */ + if (cinfo->progressive_mode) { + /* Arbitrarily estimate 2 interleaved DC scans + 3 AC scans/component. */ + nscans = 2 + 3 * cinfo->num_components; + } else { + /* For a nonprogressive multiscan file, estimate 1 scan per component. */ + nscans = cinfo->num_components; + } + cinfo->progress->pass_counter = 0L; + cinfo->progress->pass_limit = (long)cinfo->total_iMCU_rows * nscans; + cinfo->progress->completed_passes = 0; + cinfo->progress->total_passes = (cinfo->enable_2pass_quant ? 3 : 2); + /* Count the input pass as done */ + master->pass_number++; + } +#endif /* D_MULTISCAN_FILES_SUPPORTED */ +} + + +/* + * Per-pass setup. + * This is called at the beginning of each output pass. We determine which + * modules will be active during this pass and give them appropriate + * start_pass calls. We also set is_dummy_pass to indicate whether this + * is a "real" output pass or a dummy pass for color quantization. + * (In the latter case, jdapistd.c will crank the pass to completion.) + */ + +METHODDEF(void) +prepare_for_output_pass(j_decompress_ptr cinfo) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + + if (master->pub.is_dummy_pass) { +#ifdef QUANT_2PASS_SUPPORTED + /* Final pass of 2-pass quantization */ + master->pub.is_dummy_pass = FALSE; + (*cinfo->cquantize->start_pass) (cinfo, FALSE); + (*cinfo->post->start_pass) (cinfo, JBUF_CRANK_DEST); + (*cinfo->main->start_pass) (cinfo, JBUF_CRANK_DEST); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif /* QUANT_2PASS_SUPPORTED */ + } else { + if (cinfo->quantize_colors && cinfo->colormap == NULL) { + /* Select new quantization method */ + if (cinfo->two_pass_quantize && cinfo->enable_2pass_quant) { + cinfo->cquantize = master->quantizer_2pass; + master->pub.is_dummy_pass = TRUE; + } else if (cinfo->enable_1pass_quant) { + cinfo->cquantize = master->quantizer_1pass; + } else { + ERREXIT(cinfo, JERR_MODE_CHANGE); + } + } + (*cinfo->idct->start_pass) (cinfo); + (*cinfo->coef->start_output_pass) (cinfo); + if (!cinfo->raw_data_out) { + if (!master->using_merged_upsample) + (*cinfo->cconvert->start_pass) (cinfo); + (*cinfo->upsample->start_pass) (cinfo); + if (cinfo->quantize_colors) + (*cinfo->cquantize->start_pass) (cinfo, master->pub.is_dummy_pass); + (*cinfo->post->start_pass) (cinfo, + (master->pub.is_dummy_pass ? JBUF_SAVE_AND_PASS : JBUF_PASS_THRU)); + (*cinfo->main->start_pass) (cinfo, JBUF_PASS_THRU); + } + } + + /* Set up progress monitor's pass info if present */ + if (cinfo->progress != NULL) { + cinfo->progress->completed_passes = master->pass_number; + cinfo->progress->total_passes = master->pass_number + + (master->pub.is_dummy_pass ? 2 : 1); + /* In buffered-image mode, we assume one more output pass if EOI not + * yet reached, but no more passes if EOI has been reached. + */ + if (cinfo->buffered_image && !cinfo->inputctl->eoi_reached) { + cinfo->progress->total_passes += (cinfo->enable_2pass_quant ? 2 : 1); + } + } +} + + +/* + * Finish up at end of an output pass. + */ + +METHODDEF(void) +finish_output_pass(j_decompress_ptr cinfo) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + + if (cinfo->quantize_colors) + (*cinfo->cquantize->finish_pass) (cinfo); + master->pass_number++; +} + + +#ifdef D_MULTISCAN_FILES_SUPPORTED + +/* + * Switch to a new external colormap between output passes. + */ + +GLOBAL(void) +jpeg_new_colormap(j_decompress_ptr cinfo) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + + /* Prevent application from calling me at wrong times */ + if (cinfo->global_state != DSTATE_BUFIMAGE) + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + + if (cinfo->quantize_colors && cinfo->enable_external_quant && + cinfo->colormap != NULL) { + /* Select 2-pass quantizer for external colormap use */ + cinfo->cquantize = master->quantizer_2pass; + /* Notify quantizer of colormap change */ + (*cinfo->cquantize->new_color_map) (cinfo); + master->pub.is_dummy_pass = FALSE; /* just in case */ + } else + ERREXIT(cinfo, JERR_MODE_CHANGE); +} + +#endif /* D_MULTISCAN_FILES_SUPPORTED */ + + +/* + * Initialize master decompression control and select active modules. + * This is performed at the start of jpeg_start_decompress. + */ + +GLOBAL(void) +jinit_master_decompress(j_decompress_ptr cinfo) +{ + my_master_ptr master = (my_master_ptr)cinfo->master; + + master->pub.prepare_for_output_pass = prepare_for_output_pass; + master->pub.finish_output_pass = finish_output_pass; + + master->pub.is_dummy_pass = FALSE; + master->pub.jinit_upsampler_no_alloc = FALSE; + + master_selection(cinfo); +} diff --git a/src/spice2x/external/libjpeg-turbo/jdmaster.h b/src/spice2x/external/libjpeg-turbo/jdmaster.h new file mode 100644 index 0000000..76897e2 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmaster.h @@ -0,0 +1,28 @@ +/* + * jdmaster.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1995, Thomas G. Lane. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the master control structure for the JPEG decompressor. + */ + +/* Private state */ + +typedef struct { + struct jpeg_decomp_master pub; /* public fields */ + + int pass_number; /* # of passes completed */ + + boolean using_merged_upsample; /* TRUE if using merged upsample/cconvert */ + + /* Saved references to initialized quantizer modules, + * in case we need to switch modes. + */ + struct jpeg_color_quantizer *quantizer_1pass; + struct jpeg_color_quantizer *quantizer_2pass; +} my_decomp_master; + +typedef my_decomp_master *my_master_ptr; diff --git a/src/spice2x/external/libjpeg-turbo/jdmerge.c b/src/spice2x/external/libjpeg-turbo/jdmerge.c new file mode 100644 index 0000000..49f2006 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmerge.c @@ -0,0 +1,594 @@ +/* + * jdmerge.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2009, 2011, 2014-2015, 2020, 2022, D. R. Commander. + * Copyright (C) 2013, Linaro Limited. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains code for merged upsampling/color conversion. + * + * This file combines functions from jdsample.c and jdcolor.c; + * read those files first to understand what's going on. + * + * When the chroma components are to be upsampled by simple replication + * (ie, box filtering), we can save some work in color conversion by + * calculating all the output pixels corresponding to a pair of chroma + * samples at one time. In the conversion equations + * R = Y + K1 * Cr + * G = Y + K2 * Cb + K3 * Cr + * B = Y + K4 * Cb + * only the Y term varies among the group of pixels corresponding to a pair + * of chroma samples, so the rest of the terms can be calculated just once. + * At typical sampling ratios, this eliminates half or three-quarters of the + * multiplications needed for color conversion. + * + * This file currently provides implementations for the following cases: + * YCbCr => RGB color conversion only. + * Sampling ratios of 2h1v or 2h2v. + * No scaling needed at upsample time. + * Corner-aligned (non-CCIR601) sampling alignment. + * Other special cases could be added, but in most applications these are + * the only common cases. (For uncommon cases we fall back on the more + * general code in jdsample.c and jdcolor.c.) + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdmerge.h" +#include "jsimd.h" + +#ifdef UPSAMPLE_MERGING_SUPPORTED + + +#define SCALEBITS 16 /* speediest right-shift on some machines */ +#define ONE_HALF ((JLONG)1 << (SCALEBITS - 1)) +#define FIX(x) ((JLONG)((x) * (1L << SCALEBITS) + 0.5)) + + +/* Include inline routines for colorspace extensions */ + +#include "jdmrgext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE + +#define RGB_RED EXT_RGB_RED +#define RGB_GREEN EXT_RGB_GREEN +#define RGB_BLUE EXT_RGB_BLUE +#define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +#define h2v1_merged_upsample_internal extrgb_h2v1_merged_upsample_internal +#define h2v2_merged_upsample_internal extrgb_h2v2_merged_upsample_internal +#include "jdmrgext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef h2v1_merged_upsample_internal +#undef h2v2_merged_upsample_internal + +#define RGB_RED EXT_RGBX_RED +#define RGB_GREEN EXT_RGBX_GREEN +#define RGB_BLUE EXT_RGBX_BLUE +#define RGB_ALPHA 3 +#define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +#define h2v1_merged_upsample_internal extrgbx_h2v1_merged_upsample_internal +#define h2v2_merged_upsample_internal extrgbx_h2v2_merged_upsample_internal +#include "jdmrgext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef h2v1_merged_upsample_internal +#undef h2v2_merged_upsample_internal + +#define RGB_RED EXT_BGR_RED +#define RGB_GREEN EXT_BGR_GREEN +#define RGB_BLUE EXT_BGR_BLUE +#define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +#define h2v1_merged_upsample_internal extbgr_h2v1_merged_upsample_internal +#define h2v2_merged_upsample_internal extbgr_h2v2_merged_upsample_internal +#include "jdmrgext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_PIXELSIZE +#undef h2v1_merged_upsample_internal +#undef h2v2_merged_upsample_internal + +#define RGB_RED EXT_BGRX_RED +#define RGB_GREEN EXT_BGRX_GREEN +#define RGB_BLUE EXT_BGRX_BLUE +#define RGB_ALPHA 3 +#define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +#define h2v1_merged_upsample_internal extbgrx_h2v1_merged_upsample_internal +#define h2v2_merged_upsample_internal extbgrx_h2v2_merged_upsample_internal +#include "jdmrgext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef h2v1_merged_upsample_internal +#undef h2v2_merged_upsample_internal + +#define RGB_RED EXT_XBGR_RED +#define RGB_GREEN EXT_XBGR_GREEN +#define RGB_BLUE EXT_XBGR_BLUE +#define RGB_ALPHA 0 +#define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +#define h2v1_merged_upsample_internal extxbgr_h2v1_merged_upsample_internal +#define h2v2_merged_upsample_internal extxbgr_h2v2_merged_upsample_internal +#include "jdmrgext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef h2v1_merged_upsample_internal +#undef h2v2_merged_upsample_internal + +#define RGB_RED EXT_XRGB_RED +#define RGB_GREEN EXT_XRGB_GREEN +#define RGB_BLUE EXT_XRGB_BLUE +#define RGB_ALPHA 0 +#define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +#define h2v1_merged_upsample_internal extxrgb_h2v1_merged_upsample_internal +#define h2v2_merged_upsample_internal extxrgb_h2v2_merged_upsample_internal +#include "jdmrgext.c" +#undef RGB_RED +#undef RGB_GREEN +#undef RGB_BLUE +#undef RGB_ALPHA +#undef RGB_PIXELSIZE +#undef h2v1_merged_upsample_internal +#undef h2v2_merged_upsample_internal + + +/* + * Initialize tables for YCC->RGB colorspace conversion. + * This is taken directly from jdcolor.c; see that file for more info. + */ + +LOCAL(void) +build_ycc_rgb_table(j_decompress_ptr cinfo) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + int i; + JLONG x; + SHIFT_TEMPS + + upsample->Cr_r_tab = (int *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(int)); + upsample->Cb_b_tab = (int *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(int)); + upsample->Cr_g_tab = (JLONG *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(JLONG)); + upsample->Cb_g_tab = (JLONG *) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (_MAXJSAMPLE + 1) * sizeof(JLONG)); + + for (i = 0, x = -_CENTERJSAMPLE; i <= _MAXJSAMPLE; i++, x++) { + /* i is the actual input pixel value, in the range 0.._MAXJSAMPLE */ + /* The Cb or Cr value we are thinking of is x = i - _CENTERJSAMPLE */ + /* Cr=>R value is nearest int to 1.40200 * x */ + upsample->Cr_r_tab[i] = (int) + RIGHT_SHIFT(FIX(1.40200) * x + ONE_HALF, SCALEBITS); + /* Cb=>B value is nearest int to 1.77200 * x */ + upsample->Cb_b_tab[i] = (int) + RIGHT_SHIFT(FIX(1.77200) * x + ONE_HALF, SCALEBITS); + /* Cr=>G value is scaled-up -0.71414 * x */ + upsample->Cr_g_tab[i] = (-FIX(0.71414)) * x; + /* Cb=>G value is scaled-up -0.34414 * x */ + /* We also add in ONE_HALF so that need not do it in inner loop */ + upsample->Cb_g_tab[i] = (-FIX(0.34414)) * x + ONE_HALF; + } +} + + +/* + * Initialize for an upsampling pass. + */ + +METHODDEF(void) +start_pass_merged_upsample(j_decompress_ptr cinfo) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + + /* Mark the spare buffer empty */ + upsample->spare_full = FALSE; + /* Initialize total-height counter for detecting bottom of image */ + upsample->rows_to_go = cinfo->output_height; +} + + +/* + * Control routine to do upsampling (and color conversion). + * + * The control routine just handles the row buffering considerations. + */ + +METHODDEF(void) +merged_2v_upsample(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +/* 2:1 vertical sampling case: may need a spare row. */ +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + _JSAMPROW work_ptrs[2]; + JDIMENSION num_rows; /* number of rows returned to caller */ + + if (upsample->spare_full) { + /* If we have a spare row saved from a previous cycle, just return it. */ + JDIMENSION size = upsample->out_row_width; + if (cinfo->out_color_space == JCS_RGB565) + size = cinfo->output_width * 2; + _jcopy_sample_rows(&upsample->spare_row, 0, output_buf + *out_row_ctr, 0, + 1, size); + num_rows = 1; + upsample->spare_full = FALSE; + } else { + /* Figure number of rows to return to caller. */ + num_rows = 2; + /* Not more than the distance to the end of the image. */ + if (num_rows > upsample->rows_to_go) + num_rows = upsample->rows_to_go; + /* And not more than what the client can accept: */ + out_rows_avail -= *out_row_ctr; + if (num_rows > out_rows_avail) + num_rows = out_rows_avail; + /* Create output pointer array for upsampler. */ + work_ptrs[0] = output_buf[*out_row_ctr]; + if (num_rows > 1) { + work_ptrs[1] = output_buf[*out_row_ctr + 1]; + } else { + work_ptrs[1] = upsample->spare_row; + upsample->spare_full = TRUE; + } + /* Now do the upsampling. */ + (*upsample->upmethod) (cinfo, input_buf, *in_row_group_ctr, work_ptrs); + } + + /* Adjust counts */ + *out_row_ctr += num_rows; + upsample->rows_to_go -= num_rows; + /* When the buffer is emptied, declare this input row group consumed */ + if (!upsample->spare_full) + (*in_row_group_ctr)++; +} + + +METHODDEF(void) +merged_1v_upsample(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +/* 1:1 vertical sampling case: much easier, never need a spare row. */ +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + + /* Just do the upsampling. */ + (*upsample->upmethod) (cinfo, input_buf, *in_row_group_ctr, + output_buf + *out_row_ctr); + /* Adjust counts */ + (*out_row_ctr)++; + (*in_row_group_ctr)++; +} + + +/* + * These are the routines invoked by the control routines to do + * the actual upsampling/conversion. One row group is processed per call. + * + * Note: since we may be writing directly into application-supplied buffers, + * we have to be honest about the output width; we can't assume the buffer + * has been rounded up to an even width. + */ + + +/* + * Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. + */ + +METHODDEF(void) +h2v1_merged_upsample(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, _JSAMPARRAY output_buf) +{ + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + extrgb_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + extrgbx_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_BGR: + extbgr_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + extbgrx_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + extxbgr_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + extxrgb_h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + default: + h2v1_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + } +} + + +/* + * Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. + */ + +METHODDEF(void) +h2v2_merged_upsample(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, _JSAMPARRAY output_buf) +{ + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + extrgb_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + extrgbx_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_BGR: + extbgr_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + extbgrx_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + extxbgr_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + extxrgb_h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + default: + h2v2_merged_upsample_internal(cinfo, input_buf, in_row_group_ctr, + output_buf); + break; + } +} + + +/* + * RGB565 conversion + */ + +#define PACK_SHORT_565_LE(r, g, b) \ + ((((r) << 8) & 0xF800) | (((g) << 3) & 0x7E0) | ((b) >> 3)) +#define PACK_SHORT_565_BE(r, g, b) \ + (((r) & 0xF8) | ((g) >> 5) | (((g) << 11) & 0xE000) | (((b) << 5) & 0x1F00)) + +#define PACK_TWO_PIXELS_LE(l, r) ((r << 16) | l) +#define PACK_TWO_PIXELS_BE(l, r) ((l << 16) | r) + +#define WRITE_TWO_PIXELS_LE(addr, pixels) { \ + ((INT16 *)(addr))[0] = (INT16)(pixels); \ + ((INT16 *)(addr))[1] = (INT16)((pixels) >> 16); \ +} +#define WRITE_TWO_PIXELS_BE(addr, pixels) { \ + ((INT16 *)(addr))[1] = (INT16)(pixels); \ + ((INT16 *)(addr))[0] = (INT16)((pixels) >> 16); \ +} + +#define DITHER_565_R(r, dither) ((r) + ((dither) & 0xFF)) +#define DITHER_565_G(g, dither) ((g) + (((dither) & 0xFF) >> 1)) +#define DITHER_565_B(b, dither) ((b) + ((dither) & 0xFF)) + + +/* Declarations for ordered dithering + * + * We use a 4x4 ordered dither array packed into 32 bits. This array is + * sufficient for dithering RGB888 to RGB565. + */ + +#define DITHER_MASK 0x3 +#define DITHER_ROTATE(x) ((((x) & 0xFF) << 24) | (((x) >> 8) & 0x00FFFFFF)) +static const JLONG dither_matrix[4] = { + 0x0008020A, + 0x0C040E06, + 0x030B0109, + 0x0F070D05 +}; + + +/* Include inline routines for RGB565 conversion */ + +#define PACK_SHORT_565 PACK_SHORT_565_LE +#define PACK_TWO_PIXELS PACK_TWO_PIXELS_LE +#define WRITE_TWO_PIXELS WRITE_TWO_PIXELS_LE +#define h2v1_merged_upsample_565_internal h2v1_merged_upsample_565_le +#define h2v1_merged_upsample_565D_internal h2v1_merged_upsample_565D_le +#define h2v2_merged_upsample_565_internal h2v2_merged_upsample_565_le +#define h2v2_merged_upsample_565D_internal h2v2_merged_upsample_565D_le +#include "jdmrg565.c" +#undef PACK_SHORT_565 +#undef PACK_TWO_PIXELS +#undef WRITE_TWO_PIXELS +#undef h2v1_merged_upsample_565_internal +#undef h2v1_merged_upsample_565D_internal +#undef h2v2_merged_upsample_565_internal +#undef h2v2_merged_upsample_565D_internal + +#define PACK_SHORT_565 PACK_SHORT_565_BE +#define PACK_TWO_PIXELS PACK_TWO_PIXELS_BE +#define WRITE_TWO_PIXELS WRITE_TWO_PIXELS_BE +#define h2v1_merged_upsample_565_internal h2v1_merged_upsample_565_be +#define h2v1_merged_upsample_565D_internal h2v1_merged_upsample_565D_be +#define h2v2_merged_upsample_565_internal h2v2_merged_upsample_565_be +#define h2v2_merged_upsample_565D_internal h2v2_merged_upsample_565D_be +#include "jdmrg565.c" +#undef PACK_SHORT_565 +#undef PACK_TWO_PIXELS +#undef WRITE_TWO_PIXELS +#undef h2v1_merged_upsample_565_internal +#undef h2v1_merged_upsample_565D_internal +#undef h2v2_merged_upsample_565_internal +#undef h2v2_merged_upsample_565D_internal + + +static INLINE boolean is_big_endian(void) +{ + int test_value = 1; + if (*(char *)&test_value != 1) + return TRUE; + return FALSE; +} + + +METHODDEF(void) +h2v1_merged_upsample_565(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, _JSAMPARRAY output_buf) +{ + if (is_big_endian()) + h2v1_merged_upsample_565_be(cinfo, input_buf, in_row_group_ctr, + output_buf); + else + h2v1_merged_upsample_565_le(cinfo, input_buf, in_row_group_ctr, + output_buf); +} + + +METHODDEF(void) +h2v1_merged_upsample_565D(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, _JSAMPARRAY output_buf) +{ + if (is_big_endian()) + h2v1_merged_upsample_565D_be(cinfo, input_buf, in_row_group_ctr, + output_buf); + else + h2v1_merged_upsample_565D_le(cinfo, input_buf, in_row_group_ctr, + output_buf); +} + + +METHODDEF(void) +h2v2_merged_upsample_565(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, _JSAMPARRAY output_buf) +{ + if (is_big_endian()) + h2v2_merged_upsample_565_be(cinfo, input_buf, in_row_group_ctr, + output_buf); + else + h2v2_merged_upsample_565_le(cinfo, input_buf, in_row_group_ctr, + output_buf); +} + + +METHODDEF(void) +h2v2_merged_upsample_565D(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, _JSAMPARRAY output_buf) +{ + if (is_big_endian()) + h2v2_merged_upsample_565D_be(cinfo, input_buf, in_row_group_ctr, + output_buf); + else + h2v2_merged_upsample_565D_le(cinfo, input_buf, in_row_group_ctr, + output_buf); +} + + +/* + * Module initialization routine for merged upsampling/color conversion. + * + * NB: this is called under the conditions determined by use_merged_upsample() + * in jdmaster.c. That routine MUST correspond to the actual capabilities + * of this module; no safety checks are made here. + */ + +GLOBAL(void) +_jinit_merged_upsampler(j_decompress_ptr cinfo) +{ + my_merged_upsample_ptr upsample; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + upsample = (my_merged_upsample_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_merged_upsampler)); + cinfo->upsample = (struct jpeg_upsampler *)upsample; + upsample->pub.start_pass = start_pass_merged_upsample; + upsample->pub.need_context_rows = FALSE; + + upsample->out_row_width = cinfo->output_width * cinfo->out_color_components; + + if (cinfo->max_v_samp_factor == 2) { + upsample->pub._upsample = merged_2v_upsample; +#ifdef WITH_SIMD + if (jsimd_can_h2v2_merged_upsample()) + upsample->upmethod = jsimd_h2v2_merged_upsample; + else +#endif + upsample->upmethod = h2v2_merged_upsample; + if (cinfo->out_color_space == JCS_RGB565) { + if (cinfo->dither_mode != JDITHER_NONE) { + upsample->upmethod = h2v2_merged_upsample_565D; + } else { + upsample->upmethod = h2v2_merged_upsample_565; + } + } + /* Allocate a spare row buffer */ + upsample->spare_row = (_JSAMPROW) + (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, + (size_t)(upsample->out_row_width * sizeof(_JSAMPLE))); + } else { + upsample->pub._upsample = merged_1v_upsample; +#ifdef WITH_SIMD + if (jsimd_can_h2v1_merged_upsample()) + upsample->upmethod = jsimd_h2v1_merged_upsample; + else +#endif + upsample->upmethod = h2v1_merged_upsample; + if (cinfo->out_color_space == JCS_RGB565) { + if (cinfo->dither_mode != JDITHER_NONE) { + upsample->upmethod = h2v1_merged_upsample_565D; + } else { + upsample->upmethod = h2v1_merged_upsample_565; + } + } + /* No spare row needed */ + upsample->spare_row = NULL; + } + + build_ycc_rgb_table(cinfo); +} + +#endif /* UPSAMPLE_MERGING_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jdmerge.h b/src/spice2x/external/libjpeg-turbo/jdmerge.h new file mode 100644 index 0000000..73cbd60 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmerge.h @@ -0,0 +1,48 @@ +/* + * jdmerge.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + */ + +#define JPEG_INTERNALS +#include "jpeglib.h" +#include "jsamplecomp.h" + +#ifdef UPSAMPLE_MERGING_SUPPORTED + + +/* Private subobject */ + +typedef struct { + struct jpeg_upsampler pub; /* public fields */ + + /* Pointer to routine to do actual upsampling/conversion of one row group */ + void (*upmethod) (j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, _JSAMPARRAY output_buf); + + /* Private state for YCC->RGB conversion */ + int *Cr_r_tab; /* => table for Cr to R conversion */ + int *Cb_b_tab; /* => table for Cb to B conversion */ + JLONG *Cr_g_tab; /* => table for Cr to G conversion */ + JLONG *Cb_g_tab; /* => table for Cb to G conversion */ + + /* For 2:1 vertical sampling, we produce two output rows at a time. + * We need a "spare" row buffer to hold the second output row if the + * application provides just a one-row buffer; we also use the spare + * to discard the dummy last row if the image height is odd. + */ + _JSAMPROW spare_row; + boolean spare_full; /* T if spare buffer is occupied */ + + JDIMENSION out_row_width; /* samples per output row */ + JDIMENSION rows_to_go; /* counts rows remaining in image */ +} my_merged_upsampler; + +typedef my_merged_upsampler *my_merged_upsample_ptr; + +#endif /* UPSAMPLE_MERGING_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jdmrg565.c b/src/spice2x/external/libjpeg-turbo/jdmrg565.c new file mode 100644 index 0000000..0c719b9 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmrg565.c @@ -0,0 +1,355 @@ +/* + * jdmrg565.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2013, Linaro Limited. + * Copyright (C) 2014-2015, 2018, 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains code for merged upsampling/color conversion. + */ + + +INLINE +LOCAL(void) +h2v1_merged_upsample_565_internal(j_decompress_ptr cinfo, + _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + _JSAMPARRAY output_buf) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + register int y, cred, cgreen, cblue; + int cb, cr; + register _JSAMPROW outptr; + _JSAMPROW inptr0, inptr1, inptr2; + JDIMENSION col; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int *Crrtab = upsample->Cr_r_tab; + int *Cbbtab = upsample->Cb_b_tab; + JLONG *Crgtab = upsample->Cr_g_tab; + JLONG *Cbgtab = upsample->Cb_g_tab; + unsigned int r, g, b; + JLONG rgb; + SHIFT_TEMPS + + inptr0 = input_buf[0][in_row_group_ctr]; + inptr1 = input_buf[1][in_row_group_ctr]; + inptr2 = input_buf[2][in_row_group_ctr]; + outptr = output_buf[0]; + + /* Loop for each pair of output pixels */ + for (col = cinfo->output_width >> 1; col > 0; col--) { + /* Do the chroma part of the calculation */ + cb = *inptr1++; + cr = *inptr2++; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + + /* Fetch 2 Y values and emit 2 pixels */ + y = *inptr0++; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr0++; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_PIXELS(outptr, rgb); + outptr += 4; + } + + /* If image width is odd, do the last output column separately */ + if (cinfo->output_width & 1) { + cb = *inptr1; + cr = *inptr2; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + y = *inptr0; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + } +} + + +INLINE +LOCAL(void) +h2v1_merged_upsample_565D_internal(j_decompress_ptr cinfo, + _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + _JSAMPARRAY output_buf) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + register int y, cred, cgreen, cblue; + int cb, cr; + register _JSAMPROW outptr; + _JSAMPROW inptr0, inptr1, inptr2; + JDIMENSION col; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int *Crrtab = upsample->Cr_r_tab; + int *Cbbtab = upsample->Cb_b_tab; + JLONG *Crgtab = upsample->Cr_g_tab; + JLONG *Cbgtab = upsample->Cb_g_tab; + JLONG d0 = dither_matrix[cinfo->output_scanline & DITHER_MASK]; + unsigned int r, g, b; + JLONG rgb; + SHIFT_TEMPS + + inptr0 = input_buf[0][in_row_group_ctr]; + inptr1 = input_buf[1][in_row_group_ctr]; + inptr2 = input_buf[2][in_row_group_ctr]; + outptr = output_buf[0]; + + /* Loop for each pair of output pixels */ + for (col = cinfo->output_width >> 1; col > 0; col--) { + /* Do the chroma part of the calculation */ + cb = *inptr1++; + cr = *inptr2++; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + + /* Fetch 2 Y values and emit 2 pixels */ + y = *inptr0++; + r = range_limit[DITHER_565_R(y + cred, d0)]; + g = range_limit[DITHER_565_G(y + cgreen, d0)]; + b = range_limit[DITHER_565_B(y + cblue, d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr0++; + r = range_limit[DITHER_565_R(y + cred, d0)]; + g = range_limit[DITHER_565_G(y + cgreen, d0)]; + b = range_limit[DITHER_565_B(y + cblue, d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_PIXELS(outptr, rgb); + outptr += 4; + } + + /* If image width is odd, do the last output column separately */ + if (cinfo->output_width & 1) { + cb = *inptr1; + cr = *inptr2; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + y = *inptr0; + r = range_limit[DITHER_565_R(y + cred, d0)]; + g = range_limit[DITHER_565_G(y + cgreen, d0)]; + b = range_limit[DITHER_565_B(y + cblue, d0)]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr = (INT16)rgb; + } +} + + +INLINE +LOCAL(void) +h2v2_merged_upsample_565_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + _JSAMPARRAY output_buf) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + register int y, cred, cgreen, cblue; + int cb, cr; + register _JSAMPROW outptr0, outptr1; + _JSAMPROW inptr00, inptr01, inptr1, inptr2; + JDIMENSION col; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int *Crrtab = upsample->Cr_r_tab; + int *Cbbtab = upsample->Cb_b_tab; + JLONG *Crgtab = upsample->Cr_g_tab; + JLONG *Cbgtab = upsample->Cb_g_tab; + unsigned int r, g, b; + JLONG rgb; + SHIFT_TEMPS + + inptr00 = input_buf[0][in_row_group_ctr * 2]; + inptr01 = input_buf[0][in_row_group_ctr * 2 + 1]; + inptr1 = input_buf[1][in_row_group_ctr]; + inptr2 = input_buf[2][in_row_group_ctr]; + outptr0 = output_buf[0]; + outptr1 = output_buf[1]; + + /* Loop for each group of output pixels */ + for (col = cinfo->output_width >> 1; col > 0; col--) { + /* Do the chroma part of the calculation */ + cb = *inptr1++; + cr = *inptr2++; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + + /* Fetch 4 Y values and emit 4 pixels */ + y = *inptr00++; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr00++; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_PIXELS(outptr0, rgb); + outptr0 += 4; + + y = *inptr01++; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr01++; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_PIXELS(outptr1, rgb); + outptr1 += 4; + } + + /* If image width is odd, do the last output column separately */ + if (cinfo->output_width & 1) { + cb = *inptr1; + cr = *inptr2; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + + y = *inptr00; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr0 = (INT16)rgb; + + y = *inptr01; + r = range_limit[y + cred]; + g = range_limit[y + cgreen]; + b = range_limit[y + cblue]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr1 = (INT16)rgb; + } +} + + +INLINE +LOCAL(void) +h2v2_merged_upsample_565D_internal(j_decompress_ptr cinfo, + _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + _JSAMPARRAY output_buf) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + register int y, cred, cgreen, cblue; + int cb, cr; + register _JSAMPROW outptr0, outptr1; + _JSAMPROW inptr00, inptr01, inptr1, inptr2; + JDIMENSION col; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int *Crrtab = upsample->Cr_r_tab; + int *Cbbtab = upsample->Cb_b_tab; + JLONG *Crgtab = upsample->Cr_g_tab; + JLONG *Cbgtab = upsample->Cb_g_tab; + JLONG d0 = dither_matrix[cinfo->output_scanline & DITHER_MASK]; + JLONG d1 = dither_matrix[(cinfo->output_scanline + 1) & DITHER_MASK]; + unsigned int r, g, b; + JLONG rgb; + SHIFT_TEMPS + + inptr00 = input_buf[0][in_row_group_ctr * 2]; + inptr01 = input_buf[0][in_row_group_ctr * 2 + 1]; + inptr1 = input_buf[1][in_row_group_ctr]; + inptr2 = input_buf[2][in_row_group_ctr]; + outptr0 = output_buf[0]; + outptr1 = output_buf[1]; + + /* Loop for each group of output pixels */ + for (col = cinfo->output_width >> 1; col > 0; col--) { + /* Do the chroma part of the calculation */ + cb = *inptr1++; + cr = *inptr2++; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + + /* Fetch 4 Y values and emit 4 pixels */ + y = *inptr00++; + r = range_limit[DITHER_565_R(y + cred, d0)]; + g = range_limit[DITHER_565_G(y + cgreen, d0)]; + b = range_limit[DITHER_565_B(y + cblue, d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr00++; + r = range_limit[DITHER_565_R(y + cred, d0)]; + g = range_limit[DITHER_565_G(y + cgreen, d0)]; + b = range_limit[DITHER_565_B(y + cblue, d0)]; + d0 = DITHER_ROTATE(d0); + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_PIXELS(outptr0, rgb); + outptr0 += 4; + + y = *inptr01++; + r = range_limit[DITHER_565_R(y + cred, d1)]; + g = range_limit[DITHER_565_G(y + cgreen, d1)]; + b = range_limit[DITHER_565_B(y + cblue, d1)]; + d1 = DITHER_ROTATE(d1); + rgb = PACK_SHORT_565(r, g, b); + + y = *inptr01++; + r = range_limit[DITHER_565_R(y + cred, d1)]; + g = range_limit[DITHER_565_G(y + cgreen, d1)]; + b = range_limit[DITHER_565_B(y + cblue, d1)]; + d1 = DITHER_ROTATE(d1); + rgb = PACK_TWO_PIXELS(rgb, PACK_SHORT_565(r, g, b)); + + WRITE_TWO_PIXELS(outptr1, rgb); + outptr1 += 4; + } + + /* If image width is odd, do the last output column separately */ + if (cinfo->output_width & 1) { + cb = *inptr1; + cr = *inptr2; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + + y = *inptr00; + r = range_limit[DITHER_565_R(y + cred, d0)]; + g = range_limit[DITHER_565_G(y + cgreen, d0)]; + b = range_limit[DITHER_565_B(y + cblue, d0)]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr0 = (INT16)rgb; + + y = *inptr01; + r = range_limit[DITHER_565_R(y + cred, d1)]; + g = range_limit[DITHER_565_G(y + cgreen, d1)]; + b = range_limit[DITHER_565_B(y + cblue, d1)]; + rgb = PACK_SHORT_565(r, g, b); + *(INT16 *)outptr1 = (INT16)rgb; + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jdmrgext.c b/src/spice2x/external/libjpeg-turbo/jdmrgext.c new file mode 100644 index 0000000..8139e0a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdmrgext.c @@ -0,0 +1,184 @@ +/* + * jdmrgext.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2011, 2015, 2020, 2022-2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains code for merged upsampling/color conversion. + */ + + +/* This file is included by jdmerge.c */ + + +/* + * Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. + */ + +INLINE +LOCAL(void) +h2v1_merged_upsample_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + _JSAMPARRAY output_buf) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + register int y, cred, cgreen, cblue; + int cb, cr; + register _JSAMPROW outptr; + _JSAMPROW inptr0, inptr1, inptr2; + JDIMENSION col; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int *Crrtab = upsample->Cr_r_tab; + int *Cbbtab = upsample->Cb_b_tab; + JLONG *Crgtab = upsample->Cr_g_tab; + JLONG *Cbgtab = upsample->Cb_g_tab; + SHIFT_TEMPS + + inptr0 = input_buf[0][in_row_group_ctr]; + inptr1 = input_buf[1][in_row_group_ctr]; + inptr2 = input_buf[2][in_row_group_ctr]; + outptr = output_buf[0]; + /* Loop for each pair of output pixels */ + for (col = cinfo->output_width >> 1; col > 0; col--) { + /* Do the chroma part of the calculation */ + cb = *inptr1++; + cr = *inptr2++; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + /* Fetch 2 Y values and emit 2 pixels */ + y = *inptr0++; + outptr[RGB_RED] = range_limit[y + cred]; + outptr[RGB_GREEN] = range_limit[y + cgreen]; + outptr[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr += RGB_PIXELSIZE; + y = *inptr0++; + outptr[RGB_RED] = range_limit[y + cred]; + outptr[RGB_GREEN] = range_limit[y + cgreen]; + outptr[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr += RGB_PIXELSIZE; + } + /* If image width is odd, do the last output column separately */ + if (cinfo->output_width & 1) { + cb = *inptr1; + cr = *inptr2; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + y = *inptr0; + outptr[RGB_RED] = range_limit[y + cred]; + outptr[RGB_GREEN] = range_limit[y + cgreen]; + outptr[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr[RGB_ALPHA] = _MAXJSAMPLE; +#endif + } +} + + +/* + * Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. + */ + +INLINE +LOCAL(void) +h2v2_merged_upsample_internal(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + _JSAMPARRAY output_buf) +{ + my_merged_upsample_ptr upsample = (my_merged_upsample_ptr)cinfo->upsample; + register int y, cred, cgreen, cblue; + int cb, cr; + register _JSAMPROW outptr0, outptr1; + _JSAMPROW inptr00, inptr01, inptr1, inptr2; + JDIMENSION col; + /* copy these pointers into registers if possible */ + register _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int *Crrtab = upsample->Cr_r_tab; + int *Cbbtab = upsample->Cb_b_tab; + JLONG *Crgtab = upsample->Cr_g_tab; + JLONG *Cbgtab = upsample->Cb_g_tab; + SHIFT_TEMPS + + inptr00 = input_buf[0][in_row_group_ctr * 2]; + inptr01 = input_buf[0][in_row_group_ctr * 2 + 1]; + inptr1 = input_buf[1][in_row_group_ctr]; + inptr2 = input_buf[2][in_row_group_ctr]; + outptr0 = output_buf[0]; + outptr1 = output_buf[1]; + /* Loop for each group of output pixels */ + for (col = cinfo->output_width >> 1; col > 0; col--) { + /* Do the chroma part of the calculation */ + cb = *inptr1++; + cr = *inptr2++; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + /* Fetch 4 Y values and emit 4 pixels */ + y = *inptr00++; + outptr0[RGB_RED] = range_limit[y + cred]; + outptr0[RGB_GREEN] = range_limit[y + cgreen]; + outptr0[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr0[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr0 += RGB_PIXELSIZE; + y = *inptr00++; + outptr0[RGB_RED] = range_limit[y + cred]; + outptr0[RGB_GREEN] = range_limit[y + cgreen]; + outptr0[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr0[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr0 += RGB_PIXELSIZE; + y = *inptr01++; + outptr1[RGB_RED] = range_limit[y + cred]; + outptr1[RGB_GREEN] = range_limit[y + cgreen]; + outptr1[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr1[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr1 += RGB_PIXELSIZE; + y = *inptr01++; + outptr1[RGB_RED] = range_limit[y + cred]; + outptr1[RGB_GREEN] = range_limit[y + cgreen]; + outptr1[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr1[RGB_ALPHA] = _MAXJSAMPLE; +#endif + outptr1 += RGB_PIXELSIZE; + } + /* If image width is odd, do the last output column separately */ + if (cinfo->output_width & 1) { + cb = *inptr1; + cr = *inptr2; + cred = Crrtab[cr]; + cgreen = (int)RIGHT_SHIFT(Cbgtab[cb] + Crgtab[cr], SCALEBITS); + cblue = Cbbtab[cb]; + y = *inptr00; + outptr0[RGB_RED] = range_limit[y + cred]; + outptr0[RGB_GREEN] = range_limit[y + cgreen]; + outptr0[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr0[RGB_ALPHA] = _MAXJSAMPLE; +#endif + y = *inptr01; + outptr1[RGB_RED] = range_limit[y + cred]; + outptr1[RGB_GREEN] = range_limit[y + cgreen]; + outptr1[RGB_BLUE] = range_limit[y + cblue]; +#ifdef RGB_ALPHA + outptr1[RGB_ALPHA] = _MAXJSAMPLE; +#endif + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jdphuff.c b/src/spice2x/external/libjpeg-turbo/jdphuff.c new file mode 100644 index 0000000..bf97333 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdphuff.c @@ -0,0 +1,681 @@ +/* + * jdphuff.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1995-1997, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2015-2016, 2018-2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains Huffman entropy decoding routines for progressive JPEG. + * + * Much of the complexity here has to do with supporting input suspension. + * If the data source module demands suspension, we want to be able to back + * up to the start of the current MCU. To do this, we copy state variables + * into local working storage, and update them back to the permanent + * storage only upon successful completion of an MCU. + * + * NOTE: All referenced figures are from + * Recommendation ITU-T T.81 (1992) | ISO/IEC 10918-1:1994. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdhuff.h" /* Declarations shared with jd*huff.c */ +#include + + +#ifdef D_PROGRESSIVE_SUPPORTED + +/* + * Expanded entropy decoder object for progressive Huffman decoding. + * + * The savable_state subrecord contains fields that change within an MCU, + * but must not be updated permanently until we complete the MCU. + */ + +typedef struct { + unsigned int EOBRUN; /* remaining EOBs in EOBRUN */ + int last_dc_val[MAX_COMPS_IN_SCAN]; /* last DC coef for each component */ +} savable_state; + +typedef struct { + struct jpeg_entropy_decoder pub; /* public fields */ + + /* These fields are loaded into local variables at start of each MCU. + * In case of suspension, we exit WITHOUT updating them. + */ + bitread_perm_state bitstate; /* Bit buffer at start of MCU */ + savable_state saved; /* Other state at start of MCU */ + + /* These fields are NOT loaded into local working state. */ + unsigned int restarts_to_go; /* MCUs left in this restart interval */ + + /* Pointers to derived tables (these workspaces have image lifespan) */ + d_derived_tbl *derived_tbls[NUM_HUFF_TBLS]; + + d_derived_tbl *ac_derived_tbl; /* active table during an AC scan */ +} phuff_entropy_decoder; + +typedef phuff_entropy_decoder *phuff_entropy_ptr; + +/* Forward declarations */ +METHODDEF(boolean) decode_mcu_DC_first(j_decompress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(boolean) decode_mcu_AC_first(j_decompress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(boolean) decode_mcu_DC_refine(j_decompress_ptr cinfo, + JBLOCKROW *MCU_data); +METHODDEF(boolean) decode_mcu_AC_refine(j_decompress_ptr cinfo, + JBLOCKROW *MCU_data); + + +/* + * Initialize for a Huffman-compressed scan. + */ + +METHODDEF(void) +start_pass_phuff_decoder(j_decompress_ptr cinfo) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + boolean is_DC_band, bad; + int ci, coefi, tbl; + d_derived_tbl **pdtbl; + int *coef_bit_ptr, *prev_coef_bit_ptr; + jpeg_component_info *compptr; + + is_DC_band = (cinfo->Ss == 0); + + /* Validate scan parameters */ + bad = FALSE; + if (is_DC_band) { + if (cinfo->Se != 0) + bad = TRUE; + } else { + /* need not check Ss/Se < 0 since they came from unsigned bytes */ + if (cinfo->Ss > cinfo->Se || cinfo->Se >= DCTSIZE2) + bad = TRUE; + /* AC scans may have only one component */ + if (cinfo->comps_in_scan != 1) + bad = TRUE; + } + if (cinfo->Ah != 0) { + /* Successive approximation refinement scan: must have Al = Ah-1. */ + if (cinfo->Al != cinfo->Ah - 1) + bad = TRUE; + } + if (cinfo->Al > 13) /* need not check for < 0 */ + bad = TRUE; + /* Arguably the maximum Al value should be less than 13 for 8-bit precision, + * but the spec doesn't say so, and we try to be liberal about what we + * accept. Note: large Al values could result in out-of-range DC + * coefficients during early scans, leading to bizarre displays due to + * overflows in the IDCT math. But we won't crash. + */ + if (bad) + ERREXIT4(cinfo, JERR_BAD_PROGRESSION, + cinfo->Ss, cinfo->Se, cinfo->Ah, cinfo->Al); + /* Update progression status, and verify that scan order is legal. + * Note that inter-scan inconsistencies are treated as warnings + * not fatal errors ... not clear if this is right way to behave. + */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + int cindex = cinfo->cur_comp_info[ci]->component_index; + coef_bit_ptr = &cinfo->coef_bits[cindex][0]; + prev_coef_bit_ptr = &cinfo->coef_bits[cindex + cinfo->num_components][0]; + if (!is_DC_band && coef_bit_ptr[0] < 0) /* AC without prior DC scan */ + WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, 0); + for (coefi = MIN(cinfo->Ss, 1); coefi <= MAX(cinfo->Se, 9); coefi++) { + if (cinfo->input_scan_number > 1) + prev_coef_bit_ptr[coefi] = coef_bit_ptr[coefi]; + else + prev_coef_bit_ptr[coefi] = 0; + } + for (coefi = cinfo->Ss; coefi <= cinfo->Se; coefi++) { + int expected = (coef_bit_ptr[coefi] < 0) ? 0 : coef_bit_ptr[coefi]; + if (cinfo->Ah != expected) + WARNMS2(cinfo, JWRN_BOGUS_PROGRESSION, cindex, coefi); + coef_bit_ptr[coefi] = cinfo->Al; + } + } + + /* Select MCU decoding routine */ + if (cinfo->Ah == 0) { + if (is_DC_band) + entropy->pub.decode_mcu = decode_mcu_DC_first; + else + entropy->pub.decode_mcu = decode_mcu_AC_first; + } else { + if (is_DC_band) + entropy->pub.decode_mcu = decode_mcu_DC_refine; + else + entropy->pub.decode_mcu = decode_mcu_AC_refine; + } + + for (ci = 0; ci < cinfo->comps_in_scan; ci++) { + compptr = cinfo->cur_comp_info[ci]; + /* Make sure requested tables are present, and compute derived tables. + * We may build same derived table more than once, but it's not expensive. + */ + if (is_DC_band) { + if (cinfo->Ah == 0) { /* DC refinement needs no table */ + tbl = compptr->dc_tbl_no; + pdtbl = (d_derived_tbl **)(entropy->derived_tbls) + tbl; + jpeg_make_d_derived_tbl(cinfo, TRUE, tbl, pdtbl); + } + } else { + tbl = compptr->ac_tbl_no; + pdtbl = (d_derived_tbl **)(entropy->derived_tbls) + tbl; + jpeg_make_d_derived_tbl(cinfo, FALSE, tbl, pdtbl); + /* remember the single active table */ + entropy->ac_derived_tbl = entropy->derived_tbls[tbl]; + } + /* Initialize DC predictions to 0 */ + entropy->saved.last_dc_val[ci] = 0; + } + + /* Initialize bitread state variables */ + entropy->bitstate.bits_left = 0; + entropy->bitstate.get_buffer = 0; /* unnecessary, but keeps Purify quiet */ + entropy->pub.insufficient_data = FALSE; + + /* Initialize private state variables */ + entropy->saved.EOBRUN = 0; + + /* Initialize restart counter */ + entropy->restarts_to_go = cinfo->restart_interval; +} + + +/* + * Figure F.12: extend sign bit. + * On some machines, a shift and add will be faster than a table lookup. + */ + +#define AVOID_TABLES +#ifdef AVOID_TABLES + +#define NEG_1 ((unsigned)-1) +#define HUFF_EXTEND(x, s) \ + ((x) < (1 << ((s) - 1)) ? (x) + (((NEG_1) << (s)) + 1) : (x)) + +#else + +#define HUFF_EXTEND(x, s) \ + ((x) < extend_test[s] ? (x) + extend_offset[s] : (x)) + +static const int extend_test[16] = { /* entry n is 2**(n-1) */ + 0, 0x0001, 0x0002, 0x0004, 0x0008, 0x0010, 0x0020, 0x0040, 0x0080, + 0x0100, 0x0200, 0x0400, 0x0800, 0x1000, 0x2000, 0x4000 +}; + +static const int extend_offset[16] = { /* entry n is (-1 << n) + 1 */ + 0, ((-1) << 1) + 1, ((-1) << 2) + 1, ((-1) << 3) + 1, ((-1) << 4) + 1, + ((-1) << 5) + 1, ((-1) << 6) + 1, ((-1) << 7) + 1, ((-1) << 8) + 1, + ((-1) << 9) + 1, ((-1) << 10) + 1, ((-1) << 11) + 1, ((-1) << 12) + 1, + ((-1) << 13) + 1, ((-1) << 14) + 1, ((-1) << 15) + 1 +}; + +#endif /* AVOID_TABLES */ + + +/* + * Check for a restart marker & resynchronize decoder. + * Returns FALSE if must suspend. + */ + +LOCAL(boolean) +process_restart(j_decompress_ptr cinfo) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + int ci; + + /* Throw away any unused bits remaining in bit buffer; */ + /* include any full bytes in next_marker's count of discarded bytes */ + cinfo->marker->discarded_bytes += entropy->bitstate.bits_left / 8; + entropy->bitstate.bits_left = 0; + + /* Advance past the RSTn marker */ + if (!(*cinfo->marker->read_restart_marker) (cinfo)) + return FALSE; + + /* Re-initialize DC predictions to 0 */ + for (ci = 0; ci < cinfo->comps_in_scan; ci++) + entropy->saved.last_dc_val[ci] = 0; + /* Re-init EOB run count, too */ + entropy->saved.EOBRUN = 0; + + /* Reset restart counter */ + entropy->restarts_to_go = cinfo->restart_interval; + + /* Reset out-of-data flag, unless read_restart_marker left us smack up + * against a marker. In that case we will end up treating the next data + * segment as empty, and we can avoid producing bogus output pixels by + * leaving the flag set. + */ + if (cinfo->unread_marker == 0) + entropy->pub.insufficient_data = FALSE; + + return TRUE; +} + + +/* + * Huffman MCU decoding. + * Each of these routines decodes and returns one MCU's worth of + * Huffman-compressed coefficients. + * The coefficients are reordered from zigzag order into natural array order, + * but are not dequantized. + * + * The i'th block of the MCU is stored into the block pointed to by + * MCU_data[i]. WE ASSUME THIS AREA IS INITIALLY ZEROED BY THE CALLER. + * + * We return FALSE if data source requested suspension. In that case no + * changes have been made to permanent state. (Exception: some output + * coefficients may already have been assigned. This is harmless for + * spectral selection, since we'll just re-assign them on the next call. + * Successive approximation AC refinement has to be more careful, however.) + */ + +/* + * MCU decoding for DC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +METHODDEF(boolean) +decode_mcu_DC_first(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + int Al = cinfo->Al; + register int s, r; + int blkn, ci; + JBLOCKROW block; + BITREAD_STATE_VARS; + savable_state state; + d_derived_tbl *tbl; + jpeg_component_info *compptr; + + /* Process restart marker if needed; may have to suspend */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + if (!process_restart(cinfo)) + return FALSE; + } + + /* If we've run out of data, just leave the MCU set to zeroes. + * This way, we return uniform gray for the remainder of the segment. + */ + if (!entropy->pub.insufficient_data) { + + /* Load up working state */ + BITREAD_LOAD_STATE(cinfo, entropy->bitstate); + state = entropy->saved; + + /* Outer loop handles each block in the MCU */ + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data[blkn]; + ci = cinfo->MCU_membership[blkn]; + compptr = cinfo->cur_comp_info[ci]; + tbl = entropy->derived_tbls[compptr->dc_tbl_no]; + + /* Decode a single block's worth of coefficients */ + + /* Section F.2.2.1: decode the DC coefficient difference */ + HUFF_DECODE(s, br_state, tbl, return FALSE, label1); + if (s) { + CHECK_BIT_BUFFER(br_state, s, return FALSE); + r = GET_BITS(s); + s = HUFF_EXTEND(r, s); + } + + /* Convert DC difference to actual value, update last_dc_val */ + if ((state.last_dc_val[ci] >= 0 && + s > INT_MAX - state.last_dc_val[ci]) || + (state.last_dc_val[ci] < 0 && s < INT_MIN - state.last_dc_val[ci])) + ERREXIT(cinfo, JERR_BAD_DCT_COEF); + s += state.last_dc_val[ci]; + state.last_dc_val[ci] = s; + /* Scale and output the coefficient (assumes jpeg_natural_order[0]=0) */ + (*block)[0] = (JCOEF)LEFT_SHIFT(s, Al); + } + + /* Completed MCU, so update state */ + BITREAD_SAVE_STATE(cinfo, entropy->bitstate); + entropy->saved = state; + } + + /* Account for restart interval (no-op if not using restarts) */ + if (cinfo->restart_interval) + entropy->restarts_to_go--; + + return TRUE; +} + + +/* + * MCU decoding for AC initial scan (either spectral selection, + * or first pass of successive approximation). + */ + +METHODDEF(boolean) +decode_mcu_AC_first(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + int Se = cinfo->Se; + int Al = cinfo->Al; + register int s, k, r; + unsigned int EOBRUN; + JBLOCKROW block; + BITREAD_STATE_VARS; + d_derived_tbl *tbl; + + /* Process restart marker if needed; may have to suspend */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + if (!process_restart(cinfo)) + return FALSE; + } + + /* If we've run out of data, just leave the MCU set to zeroes. + * This way, we return uniform gray for the remainder of the segment. + */ + if (!entropy->pub.insufficient_data) { + + /* Load up working state. + * We can avoid loading/saving bitread state if in an EOB run. + */ + EOBRUN = entropy->saved.EOBRUN; /* only part of saved state we need */ + + /* There is always only one block per MCU */ + + if (EOBRUN > 0) /* if it's a band of zeroes... */ + EOBRUN--; /* ...process it now (we do nothing) */ + else { + BITREAD_LOAD_STATE(cinfo, entropy->bitstate); + block = MCU_data[0]; + tbl = entropy->ac_derived_tbl; + + for (k = cinfo->Ss; k <= Se; k++) { + HUFF_DECODE(s, br_state, tbl, return FALSE, label2); + r = s >> 4; + s &= 15; + if (s) { + k += r; + CHECK_BIT_BUFFER(br_state, s, return FALSE); + r = GET_BITS(s); + s = HUFF_EXTEND(r, s); + /* Scale and output coefficient in natural (dezigzagged) order */ + (*block)[jpeg_natural_order[k]] = (JCOEF)LEFT_SHIFT(s, Al); + } else { + if (r == 15) { /* ZRL */ + k += 15; /* skip 15 zeroes in band */ + } else { /* EOBr, run length is 2^r + appended bits */ + EOBRUN = 1 << r; + if (r) { /* EOBr, r > 0 */ + CHECK_BIT_BUFFER(br_state, r, return FALSE); + r = GET_BITS(r); + EOBRUN += r; + } + EOBRUN--; /* this band is processed at this moment */ + break; /* force end-of-band */ + } + } + } + + BITREAD_SAVE_STATE(cinfo, entropy->bitstate); + } + + /* Completed MCU, so update state */ + entropy->saved.EOBRUN = EOBRUN; /* only part of saved state we need */ + } + + /* Account for restart interval (no-op if not using restarts) */ + if (cinfo->restart_interval) + entropy->restarts_to_go--; + + return TRUE; +} + + +/* + * MCU decoding for DC successive approximation refinement scan. + * Note: we assume such scans can be multi-component, although the spec + * is not very clear on the point. + */ + +METHODDEF(boolean) +decode_mcu_DC_refine(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + int p1 = 1 << cinfo->Al; /* 1 in the bit position being coded */ + int blkn; + JBLOCKROW block; + BITREAD_STATE_VARS; + + /* Process restart marker if needed; may have to suspend */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + if (!process_restart(cinfo)) + return FALSE; + } + + /* Not worth the cycles to check insufficient_data here, + * since we will not change the data anyway if we read zeroes. + */ + + /* Load up working state */ + BITREAD_LOAD_STATE(cinfo, entropy->bitstate); + + /* Outer loop handles each block in the MCU */ + + for (blkn = 0; blkn < cinfo->blocks_in_MCU; blkn++) { + block = MCU_data[blkn]; + + /* Encoded data is simply the next bit of the two's-complement DC value */ + CHECK_BIT_BUFFER(br_state, 1, return FALSE); + if (GET_BITS(1)) + (*block)[0] |= p1; + /* Note: since we use |=, repeating the assignment later is safe */ + } + + /* Completed MCU, so update state */ + BITREAD_SAVE_STATE(cinfo, entropy->bitstate); + + /* Account for restart interval (no-op if not using restarts) */ + if (cinfo->restart_interval) + entropy->restarts_to_go--; + + return TRUE; +} + + +/* + * MCU decoding for AC successive approximation refinement scan. + */ + +METHODDEF(boolean) +decode_mcu_AC_refine(j_decompress_ptr cinfo, JBLOCKROW *MCU_data) +{ + phuff_entropy_ptr entropy = (phuff_entropy_ptr)cinfo->entropy; + int Se = cinfo->Se; + int p1 = 1 << cinfo->Al; /* 1 in the bit position being coded */ + int m1 = (NEG_1) << cinfo->Al; /* -1 in the bit position being coded */ + register int s, k, r; + unsigned int EOBRUN; + JBLOCKROW block; + JCOEFPTR thiscoef; + BITREAD_STATE_VARS; + d_derived_tbl *tbl; + int num_newnz; + int newnz_pos[DCTSIZE2]; + + /* Process restart marker if needed; may have to suspend */ + if (cinfo->restart_interval) { + if (entropy->restarts_to_go == 0) + if (!process_restart(cinfo)) + return FALSE; + } + + /* If we've run out of data, don't modify the MCU. + */ + if (!entropy->pub.insufficient_data) { + + /* Load up working state */ + BITREAD_LOAD_STATE(cinfo, entropy->bitstate); + EOBRUN = entropy->saved.EOBRUN; /* only part of saved state we need */ + + /* There is always only one block per MCU */ + block = MCU_data[0]; + tbl = entropy->ac_derived_tbl; + + /* If we are forced to suspend, we must undo the assignments to any newly + * nonzero coefficients in the block, because otherwise we'd get confused + * next time about which coefficients were already nonzero. + * But we need not undo addition of bits to already-nonzero coefficients; + * instead, we can test the current bit to see if we already did it. + */ + num_newnz = 0; + + /* initialize coefficient loop counter to start of band */ + k = cinfo->Ss; + + if (EOBRUN == 0) { + for (; k <= Se; k++) { + HUFF_DECODE(s, br_state, tbl, goto undoit, label3); + r = s >> 4; + s &= 15; + if (s) { + if (s != 1) /* size of new coef should always be 1 */ + WARNMS(cinfo, JWRN_HUFF_BAD_CODE); + CHECK_BIT_BUFFER(br_state, 1, goto undoit); + if (GET_BITS(1)) + s = p1; /* newly nonzero coef is positive */ + else + s = m1; /* newly nonzero coef is negative */ + } else { + if (r != 15) { + EOBRUN = 1 << r; /* EOBr, run length is 2^r + appended bits */ + if (r) { + CHECK_BIT_BUFFER(br_state, r, goto undoit); + r = GET_BITS(r); + EOBRUN += r; + } + break; /* rest of block is handled by EOB logic */ + } + /* note s = 0 for processing ZRL */ + } + /* Advance over already-nonzero coefs and r still-zero coefs, + * appending correction bits to the nonzeroes. A correction bit is 1 + * if the absolute value of the coefficient must be increased. + */ + do { + thiscoef = *block + jpeg_natural_order[k]; + if (*thiscoef != 0) { + CHECK_BIT_BUFFER(br_state, 1, goto undoit); + if (GET_BITS(1)) { + if ((*thiscoef & p1) == 0) { /* do nothing if already set it */ + if (*thiscoef >= 0) + *thiscoef += (JCOEF)p1; + else + *thiscoef += (JCOEF)m1; + } + } + } else { + if (--r < 0) + break; /* reached target zero coefficient */ + } + k++; + } while (k <= Se); + if (s) { + int pos = jpeg_natural_order[k]; + /* Output newly nonzero coefficient */ + (*block)[pos] = (JCOEF)s; + /* Remember its position in case we have to suspend */ + newnz_pos[num_newnz++] = pos; + } + } + } + + if (EOBRUN > 0) { + /* Scan any remaining coefficient positions after the end-of-band + * (the last newly nonzero coefficient, if any). Append a correction + * bit to each already-nonzero coefficient. A correction bit is 1 + * if the absolute value of the coefficient must be increased. + */ + for (; k <= Se; k++) { + thiscoef = *block + jpeg_natural_order[k]; + if (*thiscoef != 0) { + CHECK_BIT_BUFFER(br_state, 1, goto undoit); + if (GET_BITS(1)) { + if ((*thiscoef & p1) == 0) { /* do nothing if already changed it */ + if (*thiscoef >= 0) + *thiscoef += (JCOEF)p1; + else + *thiscoef += (JCOEF)m1; + } + } + } + } + /* Count one block completed in EOB run */ + EOBRUN--; + } + + /* Completed MCU, so update state */ + BITREAD_SAVE_STATE(cinfo, entropy->bitstate); + entropy->saved.EOBRUN = EOBRUN; /* only part of saved state we need */ + } + + /* Account for restart interval (no-op if not using restarts) */ + if (cinfo->restart_interval) + entropy->restarts_to_go--; + + return TRUE; + +undoit: + /* Re-zero any output coefficients that we made newly nonzero */ + while (num_newnz > 0) + (*block)[newnz_pos[--num_newnz]] = 0; + + return FALSE; +} + + +/* + * Module initialization routine for progressive Huffman entropy decoding. + */ + +GLOBAL(void) +jinit_phuff_decoder(j_decompress_ptr cinfo) +{ + phuff_entropy_ptr entropy; + int *coef_bit_ptr; + int ci, i; + + entropy = (phuff_entropy_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(phuff_entropy_decoder)); + cinfo->entropy = (struct jpeg_entropy_decoder *)entropy; + entropy->pub.start_pass = start_pass_phuff_decoder; + + /* Mark derived tables unallocated */ + for (i = 0; i < NUM_HUFF_TBLS; i++) { + entropy->derived_tbls[i] = NULL; + } + + /* Create progression status table */ + cinfo->coef_bits = (int (*)[DCTSIZE2]) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + cinfo->num_components * 2 * DCTSIZE2 * + sizeof(int)); + coef_bit_ptr = &cinfo->coef_bits[0][0]; + for (ci = 0; ci < cinfo->num_components; ci++) + for (i = 0; i < DCTSIZE2; i++) + *coef_bit_ptr++ = -1; +} + +#endif /* D_PROGRESSIVE_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jdpostct.c b/src/spice2x/external/libjpeg-turbo/jdpostct.c new file mode 100644 index 0000000..d38495f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdpostct.c @@ -0,0 +1,315 @@ +/* + * jdpostct.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022-2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the decompression postprocessing controller. + * This controller manages the upsampling, color conversion, and color + * quantization/reduction steps; specifically, it controls the buffering + * between upsample/color conversion and color quantization/reduction. + * + * If no color quantization/reduction is required, then this module has no + * work to do, and it just hands off to the upsample/color conversion code. + * An integrated upsample/convert/quantize process would replace this module + * entirely. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) + +/* Private buffer controller object */ + +typedef struct { + struct jpeg_d_post_controller pub; /* public fields */ + + /* Color quantization source buffer: this holds output data from + * the upsample/color conversion step to be passed to the quantizer. + * For two-pass color quantization, we need a full-image buffer; + * for one-pass operation, a strip buffer is sufficient. + */ + jvirt_sarray_ptr whole_image; /* virtual array, or NULL if one-pass */ + _JSAMPARRAY buffer; /* strip buffer, or current strip of virtual */ + JDIMENSION strip_height; /* buffer size in rows */ + /* for two-pass mode only: */ + JDIMENSION starting_row; /* row # of first row in current strip */ + JDIMENSION next_row; /* index of next row to fill/empty in strip */ +} my_post_controller; + +typedef my_post_controller *my_post_ptr; + + +/* Forward declarations */ +#if BITS_IN_JSAMPLE != 16 +METHODDEF(void) post_process_1pass(j_decompress_ptr cinfo, + _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, + _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +#endif +#if defined(QUANT_2PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 +METHODDEF(void) post_process_prepass(j_decompress_ptr cinfo, + _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, + _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +METHODDEF(void) post_process_2pass(j_decompress_ptr cinfo, + _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, + _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +#endif + + +/* + * Initialize for a processing pass. + */ + +METHODDEF(void) +start_pass_dpost(j_decompress_ptr cinfo, J_BUF_MODE pass_mode) +{ + my_post_ptr post = (my_post_ptr)cinfo->post; + + switch (pass_mode) { + case JBUF_PASS_THRU: +#if BITS_IN_JSAMPLE != 16 + if (cinfo->quantize_colors) { + /* Single-pass processing with color quantization. */ + post->pub._post_process_data = post_process_1pass; + /* We could be doing buffered-image output before starting a 2-pass + * color quantization; in that case, jinit_d_post_controller did not + * allocate a strip buffer. Use the virtual-array buffer as workspace. + */ + if (post->buffer == NULL) { + post->buffer = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray) + ((j_common_ptr)cinfo, post->whole_image, + (JDIMENSION)0, post->strip_height, TRUE); + } + } else +#endif + { + /* For single-pass processing without color quantization, + * I have no work to do; just call the upsampler directly. + */ + post->pub._post_process_data = cinfo->upsample->_upsample; + } + break; +#if defined(QUANT_2PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 + case JBUF_SAVE_AND_PASS: + /* First pass of 2-pass quantization */ + if (post->whole_image == NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + post->pub._post_process_data = post_process_prepass; + break; + case JBUF_CRANK_DEST: + /* Second pass of 2-pass quantization */ + if (post->whole_image == NULL) + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + post->pub._post_process_data = post_process_2pass; + break; +#endif /* defined(QUANT_2PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 */ + default: + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); + break; + } + post->starting_row = post->next_row = 0; +} + + +/* + * Process some data in the one-pass (strip buffer) case. + * This is used for color precision reduction as well as one-pass quantization. + */ + +#if BITS_IN_JSAMPLE != 16 + +METHODDEF(void) +post_process_1pass(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +{ + my_post_ptr post = (my_post_ptr)cinfo->post; + JDIMENSION num_rows, max_rows; + + /* Fill the buffer, but not more than what we can dump out in one go. */ + /* Note we rely on the upsampler to detect bottom of image. */ + max_rows = out_rows_avail - *out_row_ctr; + if (max_rows > post->strip_height) + max_rows = post->strip_height; + num_rows = 0; + (*cinfo->upsample->_upsample) (cinfo, input_buf, in_row_group_ctr, + in_row_groups_avail, post->buffer, &num_rows, + max_rows); + /* Quantize and emit data. */ + (*cinfo->cquantize->_color_quantize) (cinfo, post->buffer, + output_buf + *out_row_ctr, + (int)num_rows); + *out_row_ctr += num_rows; +} + +#endif + + +#if defined(QUANT_2PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 + +/* + * Process some data in the first pass of 2-pass quantization. + */ + +METHODDEF(void) +post_process_prepass(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +{ + my_post_ptr post = (my_post_ptr)cinfo->post; + JDIMENSION old_next_row, num_rows; + + /* Reposition virtual buffer if at start of strip. */ + if (post->next_row == 0) { + post->buffer = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray) + ((j_common_ptr)cinfo, post->whole_image, + post->starting_row, post->strip_height, TRUE); + } + + /* Upsample some data (up to a strip height's worth). */ + old_next_row = post->next_row; + (*cinfo->upsample->_upsample) (cinfo, input_buf, in_row_group_ctr, + in_row_groups_avail, post->buffer, + &post->next_row, post->strip_height); + + /* Allow quantizer to scan new data. No data is emitted, */ + /* but we advance out_row_ctr so outer loop can tell when we're done. */ + if (post->next_row > old_next_row) { + num_rows = post->next_row - old_next_row; + (*cinfo->cquantize->_color_quantize) (cinfo, post->buffer + old_next_row, + (_JSAMPARRAY)NULL, (int)num_rows); + *out_row_ctr += num_rows; + } + + /* Advance if we filled the strip. */ + if (post->next_row >= post->strip_height) { + post->starting_row += post->strip_height; + post->next_row = 0; + } +} + + +/* + * Process some data in the second pass of 2-pass quantization. + */ + +METHODDEF(void) +post_process_2pass(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, _JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail) +{ + my_post_ptr post = (my_post_ptr)cinfo->post; + JDIMENSION num_rows, max_rows; + + /* Reposition virtual buffer if at start of strip. */ + if (post->next_row == 0) { + post->buffer = (_JSAMPARRAY)(*cinfo->mem->access_virt_sarray) + ((j_common_ptr)cinfo, post->whole_image, + post->starting_row, post->strip_height, FALSE); + } + + /* Determine number of rows to emit. */ + num_rows = post->strip_height - post->next_row; /* available in strip */ + max_rows = out_rows_avail - *out_row_ctr; /* available in output area */ + if (num_rows > max_rows) + num_rows = max_rows; + /* We have to check bottom of image here, can't depend on upsampler. */ + max_rows = cinfo->output_height - post->starting_row; + if (num_rows > max_rows) + num_rows = max_rows; + + /* Quantize and emit data. */ + (*cinfo->cquantize->_color_quantize) (cinfo, post->buffer + post->next_row, + output_buf + *out_row_ctr, + (int)num_rows); + *out_row_ctr += num_rows; + + /* Advance if we filled the strip. */ + post->next_row += num_rows; + if (post->next_row >= post->strip_height) { + post->starting_row += post->strip_height; + post->next_row = 0; + } +} + +#endif /* defined(QUANT_2PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 */ + + +/* + * Initialize postprocessing controller. + */ + +GLOBAL(void) +_jinit_d_post_controller(j_decompress_ptr cinfo, boolean need_full_buffer) +{ + my_post_ptr post; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + post = (my_post_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_post_controller)); + cinfo->post = (struct jpeg_d_post_controller *)post; + post->pub.start_pass = start_pass_dpost; + post->whole_image = NULL; /* flag for no virtual arrays */ + post->buffer = NULL; /* flag for no strip buffer */ + + /* Create the quantization buffer, if needed */ + if (cinfo->quantize_colors) { +#if BITS_IN_JSAMPLE != 16 + /* The buffer strip height is max_v_samp_factor, which is typically + * an efficient number of rows for upsampling to return. + * (In the presence of output rescaling, we might want to be smarter?) + */ + post->strip_height = (JDIMENSION)cinfo->max_v_samp_factor; + if (need_full_buffer) { + /* Two-pass color quantization: need full-image storage. */ + /* We round up the number of rows to a multiple of the strip height. */ +#ifdef QUANT_2PASS_SUPPORTED + post->whole_image = (*cinfo->mem->request_virt_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, FALSE, + cinfo->output_width * cinfo->out_color_components, + (JDIMENSION)jround_up((long)cinfo->output_height, + (long)post->strip_height), + post->strip_height); +#else + ERREXIT(cinfo, JERR_BAD_BUFFER_MODE); +#endif /* QUANT_2PASS_SUPPORTED */ + } else { + /* One-pass color quantization: just make a strip buffer. */ + post->buffer = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + cinfo->output_width * cinfo->out_color_components, + post->strip_height); + } +#else + ERREXIT(cinfo, JERR_NOTIMPL); +#endif + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jdsample.c b/src/spice2x/external/libjpeg-turbo/jdsample.c new file mode 100644 index 0000000..cc8015c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdsample.c @@ -0,0 +1,540 @@ +/* + * jdsample.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2010, 2015-2016, 2022, D. R. Commander. + * Copyright (C) 2014, MIPS Technologies, Inc., California. + * Copyright (C) 2015, Google, Inc. + * Copyright (C) 2019-2020, Arm Limited. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains upsampling routines. + * + * Upsampling input data is counted in "row groups". A row group + * is defined to be (v_samp_factor * DCT_scaled_size / min_DCT_scaled_size) + * sample rows of each component. Upsampling will normally produce + * max_v_samp_factor pixel rows from each row group (but this could vary + * if the upsampler is applying a scale factor of its own). + * + * An excellent reference for image resampling is + * Digital Image Warping, George Wolberg, 1990. + * Pub. by IEEE Computer Society Press, Los Alamitos, CA. ISBN 0-8186-8944-7. + */ + +#include "jinclude.h" +#include "jdsample.h" +#include "jsimd.h" +#include "jpegapicomp.h" + + + +#if BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) + +/* + * Initialize for an upsampling pass. + */ + +METHODDEF(void) +start_pass_upsample(j_decompress_ptr cinfo) +{ + my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample; + + /* Mark the conversion buffer empty */ + upsample->next_row_out = cinfo->max_v_samp_factor; + /* Initialize total-height counter for detecting bottom of image */ + upsample->rows_to_go = cinfo->output_height; +} + + +/* + * Control routine to do upsampling (and color conversion). + * + * In this version we upsample each component independently. + * We upsample one row group into the conversion buffer, then apply + * color conversion a row at a time. + */ + +METHODDEF(void) +sep_upsample(j_decompress_ptr cinfo, _JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, JDIMENSION in_row_groups_avail, + _JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail) +{ + my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample; + int ci; + jpeg_component_info *compptr; + JDIMENSION num_rows; + + /* Fill the conversion buffer, if it's empty */ + if (upsample->next_row_out >= cinfo->max_v_samp_factor) { + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Invoke per-component upsample method. Notice we pass a POINTER + * to color_buf[ci], so that fullsize_upsample can change it. + */ + (*upsample->methods[ci]) (cinfo, compptr, + input_buf[ci] + (*in_row_group_ctr * upsample->rowgroup_height[ci]), + upsample->color_buf + ci); + } + upsample->next_row_out = 0; + } + + /* Color-convert and emit rows */ + + /* How many we have in the buffer: */ + num_rows = (JDIMENSION)(cinfo->max_v_samp_factor - upsample->next_row_out); + /* Not more than the distance to the end of the image. Need this test + * in case the image height is not a multiple of max_v_samp_factor: + */ + if (num_rows > upsample->rows_to_go) + num_rows = upsample->rows_to_go; + /* And not more than what the client can accept: */ + out_rows_avail -= *out_row_ctr; + if (num_rows > out_rows_avail) + num_rows = out_rows_avail; + + (*cinfo->cconvert->_color_convert) (cinfo, upsample->color_buf, + (JDIMENSION)upsample->next_row_out, + output_buf + *out_row_ctr, + (int)num_rows); + + /* Adjust counts */ + *out_row_ctr += num_rows; + upsample->rows_to_go -= num_rows; + upsample->next_row_out += num_rows; + /* When the buffer is emptied, declare this input row group consumed */ + if (upsample->next_row_out >= cinfo->max_v_samp_factor) + (*in_row_group_ctr)++; +} + + +/* + * These are the routines invoked by sep_upsample to upsample pixel values + * of a single component. One row group is processed per call. + */ + + +/* + * For full-size components, we just make color_buf[ci] point at the + * input buffer, and thus avoid copying any data. Note that this is + * safe only because sep_upsample doesn't declare the input row group + * "consumed" until we are done color converting and emitting it. + */ + +METHODDEF(void) +fullsize_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + *output_data_ptr = input_data; +} + + +/* + * This is a no-op version used for "uninteresting" components. + * These components will not be referenced by color conversion. + */ + +METHODDEF(void) +noop_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + *output_data_ptr = NULL; /* safety check */ +} + + +/* + * This version handles any integral sampling ratios. + * This is not used for typical JPEG files, so it need not be fast. + * Nor, for that matter, is it particularly accurate: the algorithm is + * simple replication of the input pixel onto the corresponding output + * pixels. The hi-falutin sampling literature refers to this as a + * "box filter". A box filter tends to introduce visible artifacts, + * so if you are actually going to use 3:1 or 4:1 sampling ratios + * you would be well advised to improve this code. + */ + +METHODDEF(void) +int_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + my_upsample_ptr upsample = (my_upsample_ptr)cinfo->upsample; + _JSAMPARRAY output_data = *output_data_ptr; + register _JSAMPROW inptr, outptr; + register _JSAMPLE invalue; + register int h; + _JSAMPROW outend; + int h_expand, v_expand; + int inrow, outrow; + + h_expand = upsample->h_expand[compptr->component_index]; + v_expand = upsample->v_expand[compptr->component_index]; + + inrow = outrow = 0; + while (outrow < cinfo->max_v_samp_factor) { + /* Generate one output row with proper horizontal expansion */ + inptr = input_data[inrow]; + outptr = output_data[outrow]; + outend = outptr + cinfo->output_width; + while (outptr < outend) { + invalue = *inptr++; + for (h = h_expand; h > 0; h--) { + *outptr++ = invalue; + } + } + /* Generate any additional output rows by duplicating the first one */ + if (v_expand > 1) { + _jcopy_sample_rows(output_data, outrow, output_data, outrow + 1, + v_expand - 1, cinfo->output_width); + } + inrow++; + outrow += v_expand; + } +} + + +/* + * Fast processing for the common case of 2:1 horizontal and 1:1 vertical. + * It's still a box filter. + */ + +METHODDEF(void) +h2v1_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + _JSAMPARRAY output_data = *output_data_ptr; + register _JSAMPROW inptr, outptr; + register _JSAMPLE invalue; + _JSAMPROW outend; + int inrow; + + for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) { + inptr = input_data[inrow]; + outptr = output_data[inrow]; + outend = outptr + cinfo->output_width; + while (outptr < outend) { + invalue = *inptr++; + *outptr++ = invalue; + *outptr++ = invalue; + } + } +} + + +/* + * Fast processing for the common case of 2:1 horizontal and 2:1 vertical. + * It's still a box filter. + */ + +METHODDEF(void) +h2v2_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + _JSAMPARRAY output_data = *output_data_ptr; + register _JSAMPROW inptr, outptr; + register _JSAMPLE invalue; + _JSAMPROW outend; + int inrow, outrow; + + inrow = outrow = 0; + while (outrow < cinfo->max_v_samp_factor) { + inptr = input_data[inrow]; + outptr = output_data[outrow]; + outend = outptr + cinfo->output_width; + while (outptr < outend) { + invalue = *inptr++; + *outptr++ = invalue; + *outptr++ = invalue; + } + _jcopy_sample_rows(output_data, outrow, output_data, outrow + 1, 1, + cinfo->output_width); + inrow++; + outrow += 2; + } +} + + +/* + * Fancy processing for the common case of 2:1 horizontal and 1:1 vertical. + * + * The upsampling algorithm is linear interpolation between pixel centers, + * also known as a "triangle filter". This is a good compromise between + * speed and visual quality. The centers of the output pixels are 1/4 and 3/4 + * of the way between input pixel centers. + * + * A note about the "bias" calculations: when rounding fractional values to + * integer, we do not want to always round 0.5 up to the next integer. + * If we did that, we'd introduce a noticeable bias towards larger values. + * Instead, this code is arranged so that 0.5 will be rounded up or down at + * alternate pixel locations (a simple ordered dither pattern). + */ + +METHODDEF(void) +h2v1_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + _JSAMPARRAY output_data = *output_data_ptr; + register _JSAMPROW inptr, outptr; + register int invalue; + register JDIMENSION colctr; + int inrow; + + for (inrow = 0; inrow < cinfo->max_v_samp_factor; inrow++) { + inptr = input_data[inrow]; + outptr = output_data[inrow]; + /* Special case for first column */ + invalue = *inptr++; + *outptr++ = (_JSAMPLE)invalue; + *outptr++ = (_JSAMPLE)((invalue * 3 + inptr[0] + 2) >> 2); + + for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) { + /* General case: 3/4 * nearer pixel + 1/4 * further pixel */ + invalue = (*inptr++) * 3; + *outptr++ = (_JSAMPLE)((invalue + inptr[-2] + 1) >> 2); + *outptr++ = (_JSAMPLE)((invalue + inptr[0] + 2) >> 2); + } + + /* Special case for last column */ + invalue = *inptr; + *outptr++ = (_JSAMPLE)((invalue * 3 + inptr[-1] + 1) >> 2); + *outptr++ = (_JSAMPLE)invalue; + } +} + + +/* + * Fancy processing for 1:1 horizontal and 2:1 vertical (4:4:0 subsampling). + * + * This is a less common case, but it can be encountered when losslessly + * rotating/transposing a JPEG file that uses 4:2:2 chroma subsampling. + */ + +METHODDEF(void) +h1v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + _JSAMPARRAY output_data = *output_data_ptr; + _JSAMPROW inptr0, inptr1, outptr; +#if BITS_IN_JSAMPLE == 8 + int thiscolsum, bias; +#else + JLONG thiscolsum, bias; +#endif + JDIMENSION colctr; + int inrow, outrow, v; + + inrow = outrow = 0; + while (outrow < cinfo->max_v_samp_factor) { + for (v = 0; v < 2; v++) { + /* inptr0 points to nearest input row, inptr1 points to next nearest */ + inptr0 = input_data[inrow]; + if (v == 0) { /* next nearest is row above */ + inptr1 = input_data[inrow - 1]; + bias = 1; + } else { /* next nearest is row below */ + inptr1 = input_data[inrow + 1]; + bias = 2; + } + outptr = output_data[outrow++]; + + for (colctr = 0; colctr < compptr->downsampled_width; colctr++) { + thiscolsum = (*inptr0++) * 3 + (*inptr1++); + *outptr++ = (_JSAMPLE)((thiscolsum + bias) >> 2); + } + } + inrow++; + } +} + + +/* + * Fancy processing for the common case of 2:1 horizontal and 2:1 vertical. + * Again a triangle filter; see comments for h2v1 case, above. + * + * It is OK for us to reference the adjacent input rows because we demanded + * context from the main buffer controller (see initialization code). + */ + +METHODDEF(void) +h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + _JSAMPARRAY input_data, _JSAMPARRAY *output_data_ptr) +{ + _JSAMPARRAY output_data = *output_data_ptr; + register _JSAMPROW inptr0, inptr1, outptr; +#if BITS_IN_JSAMPLE == 8 + register int thiscolsum, lastcolsum, nextcolsum; +#else + register JLONG thiscolsum, lastcolsum, nextcolsum; +#endif + register JDIMENSION colctr; + int inrow, outrow, v; + + inrow = outrow = 0; + while (outrow < cinfo->max_v_samp_factor) { + for (v = 0; v < 2; v++) { + /* inptr0 points to nearest input row, inptr1 points to next nearest */ + inptr0 = input_data[inrow]; + if (v == 0) /* next nearest is row above */ + inptr1 = input_data[inrow - 1]; + else /* next nearest is row below */ + inptr1 = input_data[inrow + 1]; + outptr = output_data[outrow++]; + + /* Special case for first column */ + thiscolsum = (*inptr0++) * 3 + (*inptr1++); + nextcolsum = (*inptr0++) * 3 + (*inptr1++); + *outptr++ = (_JSAMPLE)((thiscolsum * 4 + 8) >> 4); + *outptr++ = (_JSAMPLE)((thiscolsum * 3 + nextcolsum + 7) >> 4); + lastcolsum = thiscolsum; thiscolsum = nextcolsum; + + for (colctr = compptr->downsampled_width - 2; colctr > 0; colctr--) { + /* General case: 3/4 * nearer pixel + 1/4 * further pixel in each */ + /* dimension, thus 9/16, 3/16, 3/16, 1/16 overall */ + nextcolsum = (*inptr0++) * 3 + (*inptr1++); + *outptr++ = (_JSAMPLE)((thiscolsum * 3 + lastcolsum + 8) >> 4); + *outptr++ = (_JSAMPLE)((thiscolsum * 3 + nextcolsum + 7) >> 4); + lastcolsum = thiscolsum; thiscolsum = nextcolsum; + } + + /* Special case for last column */ + *outptr++ = (_JSAMPLE)((thiscolsum * 3 + lastcolsum + 8) >> 4); + *outptr++ = (_JSAMPLE)((thiscolsum * 4 + 7) >> 4); + } + inrow++; + } +} + + +/* + * Module initialization routine for upsampling. + */ + +GLOBAL(void) +_jinit_upsampler(j_decompress_ptr cinfo) +{ + my_upsample_ptr upsample; + int ci; + jpeg_component_info *compptr; + boolean need_buffer, do_fancy; + int h_in_group, v_in_group, h_out_group, v_out_group; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + if (!cinfo->master->jinit_upsampler_no_alloc) { + upsample = (my_upsample_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_upsampler)); + cinfo->upsample = (struct jpeg_upsampler *)upsample; + upsample->pub.start_pass = start_pass_upsample; + upsample->pub._upsample = sep_upsample; + upsample->pub.need_context_rows = FALSE; /* until we find out differently */ + } else + upsample = (my_upsample_ptr)cinfo->upsample; + + if (cinfo->CCIR601_sampling) /* this isn't supported */ + ERREXIT(cinfo, JERR_CCIR601_NOTIMPL); + + /* jdmainct.c doesn't support context rows when min_DCT_scaled_size = 1, + * so don't ask for it. + */ + do_fancy = cinfo->do_fancy_upsampling && cinfo->_min_DCT_scaled_size > 1; + + /* Verify we can handle the sampling factors, select per-component methods, + * and create storage as needed. + */ + for (ci = 0, compptr = cinfo->comp_info; ci < cinfo->num_components; + ci++, compptr++) { + /* Compute size of an "input group" after IDCT scaling. This many samples + * are to be converted to max_h_samp_factor * max_v_samp_factor pixels. + */ + h_in_group = (compptr->h_samp_factor * compptr->_DCT_scaled_size) / + cinfo->_min_DCT_scaled_size; + v_in_group = (compptr->v_samp_factor * compptr->_DCT_scaled_size) / + cinfo->_min_DCT_scaled_size; + h_out_group = cinfo->max_h_samp_factor; + v_out_group = cinfo->max_v_samp_factor; + upsample->rowgroup_height[ci] = v_in_group; /* save for use later */ + need_buffer = TRUE; + if (!compptr->component_needed) { + /* Don't bother to upsample an uninteresting component. */ + upsample->methods[ci] = noop_upsample; + need_buffer = FALSE; + } else if (h_in_group == h_out_group && v_in_group == v_out_group) { + /* Fullsize components can be processed without any work. */ + upsample->methods[ci] = fullsize_upsample; + need_buffer = FALSE; + } else if (h_in_group * 2 == h_out_group && v_in_group == v_out_group) { + /* Special cases for 2h1v upsampling */ + if (do_fancy && compptr->downsampled_width > 2) { +#ifdef WITH_SIMD + if (jsimd_can_h2v1_fancy_upsample()) + upsample->methods[ci] = jsimd_h2v1_fancy_upsample; + else +#endif + upsample->methods[ci] = h2v1_fancy_upsample; + } else { +#ifdef WITH_SIMD + if (jsimd_can_h2v1_upsample()) + upsample->methods[ci] = jsimd_h2v1_upsample; + else +#endif + upsample->methods[ci] = h2v1_upsample; + } + } else if (h_in_group == h_out_group && + v_in_group * 2 == v_out_group && do_fancy) { + /* Non-fancy upsampling is handled by the generic method */ +#if defined(WITH_SIMD) && (defined(__arm__) || defined(__aarch64__) || \ + defined(_M_ARM) || defined(_M_ARM64)) + if (jsimd_can_h1v2_fancy_upsample()) + upsample->methods[ci] = jsimd_h1v2_fancy_upsample; + else +#endif + upsample->methods[ci] = h1v2_fancy_upsample; + upsample->pub.need_context_rows = TRUE; + } else if (h_in_group * 2 == h_out_group && + v_in_group * 2 == v_out_group) { + /* Special cases for 2h2v upsampling */ + if (do_fancy && compptr->downsampled_width > 2) { +#ifdef WITH_SIMD + if (jsimd_can_h2v2_fancy_upsample()) + upsample->methods[ci] = jsimd_h2v2_fancy_upsample; + else +#endif + upsample->methods[ci] = h2v2_fancy_upsample; + upsample->pub.need_context_rows = TRUE; + } else { +#ifdef WITH_SIMD + if (jsimd_can_h2v2_upsample()) + upsample->methods[ci] = jsimd_h2v2_upsample; + else +#endif + upsample->methods[ci] = h2v2_upsample; + } + } else if ((h_out_group % h_in_group) == 0 && + (v_out_group % v_in_group) == 0) { + /* Generic integral-factors upsampling method */ +#if defined(WITH_SIMD) && defined(__mips__) + if (jsimd_can_int_upsample()) + upsample->methods[ci] = jsimd_int_upsample; + else +#endif + upsample->methods[ci] = int_upsample; + upsample->h_expand[ci] = (UINT8)(h_out_group / h_in_group); + upsample->v_expand[ci] = (UINT8)(v_out_group / v_in_group); + } else + ERREXIT(cinfo, JERR_FRACT_SAMPLE_NOTIMPL); + if (need_buffer && !cinfo->master->jinit_upsampler_no_alloc) { + upsample->color_buf[ci] = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (JDIMENSION)jround_up((long)cinfo->output_width, + (long)cinfo->max_h_samp_factor), + (JDIMENSION)cinfo->max_v_samp_factor); + } + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || defined(D_LOSSLESS_SUPPORTED) */ diff --git a/src/spice2x/external/libjpeg-turbo/jdsample.h b/src/spice2x/external/libjpeg-turbo/jdsample.h new file mode 100644 index 0000000..a8a9298 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdsample.h @@ -0,0 +1,53 @@ +/* + * jdsample.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + */ + +#define JPEG_INTERNALS +#include "jpeglib.h" +#include "jsamplecomp.h" + + +/* Pointer to routine to upsample a single component */ +typedef void (*upsample1_ptr) (j_decompress_ptr cinfo, + jpeg_component_info *compptr, + _JSAMPARRAY input_data, + _JSAMPARRAY *output_data_ptr); + +/* Private subobject */ + +typedef struct { + struct jpeg_upsampler pub; /* public fields */ + + /* Color conversion buffer. When using separate upsampling and color + * conversion steps, this buffer holds one upsampled row group until it + * has been color converted and output. + * Note: we do not allocate any storage for component(s) which are full-size, + * ie do not need rescaling. The corresponding entry of color_buf[] is + * simply set to point to the input data array, thereby avoiding copying. + */ + _JSAMPARRAY color_buf[MAX_COMPONENTS]; + + /* Per-component upsampling method pointers */ + upsample1_ptr methods[MAX_COMPONENTS]; + + int next_row_out; /* counts rows emitted from color_buf */ + JDIMENSION rows_to_go; /* counts rows remaining in image */ + + /* Height of an input row group for each component. */ + int rowgroup_height[MAX_COMPONENTS]; + + /* These arrays save pixel expansion factors so that int_expand need not + * recompute them each time. They are unused for other upsampling methods. + */ + UINT8 h_expand[MAX_COMPONENTS]; + UINT8 v_expand[MAX_COMPONENTS]; +} my_upsampler; + +typedef my_upsampler *my_upsample_ptr; diff --git a/src/spice2x/external/libjpeg-turbo/jdtrans.c b/src/spice2x/external/libjpeg-turbo/jdtrans.c new file mode 100644 index 0000000..719813f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jdtrans.c @@ -0,0 +1,162 @@ +/* + * jdtrans.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1995-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains library routines for transcoding decompression, + * that is, reading raw DCT coefficient arrays from an input JPEG file. + * The routines in jdapimin.c will also be needed by a transcoder. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jpegapicomp.h" + + +/* Forward declarations */ +LOCAL(void) transdecode_master_selection(j_decompress_ptr cinfo); + + +/* + * Read the coefficient arrays from a JPEG file. + * jpeg_read_header must be completed before calling this. + * + * The entire image is read into a set of virtual coefficient-block arrays, + * one per component. The return value is a pointer to the array of + * virtual-array descriptors. These can be manipulated directly via the + * JPEG memory manager, or handed off to jpeg_write_coefficients(). + * To release the memory occupied by the virtual arrays, call + * jpeg_finish_decompress() when done with the data. + * + * An alternative usage is to simply obtain access to the coefficient arrays + * during a buffered-image-mode decompression operation. This is allowed + * after any jpeg_finish_output() call. The arrays can be accessed until + * jpeg_finish_decompress() is called. (Note that any call to the library + * may reposition the arrays, so don't rely on access_virt_barray() results + * to stay valid across library calls.) + * + * Returns NULL if suspended. This case need be checked only if + * a suspending data source is used. + */ + +GLOBAL(jvirt_barray_ptr *) +jpeg_read_coefficients(j_decompress_ptr cinfo) +{ + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + if (cinfo->global_state == DSTATE_READY) { + /* First call: initialize active modules */ + transdecode_master_selection(cinfo); + cinfo->global_state = DSTATE_RDCOEFS; + } + if (cinfo->global_state == DSTATE_RDCOEFS) { + /* Absorb whole file into the coef buffer */ + for (;;) { + int retcode; + /* Call progress monitor hook if present */ + if (cinfo->progress != NULL) + (*cinfo->progress->progress_monitor) ((j_common_ptr)cinfo); + /* Absorb some more input */ + retcode = (*cinfo->inputctl->consume_input) (cinfo); + if (retcode == JPEG_SUSPENDED) + return NULL; + if (retcode == JPEG_REACHED_EOI) + break; + /* Advance progress counter if appropriate */ + if (cinfo->progress != NULL && + (retcode == JPEG_ROW_COMPLETED || retcode == JPEG_REACHED_SOS)) { + if (++cinfo->progress->pass_counter >= cinfo->progress->pass_limit) { + /* startup underestimated number of scans; ratchet up one scan */ + cinfo->progress->pass_limit += (long)cinfo->total_iMCU_rows; + } + } + } + /* Set state so that jpeg_finish_decompress does the right thing */ + cinfo->global_state = DSTATE_STOPPING; + } + /* At this point we should be in state DSTATE_STOPPING if being used + * standalone, or in state DSTATE_BUFIMAGE if being invoked to get access + * to the coefficients during a full buffered-image-mode decompression. + */ + if ((cinfo->global_state == DSTATE_STOPPING || + cinfo->global_state == DSTATE_BUFIMAGE) && cinfo->buffered_image) { + return cinfo->coef->coef_arrays; + } + /* Oops, improper usage */ + ERREXIT1(cinfo, JERR_BAD_STATE, cinfo->global_state); + return NULL; /* keep compiler happy */ +} + + +/* + * Master selection of decompression modules for transcoding. + * This substitutes for jdmaster.c's initialization of the full decompressor. + */ + +LOCAL(void) +transdecode_master_selection(j_decompress_ptr cinfo) +{ + /* This is effectively a buffered-image operation. */ + cinfo->buffered_image = TRUE; + +#if JPEG_LIB_VERSION >= 80 + /* Compute output image dimensions and related values. */ + jpeg_core_output_dimensions(cinfo); +#endif + + /* Entropy decoding: either Huffman or arithmetic coding. */ + if (cinfo->arith_code) { +#ifdef D_ARITH_CODING_SUPPORTED + jinit_arith_decoder(cinfo); +#else + ERREXIT(cinfo, JERR_ARITH_NOTIMPL); +#endif + } else { + if (cinfo->progressive_mode) { +#ifdef D_PROGRESSIVE_SUPPORTED + jinit_phuff_decoder(cinfo); +#else + ERREXIT(cinfo, JERR_NOT_COMPILED); +#endif + } else + jinit_huff_decoder(cinfo); + } + + /* Always get a full-image coefficient buffer. */ + if (cinfo->data_precision == 12) + j12init_d_coef_controller(cinfo, TRUE); + else + jinit_d_coef_controller(cinfo, TRUE); + + /* We can now tell the memory manager to allocate virtual arrays. */ + (*cinfo->mem->realize_virt_arrays) ((j_common_ptr)cinfo); + + /* Initialize input side of decompressor to consume first scan. */ + (*cinfo->inputctl->start_input_pass) (cinfo); + + /* Initialize progress monitoring. */ + if (cinfo->progress != NULL) { + int nscans; + /* Estimate number of scans to set pass_limit. */ + if (cinfo->progressive_mode) { + /* Arbitrarily estimate 2 interleaved DC scans + 3 AC scans/component. */ + nscans = 2 + 3 * cinfo->num_components; + } else if (cinfo->inputctl->has_multiple_scans) { + /* For a nonprogressive multiscan file, estimate 1 scan per component. */ + nscans = cinfo->num_components; + } else { + nscans = 1; + } + cinfo->progress->pass_counter = 0L; + cinfo->progress->pass_limit = (long)cinfo->total_iMCU_rows * nscans; + cinfo->progress->completed_passes = 0; + cinfo->progress->total_passes = 1; + } +} diff --git a/src/spice2x/external/libjpeg-turbo/jerror.c b/src/spice2x/external/libjpeg-turbo/jerror.c new file mode 100644 index 0000000..2133244 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jerror.c @@ -0,0 +1,243 @@ +/* + * jerror.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1998, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, 2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains simple error-reporting and trace-message routines. + * These are suitable for Unix-like systems and others where writing to + * stderr is the right thing to do. Many applications will want to replace + * some or all of these routines. + * + * If you define USE_WINDOWS_MESSAGEBOX in jconfig.h or in the makefile, + * you get a Windows-specific hack to display error messages in a dialog box. + * It ain't much, but it beats dropping error messages into the bit bucket, + * which is what happens to output to stderr under most Windows C compilers. + * + * These routines are used by both the compression and decompression code. + */ + +/* this is not a core library module, so it doesn't define JPEG_INTERNALS */ +#include "jinclude.h" +#include "jpeglib.h" +#include "jversion.h" +#include "jerror.h" + +#ifdef USE_WINDOWS_MESSAGEBOX +#include +#endif + +#ifndef EXIT_FAILURE /* define exit() codes if not provided */ +#define EXIT_FAILURE 1 +#endif + + +/* + * Create the message string table. + * We do this from the master message list in jerror.h by re-reading + * jerror.h with a suitable definition for macro JMESSAGE. + */ + +#define JMESSAGE(code, string) string, + +static const char * const jpeg_std_message_table[] = { +#include "jerror.h" + NULL +}; + + +/* + * Error exit handler: must not return to caller. + * + * Applications may override this if they want to get control back after + * an error. Typically one would longjmp somewhere instead of exiting. + * The setjmp buffer can be made a private field within an expanded error + * handler object. Note that the info needed to generate an error message + * is stored in the error object, so you can generate the message now or + * later, at your convenience. + * You should make sure that the JPEG object is cleaned up (with jpeg_abort + * or jpeg_destroy) at some point. + */ + +METHODDEF(void) +error_exit(j_common_ptr cinfo) +{ + /* Always display the message */ + (*cinfo->err->output_message) (cinfo); + + /* Let the memory manager delete any temp files before we die */ + jpeg_destroy(cinfo); + + exit(EXIT_FAILURE); +} + + +/* + * Actual output of an error or trace message. + * Applications may override this method to send JPEG messages somewhere + * other than stderr. + * + * On Windows, printing to stderr is generally completely useless, + * so we provide optional code to produce an error-dialog popup. + * Most Windows applications will still prefer to override this routine, + * but if they don't, it'll do something at least marginally useful. + * + * NOTE: to use the library in an environment that doesn't support the + * C stdio library, you may have to delete the call to fprintf() entirely, + * not just not use this routine. + */ + +METHODDEF(void) +output_message(j_common_ptr cinfo) +{ + char buffer[JMSG_LENGTH_MAX]; + + /* Create the message */ + (*cinfo->err->format_message) (cinfo, buffer); + +#ifdef USE_WINDOWS_MESSAGEBOX + /* Display it in a message dialog box */ + MessageBox(GetActiveWindow(), buffer, "JPEG Library Error", + MB_OK | MB_ICONERROR); +#else + /* Send it to stderr, adding a newline */ + fprintf(stderr, "%s\n", buffer); +#endif +} + + +/* + * Decide whether to emit a trace or warning message. + * msg_level is one of: + * -1: recoverable corrupt-data warning, may want to abort. + * 0: important advisory messages (always display to user). + * 1: first level of tracing detail. + * 2,3,...: successively more detailed tracing messages. + * An application might override this method if it wanted to abort on warnings + * or change the policy about which messages to display. + */ + +METHODDEF(void) +emit_message(j_common_ptr cinfo, int msg_level) +{ + struct jpeg_error_mgr *err = cinfo->err; + + if (msg_level < 0) { + /* It's a warning message. Since corrupt files may generate many warnings, + * the policy implemented here is to show only the first warning, + * unless trace_level >= 3. + */ + if (err->num_warnings == 0 || err->trace_level >= 3) + (*err->output_message) (cinfo); + /* Always count warnings in num_warnings. */ + err->num_warnings++; + } else { + /* It's a trace message. Show it if trace_level >= msg_level. */ + if (err->trace_level >= msg_level) + (*err->output_message) (cinfo); + } +} + + +/* + * Format a message string for the most recent JPEG error or message. + * The message is stored into buffer, which should be at least JMSG_LENGTH_MAX + * characters. Note that no '\n' character is added to the string. + * Few applications should need to override this method. + */ + +METHODDEF(void) +format_message(j_common_ptr cinfo, char *buffer) +{ + struct jpeg_error_mgr *err = cinfo->err; + int msg_code = err->msg_code; + const char *msgtext = NULL; + const char *msgptr; + char ch; + boolean isstring; + + /* Look up message string in proper table */ + if (msg_code > 0 && msg_code <= err->last_jpeg_message) { + msgtext = err->jpeg_message_table[msg_code]; + } else if (err->addon_message_table != NULL && + msg_code >= err->first_addon_message && + msg_code <= err->last_addon_message) { + msgtext = err->addon_message_table[msg_code - err->first_addon_message]; + } + + /* Defend against bogus message number */ + if (msgtext == NULL) { + err->msg_parm.i[0] = msg_code; + msgtext = err->jpeg_message_table[0]; + } + + /* Check for string parameter, as indicated by %s in the message text */ + isstring = FALSE; + msgptr = msgtext; + while ((ch = *msgptr++) != '\0') { + if (ch == '%') { + if (*msgptr == 's') isstring = TRUE; + break; + } + } + + /* Format the message into the passed buffer */ + if (isstring) + SNPRINTF(buffer, JMSG_LENGTH_MAX, msgtext, err->msg_parm.s); + else + SNPRINTF(buffer, JMSG_LENGTH_MAX, msgtext, + err->msg_parm.i[0], err->msg_parm.i[1], + err->msg_parm.i[2], err->msg_parm.i[3], + err->msg_parm.i[4], err->msg_parm.i[5], + err->msg_parm.i[6], err->msg_parm.i[7]); +} + + +/* + * Reset error state variables at start of a new image. + * This is called during compression startup to reset trace/error + * processing to default state, without losing any application-specific + * method pointers. An application might possibly want to override + * this method if it has additional error processing state. + */ + +METHODDEF(void) +reset_error_mgr(j_common_ptr cinfo) +{ + cinfo->err->num_warnings = 0; + /* trace_level is not reset since it is an application-supplied parameter */ + cinfo->err->msg_code = 0; /* may be useful as a flag for "no error" */ +} + + +/* + * Fill in the standard error-handling methods in a jpeg_error_mgr object. + * Typical call is: + * struct jpeg_compress_struct cinfo; + * struct jpeg_error_mgr err; + * + * cinfo.err = jpeg_std_error(&err); + * after which the application may override some of the methods. + */ + +GLOBAL(struct jpeg_error_mgr *) +jpeg_std_error(struct jpeg_error_mgr *err) +{ + memset(err, 0, sizeof(struct jpeg_error_mgr)); + + err->error_exit = error_exit; + err->emit_message = emit_message; + err->output_message = output_message; + err->format_message = format_message; + err->reset_error_mgr = reset_error_mgr; + + /* Initialize message table pointers */ + err->jpeg_message_table = jpeg_std_message_table; + err->last_jpeg_message = (int)JMSG_LASTMSGCODE - 1; + + return err; +} diff --git a/src/spice2x/external/libjpeg-turbo/jerror.h b/src/spice2x/external/libjpeg-turbo/jerror.h new file mode 100644 index 0000000..71ba03e --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jerror.h @@ -0,0 +1,336 @@ +/* + * jerror.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1997, Thomas G. Lane. + * Modified 1997-2009 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2014, 2017, 2021-2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file defines the error and message codes for the JPEG library. + * Edit this file to add new codes, or to translate the message strings to + * some other language. + * A set of error-reporting macros are defined too. Some applications using + * the JPEG library may wish to include this file to get the error codes + * and/or the macros. + */ + +/* + * To define the enum list of message codes, include this file without + * defining macro JMESSAGE. To create a message string table, include it + * again with a suitable JMESSAGE definition (see jerror.c for an example). + */ +#ifndef JMESSAGE +#ifndef JERROR_H +/* First time through, define the enum list */ +#define JMAKE_ENUM_LIST +#else +/* Repeated inclusions of this file are no-ops unless JMESSAGE is defined */ +#define JMESSAGE(code, string) +#endif /* JERROR_H */ +#endif /* JMESSAGE */ + +#ifdef JMAKE_ENUM_LIST + +typedef enum { + +#define JMESSAGE(code, string) code, + +#endif /* JMAKE_ENUM_LIST */ + +JMESSAGE(JMSG_NOMESSAGE, "Bogus message code %d") /* Must be first entry! */ + +/* For maintenance convenience, list is alphabetical by message code name */ +#if JPEG_LIB_VERSION < 70 +JMESSAGE(JERR_ARITH_NOTIMPL, "Sorry, arithmetic coding is not implemented") +#endif +JMESSAGE(JERR_BAD_ALIGN_TYPE, "ALIGN_TYPE is wrong, please fix") +JMESSAGE(JERR_BAD_ALLOC_CHUNK, "MAX_ALLOC_CHUNK is wrong, please fix") +JMESSAGE(JERR_BAD_BUFFER_MODE, "Bogus buffer control mode") +JMESSAGE(JERR_BAD_COMPONENT_ID, "Invalid component ID %d in SOS") +#if JPEG_LIB_VERSION >= 70 +JMESSAGE(JERR_BAD_CROP_SPEC, "Invalid crop request") +#endif +JMESSAGE(JERR_BAD_DCT_COEF, + "DCT coefficient (lossy) or spatial difference (lossless) out of range") +JMESSAGE(JERR_BAD_DCTSIZE, "IDCT output block size %d not supported") +#if JPEG_LIB_VERSION >= 70 +JMESSAGE(JERR_BAD_DROP_SAMPLING, + "Component index %d: mismatching sampling ratio %d:%d, %d:%d, %c") +#endif +JMESSAGE(JERR_BAD_HUFF_TABLE, "Bogus Huffman table definition") +JMESSAGE(JERR_BAD_IN_COLORSPACE, "Bogus input colorspace") +JMESSAGE(JERR_BAD_J_COLORSPACE, "Bogus JPEG colorspace") +JMESSAGE(JERR_BAD_LENGTH, "Bogus marker length") +JMESSAGE(JERR_BAD_LIB_VERSION, + "Wrong JPEG library version: library is %d, caller expects %d") +JMESSAGE(JERR_BAD_MCU_SIZE, "Sampling factors too large for interleaved scan") +JMESSAGE(JERR_BAD_POOL_ID, "Invalid memory pool code %d") +JMESSAGE(JERR_BAD_PRECISION, "Unsupported JPEG data precision %d") +JMESSAGE(JERR_BAD_PROGRESSION, + "Invalid progressive/lossless parameters Ss=%d Se=%d Ah=%d Al=%d") +JMESSAGE(JERR_BAD_PROG_SCRIPT, + "Invalid progressive/lossless parameters at scan script entry %d") +JMESSAGE(JERR_BAD_SAMPLING, "Bogus sampling factors") +JMESSAGE(JERR_BAD_SCAN_SCRIPT, "Invalid scan script at entry %d") +JMESSAGE(JERR_BAD_STATE, "Improper call to JPEG library in state %d") +JMESSAGE(JERR_BAD_STRUCT_SIZE, + "JPEG parameter struct mismatch: library thinks size is %u, caller expects %u") +JMESSAGE(JERR_BAD_VIRTUAL_ACCESS, "Bogus virtual array access") +JMESSAGE(JERR_BUFFER_SIZE, "Buffer passed to JPEG library is too small") +JMESSAGE(JERR_CANT_SUSPEND, "Suspension not allowed here") +JMESSAGE(JERR_CCIR601_NOTIMPL, "CCIR601 sampling not implemented yet") +JMESSAGE(JERR_COMPONENT_COUNT, "Too many color components: %d, max %d") +JMESSAGE(JERR_CONVERSION_NOTIMPL, "Unsupported color conversion request") +JMESSAGE(JERR_DAC_INDEX, "Bogus DAC index %d") +JMESSAGE(JERR_DAC_VALUE, "Bogus DAC value 0x%x") +JMESSAGE(JERR_DHT_INDEX, "Bogus DHT index %d") +JMESSAGE(JERR_DQT_INDEX, "Bogus DQT index %d") +JMESSAGE(JERR_EMPTY_IMAGE, "Empty JPEG image (DNL not supported)") +JMESSAGE(JERR_EMS_READ, "Read from EMS failed") +JMESSAGE(JERR_EMS_WRITE, "Write to EMS failed") +JMESSAGE(JERR_EOI_EXPECTED, "Didn't expect more than one scan") +JMESSAGE(JERR_FILE_READ, "Input file read error") +JMESSAGE(JERR_FILE_WRITE, "Output file write error --- out of disk space?") +JMESSAGE(JERR_FRACT_SAMPLE_NOTIMPL, "Fractional sampling not implemented yet") +JMESSAGE(JERR_HUFF_CLEN_OVERFLOW, "Huffman code size table overflow") +JMESSAGE(JERR_HUFF_MISSING_CODE, "Missing Huffman code table entry") +JMESSAGE(JERR_IMAGE_TOO_BIG, "Maximum supported image dimension is %u pixels") +JMESSAGE(JERR_INPUT_EMPTY, "Empty input file") +JMESSAGE(JERR_INPUT_EOF, "Premature end of input file") +JMESSAGE(JERR_MISMATCHED_QUANT_TABLE, + "Cannot transcode due to multiple use of quantization table %d") +JMESSAGE(JERR_MISSING_DATA, "Scan script does not transmit all data") +JMESSAGE(JERR_MODE_CHANGE, "Invalid color quantization mode change") +JMESSAGE(JERR_NOTIMPL, "Requested features are incompatible") +JMESSAGE(JERR_NOT_COMPILED, "Requested feature was omitted at compile time") +#if JPEG_LIB_VERSION >= 70 +JMESSAGE(JERR_NO_ARITH_TABLE, "Arithmetic table 0x%02x was not defined") +#endif +JMESSAGE(JERR_NO_BACKING_STORE, "Memory limit exceeded") +JMESSAGE(JERR_NO_HUFF_TABLE, "Huffman table 0x%02x was not defined") +JMESSAGE(JERR_NO_IMAGE, "JPEG datastream contains no image") +JMESSAGE(JERR_NO_QUANT_TABLE, "Quantization table 0x%02x was not defined") +JMESSAGE(JERR_NO_SOI, "Not a JPEG file: starts with 0x%02x 0x%02x") +JMESSAGE(JERR_OUT_OF_MEMORY, "Insufficient memory (case %d)") +JMESSAGE(JERR_QUANT_COMPONENTS, + "Cannot quantize more than %d color components") +JMESSAGE(JERR_QUANT_FEW_COLORS, "Cannot quantize to fewer than %d colors") +JMESSAGE(JERR_QUANT_MANY_COLORS, "Cannot quantize to more than %d colors") +JMESSAGE(JERR_SOF_DUPLICATE, "Invalid JPEG file structure: two SOF markers") +JMESSAGE(JERR_SOF_NO_SOS, "Invalid JPEG file structure: missing SOS marker") +JMESSAGE(JERR_SOF_UNSUPPORTED, "Unsupported JPEG process: SOF type 0x%02x") +JMESSAGE(JERR_SOI_DUPLICATE, "Invalid JPEG file structure: two SOI markers") +JMESSAGE(JERR_SOS_NO_SOF, "Invalid JPEG file structure: SOS before SOF") +JMESSAGE(JERR_TFILE_CREATE, "Failed to create temporary file %s") +JMESSAGE(JERR_TFILE_READ, "Read failed on temporary file") +JMESSAGE(JERR_TFILE_SEEK, "Seek failed on temporary file") +JMESSAGE(JERR_TFILE_WRITE, + "Write failed on temporary file --- out of disk space?") +JMESSAGE(JERR_TOO_LITTLE_DATA, "Application transferred too few scanlines") +JMESSAGE(JERR_UNKNOWN_MARKER, "Unsupported marker type 0x%02x") +JMESSAGE(JERR_VIRTUAL_BUG, "Virtual array controller messed up") +JMESSAGE(JERR_WIDTH_OVERFLOW, "Image too wide for this implementation") +JMESSAGE(JERR_XMS_READ, "Read from XMS failed") +JMESSAGE(JERR_XMS_WRITE, "Write to XMS failed") +JMESSAGE(JMSG_COPYRIGHT, JCOPYRIGHT_SHORT) +JMESSAGE(JMSG_VERSION, JVERSION) +JMESSAGE(JTRC_16BIT_TABLES, + "Caution: quantization tables are too coarse for baseline JPEG") +JMESSAGE(JTRC_ADOBE, + "Adobe APP14 marker: version %d, flags 0x%04x 0x%04x, transform %d") +JMESSAGE(JTRC_APP0, "Unknown APP0 marker (not JFIF), length %u") +JMESSAGE(JTRC_APP14, "Unknown APP14 marker (not Adobe), length %u") +JMESSAGE(JTRC_DAC, "Define Arithmetic Table 0x%02x: 0x%02x") +JMESSAGE(JTRC_DHT, "Define Huffman Table 0x%02x") +JMESSAGE(JTRC_DQT, "Define Quantization Table %d precision %d") +JMESSAGE(JTRC_DRI, "Define Restart Interval %u") +JMESSAGE(JTRC_EMS_CLOSE, "Freed EMS handle %u") +JMESSAGE(JTRC_EMS_OPEN, "Obtained EMS handle %u") +JMESSAGE(JTRC_EOI, "End Of Image") +JMESSAGE(JTRC_HUFFBITS, " %3d %3d %3d %3d %3d %3d %3d %3d") +JMESSAGE(JTRC_JFIF, "JFIF APP0 marker: version %d.%02d, density %dx%d %d") +JMESSAGE(JTRC_JFIF_BADTHUMBNAILSIZE, + "Warning: thumbnail image size does not match data length %u") +JMESSAGE(JTRC_JFIF_EXTENSION, "JFIF extension marker: type 0x%02x, length %u") +JMESSAGE(JTRC_JFIF_THUMBNAIL, " with %d x %d thumbnail image") +JMESSAGE(JTRC_MISC_MARKER, "Miscellaneous marker 0x%02x, length %u") +JMESSAGE(JTRC_PARMLESS_MARKER, "Unexpected marker 0x%02x") +JMESSAGE(JTRC_QUANTVALS, " %4u %4u %4u %4u %4u %4u %4u %4u") +JMESSAGE(JTRC_QUANT_3_NCOLORS, "Quantizing to %d = %d*%d*%d colors") +JMESSAGE(JTRC_QUANT_NCOLORS, "Quantizing to %d colors") +JMESSAGE(JTRC_QUANT_SELECTED, "Selected %d colors for quantization") +JMESSAGE(JTRC_RECOVERY_ACTION, "At marker 0x%02x, recovery action %d") +JMESSAGE(JTRC_RST, "RST%d") +JMESSAGE(JTRC_SMOOTH_NOTIMPL, + "Smoothing not supported with nonstandard sampling ratios") +JMESSAGE(JTRC_SOF, "Start Of Frame 0x%02x: width=%u, height=%u, components=%d") +JMESSAGE(JTRC_SOF_COMPONENT, " Component %d: %dhx%dv q=%d") +JMESSAGE(JTRC_SOI, "Start of Image") +JMESSAGE(JTRC_SOS, "Start Of Scan: %d components") +JMESSAGE(JTRC_SOS_COMPONENT, " Component %d: dc=%d ac=%d") +JMESSAGE(JTRC_SOS_PARAMS, " Ss=%d, Se=%d, Ah=%d, Al=%d") +JMESSAGE(JTRC_TFILE_CLOSE, "Closed temporary file %s") +JMESSAGE(JTRC_TFILE_OPEN, "Opened temporary file %s") +JMESSAGE(JTRC_THUMB_JPEG, + "JFIF extension marker: JPEG-compressed thumbnail image, length %u") +JMESSAGE(JTRC_THUMB_PALETTE, + "JFIF extension marker: palette thumbnail image, length %u") +JMESSAGE(JTRC_THUMB_RGB, + "JFIF extension marker: RGB thumbnail image, length %u") +JMESSAGE(JTRC_UNKNOWN_IDS, + "Unrecognized component IDs %d %d %d, assuming YCbCr (lossy) or RGB (lossless)") +JMESSAGE(JTRC_XMS_CLOSE, "Freed XMS handle %u") +JMESSAGE(JTRC_XMS_OPEN, "Obtained XMS handle %u") +JMESSAGE(JWRN_ADOBE_XFORM, "Unknown Adobe color transform code %d") +#if JPEG_LIB_VERSION >= 70 +JMESSAGE(JWRN_ARITH_BAD_CODE, "Corrupt JPEG data: bad arithmetic code") +#endif +JMESSAGE(JWRN_BOGUS_PROGRESSION, + "Inconsistent progression sequence for component %d coefficient %d") +JMESSAGE(JWRN_EXTRANEOUS_DATA, + "Corrupt JPEG data: %u extraneous bytes before marker 0x%02x") +JMESSAGE(JWRN_HIT_MARKER, "Corrupt JPEG data: premature end of data segment") +JMESSAGE(JWRN_HUFF_BAD_CODE, "Corrupt JPEG data: bad Huffman code") +JMESSAGE(JWRN_JFIF_MAJOR, "Warning: unknown JFIF revision number %d.%02d") +JMESSAGE(JWRN_JPEG_EOF, "Premature end of JPEG file") +JMESSAGE(JWRN_MUST_RESYNC, + "Corrupt JPEG data: found marker 0x%02x instead of RST%d") +JMESSAGE(JWRN_NOT_SEQUENTIAL, "Invalid SOS parameters for sequential JPEG") +JMESSAGE(JWRN_TOO_MUCH_DATA, "Application transferred too many scanlines") +#if JPEG_LIB_VERSION < 70 +JMESSAGE(JERR_BAD_CROP_SPEC, "Invalid crop request") +#if defined(C_ARITH_CODING_SUPPORTED) || defined(D_ARITH_CODING_SUPPORTED) +JMESSAGE(JERR_NO_ARITH_TABLE, "Arithmetic table 0x%02x was not defined") +JMESSAGE(JWRN_ARITH_BAD_CODE, "Corrupt JPEG data: bad arithmetic code") +#endif +#endif +JMESSAGE(JWRN_BOGUS_ICC, "Corrupt JPEG data: bad ICC marker") +#if JPEG_LIB_VERSION < 70 +JMESSAGE(JERR_BAD_DROP_SAMPLING, + "Component index %d: mismatching sampling ratio %d:%d, %d:%d, %c") +#endif +JMESSAGE(JERR_BAD_RESTART, + "Invalid restart interval %d; must be an integer multiple of the number of MCUs in an MCU row (%d)") + +#ifdef JMAKE_ENUM_LIST + + JMSG_LASTMSGCODE +} J_MESSAGE_CODE; + +#undef JMAKE_ENUM_LIST +#endif /* JMAKE_ENUM_LIST */ + +/* Zap JMESSAGE macro so that future re-inclusions do nothing by default */ +#undef JMESSAGE + + +#ifndef JERROR_H +#define JERROR_H + +/* Macros to simplify using the error and trace message stuff */ +/* The first parameter is either type of cinfo pointer */ + +/* Fatal errors (print message and exit) */ +#define ERREXIT(cinfo, code) \ + ((cinfo)->err->msg_code = (code), \ + (*(cinfo)->err->error_exit) ((j_common_ptr)(cinfo))) +#define ERREXIT1(cinfo, code, p1) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (*(cinfo)->err->error_exit) ((j_common_ptr)(cinfo))) +#define ERREXIT2(cinfo, code, p1, p2) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (cinfo)->err->msg_parm.i[1] = (p2), \ + (*(cinfo)->err->error_exit) ((j_common_ptr)(cinfo))) +#define ERREXIT3(cinfo, code, p1, p2, p3) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (cinfo)->err->msg_parm.i[1] = (p2), \ + (cinfo)->err->msg_parm.i[2] = (p3), \ + (*(cinfo)->err->error_exit) ((j_common_ptr)(cinfo))) +#define ERREXIT4(cinfo, code, p1, p2, p3, p4) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (cinfo)->err->msg_parm.i[1] = (p2), \ + (cinfo)->err->msg_parm.i[2] = (p3), \ + (cinfo)->err->msg_parm.i[3] = (p4), \ + (*(cinfo)->err->error_exit) ((j_common_ptr)(cinfo))) +#define ERREXIT6(cinfo, code, p1, p2, p3, p4, p5, p6) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (cinfo)->err->msg_parm.i[1] = (p2), \ + (cinfo)->err->msg_parm.i[2] = (p3), \ + (cinfo)->err->msg_parm.i[3] = (p4), \ + (cinfo)->err->msg_parm.i[4] = (p5), \ + (cinfo)->err->msg_parm.i[5] = (p6), \ + (*(cinfo)->err->error_exit) ((j_common_ptr)(cinfo))) +#define ERREXITS(cinfo, code, str) \ + ((cinfo)->err->msg_code = (code), \ + strncpy((cinfo)->err->msg_parm.s, (str), JMSG_STR_PARM_MAX), \ + (cinfo)->err->msg_parm.s[JMSG_STR_PARM_MAX - 1] = '\0', \ + (*(cinfo)->err->error_exit) ((j_common_ptr)(cinfo))) + +#define MAKESTMT(stuff) do { stuff } while (0) + +/* Nonfatal errors (we can keep going, but the data is probably corrupt) */ +#define WARNMS(cinfo, code) \ + ((cinfo)->err->msg_code = (code), \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), -1)) +#define WARNMS1(cinfo, code, p1) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), -1)) +#define WARNMS2(cinfo, code, p1, p2) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (cinfo)->err->msg_parm.i[1] = (p2), \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), -1)) + +/* Informational/debugging messages */ +#define TRACEMS(cinfo, lvl, code) \ + ((cinfo)->err->msg_code = (code), \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl))) +#define TRACEMS1(cinfo, lvl, code, p1) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl))) +#define TRACEMS2(cinfo, lvl, code, p1, p2) \ + ((cinfo)->err->msg_code = (code), \ + (cinfo)->err->msg_parm.i[0] = (p1), \ + (cinfo)->err->msg_parm.i[1] = (p2), \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl))) +#define TRACEMS3(cinfo, lvl, code, p1, p2, p3) \ + MAKESTMT(int *_mp = (cinfo)->err->msg_parm.i; \ + _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); \ + (cinfo)->err->msg_code = (code); \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl)); ) +#define TRACEMS4(cinfo, lvl, code, p1, p2, p3, p4) \ + MAKESTMT(int *_mp = (cinfo)->err->msg_parm.i; \ + _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); _mp[3] = (p4); \ + (cinfo)->err->msg_code = (code); \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl)); ) +#define TRACEMS5(cinfo, lvl, code, p1, p2, p3, p4, p5) \ + MAKESTMT(int *_mp = (cinfo)->err->msg_parm.i; \ + _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); _mp[3] = (p4); \ + _mp[4] = (p5); \ + (cinfo)->err->msg_code = (code); \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl)); ) +#define TRACEMS8(cinfo, lvl, code, p1, p2, p3, p4, p5, p6, p7, p8) \ + MAKESTMT(int *_mp = (cinfo)->err->msg_parm.i; \ + _mp[0] = (p1); _mp[1] = (p2); _mp[2] = (p3); _mp[3] = (p4); \ + _mp[4] = (p5); _mp[5] = (p6); _mp[6] = (p7); _mp[7] = (p8); \ + (cinfo)->err->msg_code = (code); \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl)); ) +#define TRACEMSS(cinfo, lvl, code, str) \ + ((cinfo)->err->msg_code = (code), \ + strncpy((cinfo)->err->msg_parm.s, (str), JMSG_STR_PARM_MAX), \ + (cinfo)->err->msg_parm.s[JMSG_STR_PARM_MAX - 1] = '\0', \ + (*(cinfo)->err->emit_message) ((j_common_ptr)(cinfo), (lvl))) + +#endif /* JERROR_H */ diff --git a/src/spice2x/external/libjpeg-turbo/jfdctflt.c b/src/spice2x/external/libjpeg-turbo/jfdctflt.c new file mode 100644 index 0000000..ab6f6d0 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jfdctflt.c @@ -0,0 +1,169 @@ +/* + * jfdctflt.c + * + * Copyright (C) 1994-1996, Thomas G. Lane. + * This file is part of the Independent JPEG Group's software. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains a floating-point implementation of the + * forward DCT (Discrete Cosine Transform). + * + * This implementation should be more accurate than either of the integer + * DCT implementations. However, it may not give the same results on all + * machines because of differences in roundoff behavior. Speed will depend + * on the hardware's floating point capacity. + * + * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT + * on each column. Direct algorithms are also available, but they are + * much more complex and seem not to be any faster when reduced to code. + * + * This implementation is based on Arai, Agui, and Nakajima's algorithm for + * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in + * Japanese, but the algorithm is described in the Pennebaker & Mitchell + * JPEG textbook (see REFERENCES section in file README.ijg). The following + * code is based directly on figure 4-8 in P&M. + * While an 8-point DCT cannot be done in less than 11 multiplies, it is + * possible to arrange the computation so that many of the multiplies are + * simple scalings of the final outputs. These multiplies can then be + * folded into the multiplications or divisions by the JPEG quantization + * table entries. The AA&N method leaves only 5 multiplies and 29 adds + * to be done in the DCT itself. + * The primary disadvantage of this method is that with a fixed-point + * implementation, accuracy is lost due to imprecise representation of the + * scaled quantization values. However, that problem does not arise if + * we use floating point arithmetic. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ + +#ifdef DCT_FLOAT_SUPPORTED + + +/* + * This module is specialized to the case DCTSIZE = 8. + */ + +#if DCTSIZE != 8 + Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ +#endif + + +/* + * Perform the forward DCT on one block of samples. + */ + +GLOBAL(void) +jpeg_fdct_float(FAST_FLOAT *data) +{ + FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + FAST_FLOAT tmp10, tmp11, tmp12, tmp13; + FAST_FLOAT z1, z2, z3, z4, z5, z11, z13; + FAST_FLOAT *dataptr; + int ctr; + + /* Pass 1: process rows. */ + + dataptr = data; + for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) { + tmp0 = dataptr[0] + dataptr[7]; + tmp7 = dataptr[0] - dataptr[7]; + tmp1 = dataptr[1] + dataptr[6]; + tmp6 = dataptr[1] - dataptr[6]; + tmp2 = dataptr[2] + dataptr[5]; + tmp5 = dataptr[2] - dataptr[5]; + tmp3 = dataptr[3] + dataptr[4]; + tmp4 = dataptr[3] - dataptr[4]; + + /* Even part */ + + tmp10 = tmp0 + tmp3; /* phase 2 */ + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + dataptr[0] = tmp10 + tmp11; /* phase 3 */ + dataptr[4] = tmp10 - tmp11; + + z1 = (tmp12 + tmp13) * ((FAST_FLOAT)0.707106781); /* c4 */ + dataptr[2] = tmp13 + z1; /* phase 5 */ + dataptr[6] = tmp13 - z1; + + /* Odd part */ + + tmp10 = tmp4 + tmp5; /* phase 2 */ + tmp11 = tmp5 + tmp6; + tmp12 = tmp6 + tmp7; + + /* The rotator is modified from fig 4-8 to avoid extra negations. */ + z5 = (tmp10 - tmp12) * ((FAST_FLOAT)0.382683433); /* c6 */ + z2 = ((FAST_FLOAT)0.541196100) * tmp10 + z5; /* c2-c6 */ + z4 = ((FAST_FLOAT)1.306562965) * tmp12 + z5; /* c2+c6 */ + z3 = tmp11 * ((FAST_FLOAT)0.707106781); /* c4 */ + + z11 = tmp7 + z3; /* phase 5 */ + z13 = tmp7 - z3; + + dataptr[5] = z13 + z2; /* phase 6 */ + dataptr[3] = z13 - z2; + dataptr[1] = z11 + z4; + dataptr[7] = z11 - z4; + + dataptr += DCTSIZE; /* advance pointer to next row */ + } + + /* Pass 2: process columns. */ + + dataptr = data; + for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) { + tmp0 = dataptr[DCTSIZE * 0] + dataptr[DCTSIZE * 7]; + tmp7 = dataptr[DCTSIZE * 0] - dataptr[DCTSIZE * 7]; + tmp1 = dataptr[DCTSIZE * 1] + dataptr[DCTSIZE * 6]; + tmp6 = dataptr[DCTSIZE * 1] - dataptr[DCTSIZE * 6]; + tmp2 = dataptr[DCTSIZE * 2] + dataptr[DCTSIZE * 5]; + tmp5 = dataptr[DCTSIZE * 2] - dataptr[DCTSIZE * 5]; + tmp3 = dataptr[DCTSIZE * 3] + dataptr[DCTSIZE * 4]; + tmp4 = dataptr[DCTSIZE * 3] - dataptr[DCTSIZE * 4]; + + /* Even part */ + + tmp10 = tmp0 + tmp3; /* phase 2 */ + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + dataptr[DCTSIZE * 0] = tmp10 + tmp11; /* phase 3 */ + dataptr[DCTSIZE * 4] = tmp10 - tmp11; + + z1 = (tmp12 + tmp13) * ((FAST_FLOAT)0.707106781); /* c4 */ + dataptr[DCTSIZE * 2] = tmp13 + z1; /* phase 5 */ + dataptr[DCTSIZE * 6] = tmp13 - z1; + + /* Odd part */ + + tmp10 = tmp4 + tmp5; /* phase 2 */ + tmp11 = tmp5 + tmp6; + tmp12 = tmp6 + tmp7; + + /* The rotator is modified from fig 4-8 to avoid extra negations. */ + z5 = (tmp10 - tmp12) * ((FAST_FLOAT)0.382683433); /* c6 */ + z2 = ((FAST_FLOAT)0.541196100) * tmp10 + z5; /* c2-c6 */ + z4 = ((FAST_FLOAT)1.306562965) * tmp12 + z5; /* c2+c6 */ + z3 = tmp11 * ((FAST_FLOAT)0.707106781); /* c4 */ + + z11 = tmp7 + z3; /* phase 5 */ + z13 = tmp7 - z3; + + dataptr[DCTSIZE * 5] = z13 + z2; /* phase 6 */ + dataptr[DCTSIZE * 3] = z13 - z2; + dataptr[DCTSIZE * 1] = z11 + z4; + dataptr[DCTSIZE * 7] = z11 - z4; + + dataptr++; /* advance pointer to next column */ + } +} + +#endif /* DCT_FLOAT_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jfdctfst.c b/src/spice2x/external/libjpeg-turbo/jfdctfst.c new file mode 100644 index 0000000..26070d1 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jfdctfst.c @@ -0,0 +1,227 @@ +/* + * jfdctfst.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains a fast, not so accurate integer implementation of the + * forward DCT (Discrete Cosine Transform). + * + * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT + * on each column. Direct algorithms are also available, but they are + * much more complex and seem not to be any faster when reduced to code. + * + * This implementation is based on Arai, Agui, and Nakajima's algorithm for + * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in + * Japanese, but the algorithm is described in the Pennebaker & Mitchell + * JPEG textbook (see REFERENCES section in file README.ijg). The following + * code is based directly on figure 4-8 in P&M. + * While an 8-point DCT cannot be done in less than 11 multiplies, it is + * possible to arrange the computation so that many of the multiplies are + * simple scalings of the final outputs. These multiplies can then be + * folded into the multiplications or divisions by the JPEG quantization + * table entries. The AA&N method leaves only 5 multiplies and 29 adds + * to be done in the DCT itself. + * The primary disadvantage of this method is that with fixed-point math, + * accuracy is lost due to imprecise representation of the scaled + * quantization values. The smaller the quantization table entry, the less + * precise the scaled value, so this implementation does worse with high- + * quality-setting files than with low-quality ones. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ + +#ifdef DCT_IFAST_SUPPORTED + + +/* + * This module is specialized to the case DCTSIZE = 8. + */ + +#if DCTSIZE != 8 + Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ +#endif + + +/* Scaling decisions are generally the same as in the LL&M algorithm; + * see jfdctint.c for more details. However, we choose to descale + * (right shift) multiplication products as soon as they are formed, + * rather than carrying additional fractional bits into subsequent additions. + * This compromises accuracy slightly, but it lets us save a few shifts. + * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples) + * everywhere except in the multiplications proper; this saves a good deal + * of work on 16-bit-int machines. + * + * Again to save a few shifts, the intermediate results between pass 1 and + * pass 2 are not upscaled, but are represented only to integral precision. + * + * A final compromise is to represent the multiplicative constants to only + * 8 fractional bits, rather than 13. This saves some shifting work on some + * machines, and may also reduce the cost of multiplication (since there + * are fewer one-bits in the constants). + */ + +#define CONST_BITS 8 + + +/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus + * causing a lot of useless floating-point operations at run time. + * To get around this we use the following pre-calculated constants. + * If you change CONST_BITS you may want to add appropriate values. + * (With a reasonable C compiler, you can just rely on the FIX() macro...) + */ + +#if CONST_BITS == 8 +#define FIX_0_382683433 ((JLONG)98) /* FIX(0.382683433) */ +#define FIX_0_541196100 ((JLONG)139) /* FIX(0.541196100) */ +#define FIX_0_707106781 ((JLONG)181) /* FIX(0.707106781) */ +#define FIX_1_306562965 ((JLONG)334) /* FIX(1.306562965) */ +#else +#define FIX_0_382683433 FIX(0.382683433) +#define FIX_0_541196100 FIX(0.541196100) +#define FIX_0_707106781 FIX(0.707106781) +#define FIX_1_306562965 FIX(1.306562965) +#endif + + +/* We can gain a little more speed, with a further compromise in accuracy, + * by omitting the addition in a descaling shift. This yields an incorrectly + * rounded result half the time... + */ + +#ifndef USE_ACCURATE_ROUNDING +#undef DESCALE +#define DESCALE(x, n) RIGHT_SHIFT(x, n) +#endif + + +/* Multiply a DCTELEM variable by an JLONG constant, and immediately + * descale to yield a DCTELEM result. + */ + +#define MULTIPLY(var, const) ((DCTELEM)DESCALE((var) * (const), CONST_BITS)) + + +/* + * Perform the forward DCT on one block of samples. + */ + +GLOBAL(void) +_jpeg_fdct_ifast(DCTELEM *data) +{ + DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + DCTELEM tmp10, tmp11, tmp12, tmp13; + DCTELEM z1, z2, z3, z4, z5, z11, z13; + DCTELEM *dataptr; + int ctr; + SHIFT_TEMPS + + /* Pass 1: process rows. */ + + dataptr = data; + for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) { + tmp0 = dataptr[0] + dataptr[7]; + tmp7 = dataptr[0] - dataptr[7]; + tmp1 = dataptr[1] + dataptr[6]; + tmp6 = dataptr[1] - dataptr[6]; + tmp2 = dataptr[2] + dataptr[5]; + tmp5 = dataptr[2] - dataptr[5]; + tmp3 = dataptr[3] + dataptr[4]; + tmp4 = dataptr[3] - dataptr[4]; + + /* Even part */ + + tmp10 = tmp0 + tmp3; /* phase 2 */ + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + dataptr[0] = tmp10 + tmp11; /* phase 3 */ + dataptr[4] = tmp10 - tmp11; + + z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */ + dataptr[2] = tmp13 + z1; /* phase 5 */ + dataptr[6] = tmp13 - z1; + + /* Odd part */ + + tmp10 = tmp4 + tmp5; /* phase 2 */ + tmp11 = tmp5 + tmp6; + tmp12 = tmp6 + tmp7; + + /* The rotator is modified from fig 4-8 to avoid extra negations. */ + z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */ + z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */ + z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */ + z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */ + + z11 = tmp7 + z3; /* phase 5 */ + z13 = tmp7 - z3; + + dataptr[5] = z13 + z2; /* phase 6 */ + dataptr[3] = z13 - z2; + dataptr[1] = z11 + z4; + dataptr[7] = z11 - z4; + + dataptr += DCTSIZE; /* advance pointer to next row */ + } + + /* Pass 2: process columns. */ + + dataptr = data; + for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) { + tmp0 = dataptr[DCTSIZE * 0] + dataptr[DCTSIZE * 7]; + tmp7 = dataptr[DCTSIZE * 0] - dataptr[DCTSIZE * 7]; + tmp1 = dataptr[DCTSIZE * 1] + dataptr[DCTSIZE * 6]; + tmp6 = dataptr[DCTSIZE * 1] - dataptr[DCTSIZE * 6]; + tmp2 = dataptr[DCTSIZE * 2] + dataptr[DCTSIZE * 5]; + tmp5 = dataptr[DCTSIZE * 2] - dataptr[DCTSIZE * 5]; + tmp3 = dataptr[DCTSIZE * 3] + dataptr[DCTSIZE * 4]; + tmp4 = dataptr[DCTSIZE * 3] - dataptr[DCTSIZE * 4]; + + /* Even part */ + + tmp10 = tmp0 + tmp3; /* phase 2 */ + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + dataptr[DCTSIZE * 0] = tmp10 + tmp11; /* phase 3 */ + dataptr[DCTSIZE * 4] = tmp10 - tmp11; + + z1 = MULTIPLY(tmp12 + tmp13, FIX_0_707106781); /* c4 */ + dataptr[DCTSIZE * 2] = tmp13 + z1; /* phase 5 */ + dataptr[DCTSIZE * 6] = tmp13 - z1; + + /* Odd part */ + + tmp10 = tmp4 + tmp5; /* phase 2 */ + tmp11 = tmp5 + tmp6; + tmp12 = tmp6 + tmp7; + + /* The rotator is modified from fig 4-8 to avoid extra negations. */ + z5 = MULTIPLY(tmp10 - tmp12, FIX_0_382683433); /* c6 */ + z2 = MULTIPLY(tmp10, FIX_0_541196100) + z5; /* c2-c6 */ + z4 = MULTIPLY(tmp12, FIX_1_306562965) + z5; /* c2+c6 */ + z3 = MULTIPLY(tmp11, FIX_0_707106781); /* c4 */ + + z11 = tmp7 + z3; /* phase 5 */ + z13 = tmp7 - z3; + + dataptr[DCTSIZE * 5] = z13 + z2; /* phase 6 */ + dataptr[DCTSIZE * 3] = z13 - z2; + dataptr[DCTSIZE * 1] = z11 + z4; + dataptr[DCTSIZE * 7] = z11 - z4; + + dataptr++; /* advance pointer to next column */ + } +} + +#endif /* DCT_IFAST_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jfdctint.c b/src/spice2x/external/libjpeg-turbo/jfdctint.c new file mode 100644 index 0000000..974013f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jfdctint.c @@ -0,0 +1,288 @@ +/* + * jfdctint.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains a slower but more accurate integer implementation of the + * forward DCT (Discrete Cosine Transform). + * + * A 2-D DCT can be done by 1-D DCT on each row followed by 1-D DCT + * on each column. Direct algorithms are also available, but they are + * much more complex and seem not to be any faster when reduced to code. + * + * This implementation is based on an algorithm described in + * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT + * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics, + * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991. + * The primary algorithm described there uses 11 multiplies and 29 adds. + * We use their alternate method with 12 multiplies and 32 adds. + * The advantage of this method is that no data path contains more than one + * multiplication; this allows a very simple and accurate implementation in + * scaled fixed-point arithmetic, with a minimal number of shifts. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ + +#ifdef DCT_ISLOW_SUPPORTED + + +/* + * This module is specialized to the case DCTSIZE = 8. + */ + +#if DCTSIZE != 8 + Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ +#endif + + +/* + * The poop on this scaling stuff is as follows: + * + * Each 1-D DCT step produces outputs which are a factor of sqrt(N) + * larger than the true DCT outputs. The final outputs are therefore + * a factor of N larger than desired; since N=8 this can be cured by + * a simple right shift at the end of the algorithm. The advantage of + * this arrangement is that we save two multiplications per 1-D DCT, + * because the y0 and y4 outputs need not be divided by sqrt(N). + * In the IJG code, this factor of 8 is removed by the quantization step + * (in jcdctmgr.c), NOT in this module. + * + * We have to do addition and subtraction of the integer inputs, which + * is no problem, and multiplication by fractional constants, which is + * a problem to do in integer arithmetic. We multiply all the constants + * by CONST_SCALE and convert them to integer constants (thus retaining + * CONST_BITS bits of precision in the constants). After doing a + * multiplication we have to divide the product by CONST_SCALE, with proper + * rounding, to produce the correct output. This division can be done + * cheaply as a right shift of CONST_BITS bits. We postpone shifting + * as long as possible so that partial sums can be added together with + * full fractional precision. + * + * The outputs of the first pass are scaled up by PASS1_BITS bits so that + * they are represented to better-than-integral precision. These outputs + * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word + * with the recommended scaling. (For 12-bit sample data, the intermediate + * array is JLONG anyway.) + * + * To avoid overflow of the 32-bit intermediate results in pass 2, we must + * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis + * shows that the values given below are the most effective. + */ + +#if BITS_IN_JSAMPLE == 8 +#define CONST_BITS 13 +#define PASS1_BITS 2 +#else +#define CONST_BITS 13 +#define PASS1_BITS 1 /* lose a little precision to avoid overflow */ +#endif + +/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus + * causing a lot of useless floating-point operations at run time. + * To get around this we use the following pre-calculated constants. + * If you change CONST_BITS you may want to add appropriate values. + * (With a reasonable C compiler, you can just rely on the FIX() macro...) + */ + +#if CONST_BITS == 13 +#define FIX_0_298631336 ((JLONG)2446) /* FIX(0.298631336) */ +#define FIX_0_390180644 ((JLONG)3196) /* FIX(0.390180644) */ +#define FIX_0_541196100 ((JLONG)4433) /* FIX(0.541196100) */ +#define FIX_0_765366865 ((JLONG)6270) /* FIX(0.765366865) */ +#define FIX_0_899976223 ((JLONG)7373) /* FIX(0.899976223) */ +#define FIX_1_175875602 ((JLONG)9633) /* FIX(1.175875602) */ +#define FIX_1_501321110 ((JLONG)12299) /* FIX(1.501321110) */ +#define FIX_1_847759065 ((JLONG)15137) /* FIX(1.847759065) */ +#define FIX_1_961570560 ((JLONG)16069) /* FIX(1.961570560) */ +#define FIX_2_053119869 ((JLONG)16819) /* FIX(2.053119869) */ +#define FIX_2_562915447 ((JLONG)20995) /* FIX(2.562915447) */ +#define FIX_3_072711026 ((JLONG)25172) /* FIX(3.072711026) */ +#else +#define FIX_0_298631336 FIX(0.298631336) +#define FIX_0_390180644 FIX(0.390180644) +#define FIX_0_541196100 FIX(0.541196100) +#define FIX_0_765366865 FIX(0.765366865) +#define FIX_0_899976223 FIX(0.899976223) +#define FIX_1_175875602 FIX(1.175875602) +#define FIX_1_501321110 FIX(1.501321110) +#define FIX_1_847759065 FIX(1.847759065) +#define FIX_1_961570560 FIX(1.961570560) +#define FIX_2_053119869 FIX(2.053119869) +#define FIX_2_562915447 FIX(2.562915447) +#define FIX_3_072711026 FIX(3.072711026) +#endif + + +/* Multiply an JLONG variable by an JLONG constant to yield an JLONG result. + * For 8-bit samples with the recommended scaling, all the variable + * and constant values involved are no more than 16 bits wide, so a + * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. + * For 12-bit samples, a full 32-bit multiplication will be needed. + */ + +#if BITS_IN_JSAMPLE == 8 +#define MULTIPLY(var, const) MULTIPLY16C16(var, const) +#else +#define MULTIPLY(var, const) ((var) * (const)) +#endif + + +/* + * Perform the forward DCT on one block of samples. + */ + +GLOBAL(void) +_jpeg_fdct_islow(DCTELEM *data) +{ + JLONG tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + JLONG tmp10, tmp11, tmp12, tmp13; + JLONG z1, z2, z3, z4, z5; + DCTELEM *dataptr; + int ctr; + SHIFT_TEMPS + + /* Pass 1: process rows. */ + /* Note results are scaled up by sqrt(8) compared to a true DCT; */ + /* furthermore, we scale the results by 2**PASS1_BITS. */ + + dataptr = data; + for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) { + tmp0 = dataptr[0] + dataptr[7]; + tmp7 = dataptr[0] - dataptr[7]; + tmp1 = dataptr[1] + dataptr[6]; + tmp6 = dataptr[1] - dataptr[6]; + tmp2 = dataptr[2] + dataptr[5]; + tmp5 = dataptr[2] - dataptr[5]; + tmp3 = dataptr[3] + dataptr[4]; + tmp4 = dataptr[3] - dataptr[4]; + + /* Even part per LL&M figure 1 --- note that published figure is faulty; + * rotator "sqrt(2)*c1" should be "sqrt(2)*c6". + */ + + tmp10 = tmp0 + tmp3; + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + dataptr[0] = (DCTELEM)LEFT_SHIFT(tmp10 + tmp11, PASS1_BITS); + dataptr[4] = (DCTELEM)LEFT_SHIFT(tmp10 - tmp11, PASS1_BITS); + + z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); + dataptr[2] = (DCTELEM)DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), + CONST_BITS - PASS1_BITS); + dataptr[6] = (DCTELEM)DESCALE(z1 + MULTIPLY(tmp12, -FIX_1_847759065), + CONST_BITS - PASS1_BITS); + + /* Odd part per figure 8 --- note paper omits factor of sqrt(2). + * cK represents cos(K*pi/16). + * i0..i3 in the paper are tmp4..tmp7 here. + */ + + z1 = tmp4 + tmp7; + z2 = tmp5 + tmp6; + z3 = tmp4 + tmp6; + z4 = tmp5 + tmp7; + z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */ + + tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */ + tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */ + tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */ + tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */ + z1 = MULTIPLY(z1, -FIX_0_899976223); /* sqrt(2) * ( c7-c3) */ + z2 = MULTIPLY(z2, -FIX_2_562915447); /* sqrt(2) * (-c1-c3) */ + z3 = MULTIPLY(z3, -FIX_1_961570560); /* sqrt(2) * (-c3-c5) */ + z4 = MULTIPLY(z4, -FIX_0_390180644); /* sqrt(2) * ( c5-c3) */ + + z3 += z5; + z4 += z5; + + dataptr[7] = (DCTELEM)DESCALE(tmp4 + z1 + z3, CONST_BITS - PASS1_BITS); + dataptr[5] = (DCTELEM)DESCALE(tmp5 + z2 + z4, CONST_BITS - PASS1_BITS); + dataptr[3] = (DCTELEM)DESCALE(tmp6 + z2 + z3, CONST_BITS - PASS1_BITS); + dataptr[1] = (DCTELEM)DESCALE(tmp7 + z1 + z4, CONST_BITS - PASS1_BITS); + + dataptr += DCTSIZE; /* advance pointer to next row */ + } + + /* Pass 2: process columns. + * We remove the PASS1_BITS scaling, but leave the results scaled up + * by an overall factor of 8. + */ + + dataptr = data; + for (ctr = DCTSIZE - 1; ctr >= 0; ctr--) { + tmp0 = dataptr[DCTSIZE * 0] + dataptr[DCTSIZE * 7]; + tmp7 = dataptr[DCTSIZE * 0] - dataptr[DCTSIZE * 7]; + tmp1 = dataptr[DCTSIZE * 1] + dataptr[DCTSIZE * 6]; + tmp6 = dataptr[DCTSIZE * 1] - dataptr[DCTSIZE * 6]; + tmp2 = dataptr[DCTSIZE * 2] + dataptr[DCTSIZE * 5]; + tmp5 = dataptr[DCTSIZE * 2] - dataptr[DCTSIZE * 5]; + tmp3 = dataptr[DCTSIZE * 3] + dataptr[DCTSIZE * 4]; + tmp4 = dataptr[DCTSIZE * 3] - dataptr[DCTSIZE * 4]; + + /* Even part per LL&M figure 1 --- note that published figure is faulty; + * rotator "sqrt(2)*c1" should be "sqrt(2)*c6". + */ + + tmp10 = tmp0 + tmp3; + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + dataptr[DCTSIZE * 0] = (DCTELEM)DESCALE(tmp10 + tmp11, PASS1_BITS); + dataptr[DCTSIZE * 4] = (DCTELEM)DESCALE(tmp10 - tmp11, PASS1_BITS); + + z1 = MULTIPLY(tmp12 + tmp13, FIX_0_541196100); + dataptr[DCTSIZE * 2] = + (DCTELEM)DESCALE(z1 + MULTIPLY(tmp13, FIX_0_765366865), + CONST_BITS + PASS1_BITS); + dataptr[DCTSIZE * 6] = + (DCTELEM)DESCALE(z1 + MULTIPLY(tmp12, -FIX_1_847759065), + CONST_BITS + PASS1_BITS); + + /* Odd part per figure 8 --- note paper omits factor of sqrt(2). + * cK represents cos(K*pi/16). + * i0..i3 in the paper are tmp4..tmp7 here. + */ + + z1 = tmp4 + tmp7; + z2 = tmp5 + tmp6; + z3 = tmp4 + tmp6; + z4 = tmp5 + tmp7; + z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */ + + tmp4 = MULTIPLY(tmp4, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */ + tmp5 = MULTIPLY(tmp5, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */ + tmp6 = MULTIPLY(tmp6, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */ + tmp7 = MULTIPLY(tmp7, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */ + z1 = MULTIPLY(z1, -FIX_0_899976223); /* sqrt(2) * ( c7-c3) */ + z2 = MULTIPLY(z2, -FIX_2_562915447); /* sqrt(2) * (-c1-c3) */ + z3 = MULTIPLY(z3, -FIX_1_961570560); /* sqrt(2) * (-c3-c5) */ + z4 = MULTIPLY(z4, -FIX_0_390180644); /* sqrt(2) * ( c5-c3) */ + + z3 += z5; + z4 += z5; + + dataptr[DCTSIZE * 7] = (DCTELEM)DESCALE(tmp4 + z1 + z3, + CONST_BITS + PASS1_BITS); + dataptr[DCTSIZE * 5] = (DCTELEM)DESCALE(tmp5 + z2 + z4, + CONST_BITS + PASS1_BITS); + dataptr[DCTSIZE * 3] = (DCTELEM)DESCALE(tmp6 + z2 + z3, + CONST_BITS + PASS1_BITS); + dataptr[DCTSIZE * 1] = (DCTELEM)DESCALE(tmp7 + z1 + z4, + CONST_BITS + PASS1_BITS); + + dataptr++; /* advance pointer to next column */ + } +} + +#endif /* DCT_ISLOW_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jidctflt.c b/src/spice2x/external/libjpeg-turbo/jidctflt.c new file mode 100644 index 0000000..ee3a31a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jidctflt.c @@ -0,0 +1,240 @@ +/* + * jidctflt.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1998, Thomas G. Lane. + * Modified 2010 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2014, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains a floating-point implementation of the + * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine + * must also perform dequantization of the input coefficients. + * + * This implementation should be more accurate than either of the integer + * IDCT implementations. However, it may not give the same results on all + * machines because of differences in roundoff behavior. Speed will depend + * on the hardware's floating point capacity. + * + * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT + * on each row (or vice versa, but it's more convenient to emit a row at + * a time). Direct algorithms are also available, but they are much more + * complex and seem not to be any faster when reduced to code. + * + * This implementation is based on Arai, Agui, and Nakajima's algorithm for + * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in + * Japanese, but the algorithm is described in the Pennebaker & Mitchell + * JPEG textbook (see REFERENCES section in file README.ijg). The following + * code is based directly on figure 4-8 in P&M. + * While an 8-point DCT cannot be done in less than 11 multiplies, it is + * possible to arrange the computation so that many of the multiplies are + * simple scalings of the final outputs. These multiplies can then be + * folded into the multiplications or divisions by the JPEG quantization + * table entries. The AA&N method leaves only 5 multiplies and 29 adds + * to be done in the DCT itself. + * The primary disadvantage of this method is that with a fixed-point + * implementation, accuracy is lost due to imprecise representation of the + * scaled quantization values. However, that problem does not arise if + * we use floating point arithmetic. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ + +#ifdef DCT_FLOAT_SUPPORTED + + +/* + * This module is specialized to the case DCTSIZE = 8. + */ + +#if DCTSIZE != 8 + Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ +#endif + + +/* Dequantize a coefficient by multiplying it by the multiplier-table + * entry; produce a float result. + */ + +#define DEQUANTIZE(coef, quantval) (((FAST_FLOAT)(coef)) * (quantval)) + + +/* + * Perform dequantization and inverse DCT on one block of coefficients. + */ + +GLOBAL(void) +_jpeg_idct_float(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + FAST_FLOAT tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + FAST_FLOAT tmp10, tmp11, tmp12, tmp13; + FAST_FLOAT z5, z10, z11, z12, z13; + JCOEFPTR inptr; + FLOAT_MULT_TYPE *quantptr; + FAST_FLOAT *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int ctr; + FAST_FLOAT workspace[DCTSIZE2]; /* buffers data between passes */ +#define _0_125 ((FLOAT_MULT_TYPE)0.125) + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (FLOAT_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = DCTSIZE; ctr > 0; ctr--) { + /* Due to quantization, we will usually find that many of the input + * coefficients are zero, especially the AC terms. We can exploit this + * by short-circuiting the IDCT calculation for any column in which all + * the AC terms are zero. In that case each output is equal to the + * DC coefficient (with scale factor as needed). + * With typical images and quantization tables, half or more of the + * column DCT calculations can be simplified this way. + */ + + if (inptr[DCTSIZE * 1] == 0 && inptr[DCTSIZE * 2] == 0 && + inptr[DCTSIZE * 3] == 0 && inptr[DCTSIZE * 4] == 0 && + inptr[DCTSIZE * 5] == 0 && inptr[DCTSIZE * 6] == 0 && + inptr[DCTSIZE * 7] == 0) { + /* AC terms all zero */ + FAST_FLOAT dcval = DEQUANTIZE(inptr[DCTSIZE * 0], + quantptr[DCTSIZE * 0] * _0_125); + + wsptr[DCTSIZE * 0] = dcval; + wsptr[DCTSIZE * 1] = dcval; + wsptr[DCTSIZE * 2] = dcval; + wsptr[DCTSIZE * 3] = dcval; + wsptr[DCTSIZE * 4] = dcval; + wsptr[DCTSIZE * 5] = dcval; + wsptr[DCTSIZE * 6] = dcval; + wsptr[DCTSIZE * 7] = dcval; + + inptr++; /* advance pointers to next column */ + quantptr++; + wsptr++; + continue; + } + + /* Even part */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0] * _0_125); + tmp1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2] * _0_125); + tmp2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4] * _0_125); + tmp3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6] * _0_125); + + tmp10 = tmp0 + tmp2; /* phase 3 */ + tmp11 = tmp0 - tmp2; + + tmp13 = tmp1 + tmp3; /* phases 5-3 */ + tmp12 = (tmp1 - tmp3) * ((FAST_FLOAT)1.414213562) - tmp13; /* 2*c4 */ + + tmp0 = tmp10 + tmp13; /* phase 2 */ + tmp3 = tmp10 - tmp13; + tmp1 = tmp11 + tmp12; + tmp2 = tmp11 - tmp12; + + /* Odd part */ + + tmp4 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1] * _0_125); + tmp5 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3] * _0_125); + tmp6 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5] * _0_125); + tmp7 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7] * _0_125); + + z13 = tmp6 + tmp5; /* phase 6 */ + z10 = tmp6 - tmp5; + z11 = tmp4 + tmp7; + z12 = tmp4 - tmp7; + + tmp7 = z11 + z13; /* phase 5 */ + tmp11 = (z11 - z13) * ((FAST_FLOAT)1.414213562); /* 2*c4 */ + + z5 = (z10 + z12) * ((FAST_FLOAT)1.847759065); /* 2*c2 */ + tmp10 = z5 - z12 * ((FAST_FLOAT)1.082392200); /* 2*(c2-c6) */ + tmp12 = z5 - z10 * ((FAST_FLOAT)2.613125930); /* 2*(c2+c6) */ + + tmp6 = tmp12 - tmp7; /* phase 2 */ + tmp5 = tmp11 - tmp6; + tmp4 = tmp10 - tmp5; + + wsptr[DCTSIZE * 0] = tmp0 + tmp7; + wsptr[DCTSIZE * 7] = tmp0 - tmp7; + wsptr[DCTSIZE * 1] = tmp1 + tmp6; + wsptr[DCTSIZE * 6] = tmp1 - tmp6; + wsptr[DCTSIZE * 2] = tmp2 + tmp5; + wsptr[DCTSIZE * 5] = tmp2 - tmp5; + wsptr[DCTSIZE * 3] = tmp3 + tmp4; + wsptr[DCTSIZE * 4] = tmp3 - tmp4; + + inptr++; /* advance pointers to next column */ + quantptr++; + wsptr++; + } + + /* Pass 2: process rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < DCTSIZE; ctr++) { + outptr = output_buf[ctr] + output_col; + /* Rows of zeroes can be exploited in the same way as we did with columns. + * However, the column calculation has created many nonzero AC terms, so + * the simplification applies less often (typically 5% to 10% of the time). + * And testing floats for zero is relatively expensive, so we don't bother. + */ + + /* Even part */ + + /* Apply signed->unsigned and prepare float->int conversion */ + z5 = wsptr[0] + ((FAST_FLOAT)_CENTERJSAMPLE + (FAST_FLOAT)0.5); + tmp10 = z5 + wsptr[4]; + tmp11 = z5 - wsptr[4]; + + tmp13 = wsptr[2] + wsptr[6]; + tmp12 = (wsptr[2] - wsptr[6]) * ((FAST_FLOAT)1.414213562) - tmp13; + + tmp0 = tmp10 + tmp13; + tmp3 = tmp10 - tmp13; + tmp1 = tmp11 + tmp12; + tmp2 = tmp11 - tmp12; + + /* Odd part */ + + z13 = wsptr[5] + wsptr[3]; + z10 = wsptr[5] - wsptr[3]; + z11 = wsptr[1] + wsptr[7]; + z12 = wsptr[1] - wsptr[7]; + + tmp7 = z11 + z13; + tmp11 = (z11 - z13) * ((FAST_FLOAT)1.414213562); + + z5 = (z10 + z12) * ((FAST_FLOAT)1.847759065); /* 2*c2 */ + tmp10 = z5 - z12 * ((FAST_FLOAT)1.082392200); /* 2*(c2-c6) */ + tmp12 = z5 - z10 * ((FAST_FLOAT)2.613125930); /* 2*(c2+c6) */ + + tmp6 = tmp12 - tmp7; + tmp5 = tmp11 - tmp6; + tmp4 = tmp10 - tmp5; + + /* Final output stage: float->int conversion and range-limit */ + + outptr[0] = range_limit[((int)(tmp0 + tmp7)) & RANGE_MASK]; + outptr[7] = range_limit[((int)(tmp0 - tmp7)) & RANGE_MASK]; + outptr[1] = range_limit[((int)(tmp1 + tmp6)) & RANGE_MASK]; + outptr[6] = range_limit[((int)(tmp1 - tmp6)) & RANGE_MASK]; + outptr[2] = range_limit[((int)(tmp2 + tmp5)) & RANGE_MASK]; + outptr[5] = range_limit[((int)(tmp2 - tmp5)) & RANGE_MASK]; + outptr[3] = range_limit[((int)(tmp3 + tmp4)) & RANGE_MASK]; + outptr[4] = range_limit[((int)(tmp3 - tmp4)) & RANGE_MASK]; + + wsptr += DCTSIZE; /* advance pointer to next row */ + } +} + +#endif /* DCT_FLOAT_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jidctfst.c b/src/spice2x/external/libjpeg-turbo/jidctfst.c new file mode 100644 index 0000000..68119b9 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jidctfst.c @@ -0,0 +1,371 @@ +/* + * jidctfst.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1998, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains a fast, not so accurate integer implementation of the + * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine + * must also perform dequantization of the input coefficients. + * + * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT + * on each row (or vice versa, but it's more convenient to emit a row at + * a time). Direct algorithms are also available, but they are much more + * complex and seem not to be any faster when reduced to code. + * + * This implementation is based on Arai, Agui, and Nakajima's algorithm for + * scaled DCT. Their original paper (Trans. IEICE E-71(11):1095) is in + * Japanese, but the algorithm is described in the Pennebaker & Mitchell + * JPEG textbook (see REFERENCES section in file README.ijg). The following + * code is based directly on figure 4-8 in P&M. + * While an 8-point DCT cannot be done in less than 11 multiplies, it is + * possible to arrange the computation so that many of the multiplies are + * simple scalings of the final outputs. These multiplies can then be + * folded into the multiplications or divisions by the JPEG quantization + * table entries. The AA&N method leaves only 5 multiplies and 29 adds + * to be done in the DCT itself. + * The primary disadvantage of this method is that with fixed-point math, + * accuracy is lost due to imprecise representation of the scaled + * quantization values. The smaller the quantization table entry, the less + * precise the scaled value, so this implementation does worse with high- + * quality-setting files than with low-quality ones. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ + +#ifdef DCT_IFAST_SUPPORTED + + +/* + * This module is specialized to the case DCTSIZE = 8. + */ + +#if DCTSIZE != 8 + Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ +#endif + + +/* Scaling decisions are generally the same as in the LL&M algorithm; + * see jidctint.c for more details. However, we choose to descale + * (right shift) multiplication products as soon as they are formed, + * rather than carrying additional fractional bits into subsequent additions. + * This compromises accuracy slightly, but it lets us save a few shifts. + * More importantly, 16-bit arithmetic is then adequate (for 8-bit samples) + * everywhere except in the multiplications proper; this saves a good deal + * of work on 16-bit-int machines. + * + * The dequantized coefficients are not integers because the AA&N scaling + * factors have been incorporated. We represent them scaled up by PASS1_BITS, + * so that the first and second IDCT rounds have the same input scaling. + * For 8-bit samples, we choose IFAST_SCALE_BITS = PASS1_BITS so as to + * avoid a descaling shift; this compromises accuracy rather drastically + * for small quantization table entries, but it saves a lot of shifts. + * For 12-bit samples, there's no hope of using 16x16 multiplies anyway, + * so we use a much larger scaling factor to preserve accuracy. + * + * A final compromise is to represent the multiplicative constants to only + * 8 fractional bits, rather than 13. This saves some shifting work on some + * machines, and may also reduce the cost of multiplication (since there + * are fewer one-bits in the constants). + */ + +#if BITS_IN_JSAMPLE == 8 +#define CONST_BITS 8 +#define PASS1_BITS 2 +#else +#define CONST_BITS 8 +#define PASS1_BITS 1 /* lose a little precision to avoid overflow */ +#endif + +/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus + * causing a lot of useless floating-point operations at run time. + * To get around this we use the following pre-calculated constants. + * If you change CONST_BITS you may want to add appropriate values. + * (With a reasonable C compiler, you can just rely on the FIX() macro...) + */ + +#if CONST_BITS == 8 +#define FIX_1_082392200 ((JLONG)277) /* FIX(1.082392200) */ +#define FIX_1_414213562 ((JLONG)362) /* FIX(1.414213562) */ +#define FIX_1_847759065 ((JLONG)473) /* FIX(1.847759065) */ +#define FIX_2_613125930 ((JLONG)669) /* FIX(2.613125930) */ +#else +#define FIX_1_082392200 FIX(1.082392200) +#define FIX_1_414213562 FIX(1.414213562) +#define FIX_1_847759065 FIX(1.847759065) +#define FIX_2_613125930 FIX(2.613125930) +#endif + + +/* We can gain a little more speed, with a further compromise in accuracy, + * by omitting the addition in a descaling shift. This yields an incorrectly + * rounded result half the time... + */ + +#ifndef USE_ACCURATE_ROUNDING +#undef DESCALE +#define DESCALE(x, n) RIGHT_SHIFT(x, n) +#endif + + +/* Multiply a DCTELEM variable by an JLONG constant, and immediately + * descale to yield a DCTELEM result. + */ + +#define MULTIPLY(var, const) ((DCTELEM)DESCALE((var) * (const), CONST_BITS)) + + +/* Dequantize a coefficient by multiplying it by the multiplier-table + * entry; produce a DCTELEM result. For 8-bit data a 16x16->16 + * multiplication will do. For 12-bit data, the multiplier table is + * declared JLONG, so a 32-bit multiply will be used. + */ + +#if BITS_IN_JSAMPLE == 8 +#define DEQUANTIZE(coef, quantval) (((IFAST_MULT_TYPE)(coef)) * (quantval)) +#else +#define DEQUANTIZE(coef, quantval) \ + DESCALE((coef) * (quantval), IFAST_SCALE_BITS - PASS1_BITS) +#endif + + +/* Like DESCALE, but applies to a DCTELEM and produces an int. + * We assume that int right shift is unsigned if JLONG right shift is. + */ + +#ifdef RIGHT_SHIFT_IS_UNSIGNED +#define ISHIFT_TEMPS DCTELEM ishift_temp; +#if BITS_IN_JSAMPLE == 8 +#define DCTELEMBITS 16 /* DCTELEM may be 16 or 32 bits */ +#else +#define DCTELEMBITS 32 /* DCTELEM must be 32 bits */ +#endif +#define IRIGHT_SHIFT(x, shft) \ + ((ishift_temp = (x)) < 0 ? \ + (ishift_temp >> (shft)) | ((~((DCTELEM)0)) << (DCTELEMBITS - (shft))) : \ + (ishift_temp >> (shft))) +#else +#define ISHIFT_TEMPS +#define IRIGHT_SHIFT(x, shft) ((x) >> (shft)) +#endif + +#ifdef USE_ACCURATE_ROUNDING +#define IDESCALE(x, n) ((int)IRIGHT_SHIFT((x) + (1 << ((n) - 1)), n)) +#else +#define IDESCALE(x, n) ((int)IRIGHT_SHIFT(x, n)) +#endif + + +/* + * Perform dequantization and inverse DCT on one block of coefficients. + */ + +GLOBAL(void) +_jpeg_idct_ifast(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + DCTELEM tmp0, tmp1, tmp2, tmp3, tmp4, tmp5, tmp6, tmp7; + DCTELEM tmp10, tmp11, tmp12, tmp13; + DCTELEM z5, z10, z11, z12, z13; + JCOEFPTR inptr; + IFAST_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[DCTSIZE2]; /* buffers data between passes */ + SHIFT_TEMPS /* for DESCALE */ + ISHIFT_TEMPS /* for IDESCALE */ + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (IFAST_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = DCTSIZE; ctr > 0; ctr--) { + /* Due to quantization, we will usually find that many of the input + * coefficients are zero, especially the AC terms. We can exploit this + * by short-circuiting the IDCT calculation for any column in which all + * the AC terms are zero. In that case each output is equal to the + * DC coefficient (with scale factor as needed). + * With typical images and quantization tables, half or more of the + * column DCT calculations can be simplified this way. + */ + + if (inptr[DCTSIZE * 1] == 0 && inptr[DCTSIZE * 2] == 0 && + inptr[DCTSIZE * 3] == 0 && inptr[DCTSIZE * 4] == 0 && + inptr[DCTSIZE * 5] == 0 && inptr[DCTSIZE * 6] == 0 && + inptr[DCTSIZE * 7] == 0) { + /* AC terms all zero */ + int dcval = (int)DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + + wsptr[DCTSIZE * 0] = dcval; + wsptr[DCTSIZE * 1] = dcval; + wsptr[DCTSIZE * 2] = dcval; + wsptr[DCTSIZE * 3] = dcval; + wsptr[DCTSIZE * 4] = dcval; + wsptr[DCTSIZE * 5] = dcval; + wsptr[DCTSIZE * 6] = dcval; + wsptr[DCTSIZE * 7] = dcval; + + inptr++; /* advance pointers to next column */ + quantptr++; + wsptr++; + continue; + } + + /* Even part */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + tmp2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + tmp3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + tmp10 = tmp0 + tmp2; /* phase 3 */ + tmp11 = tmp0 - tmp2; + + tmp13 = tmp1 + tmp3; /* phases 5-3 */ + tmp12 = MULTIPLY(tmp1 - tmp3, FIX_1_414213562) - tmp13; /* 2*c4 */ + + tmp0 = tmp10 + tmp13; /* phase 2 */ + tmp3 = tmp10 - tmp13; + tmp1 = tmp11 + tmp12; + tmp2 = tmp11 - tmp12; + + /* Odd part */ + + tmp4 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + tmp5 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + tmp6 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + tmp7 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + z13 = tmp6 + tmp5; /* phase 6 */ + z10 = tmp6 - tmp5; + z11 = tmp4 + tmp7; + z12 = tmp4 - tmp7; + + tmp7 = z11 + z13; /* phase 5 */ + tmp11 = MULTIPLY(z11 - z13, FIX_1_414213562); /* 2*c4 */ + + z5 = MULTIPLY(z10 + z12, FIX_1_847759065); /* 2*c2 */ + tmp10 = MULTIPLY(z12, FIX_1_082392200) - z5; /* 2*(c2-c6) */ + tmp12 = MULTIPLY(z10, -FIX_2_613125930) + z5; /* -2*(c2+c6) */ + + tmp6 = tmp12 - tmp7; /* phase 2 */ + tmp5 = tmp11 - tmp6; + tmp4 = tmp10 + tmp5; + + wsptr[DCTSIZE * 0] = (int)(tmp0 + tmp7); + wsptr[DCTSIZE * 7] = (int)(tmp0 - tmp7); + wsptr[DCTSIZE * 1] = (int)(tmp1 + tmp6); + wsptr[DCTSIZE * 6] = (int)(tmp1 - tmp6); + wsptr[DCTSIZE * 2] = (int)(tmp2 + tmp5); + wsptr[DCTSIZE * 5] = (int)(tmp2 - tmp5); + wsptr[DCTSIZE * 4] = (int)(tmp3 + tmp4); + wsptr[DCTSIZE * 3] = (int)(tmp3 - tmp4); + + inptr++; /* advance pointers to next column */ + quantptr++; + wsptr++; + } + + /* Pass 2: process rows from work array, store into output array. */ + /* Note that we must descale the results by a factor of 8 == 2**3, */ + /* and also undo the PASS1_BITS scaling. */ + + wsptr = workspace; + for (ctr = 0; ctr < DCTSIZE; ctr++) { + outptr = output_buf[ctr] + output_col; + /* Rows of zeroes can be exploited in the same way as we did with columns. + * However, the column calculation has created many nonzero AC terms, so + * the simplification applies less often (typically 5% to 10% of the time). + * On machines with very fast multiplication, it's possible that the + * test takes more time than it's worth. In that case this section + * may be commented out. + */ + +#ifndef NO_ZERO_ROW_TEST + if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 && + wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) { + /* AC terms all zero */ + _JSAMPLE dcval = + range_limit[IDESCALE(wsptr[0], PASS1_BITS + 3) & RANGE_MASK]; + + outptr[0] = dcval; + outptr[1] = dcval; + outptr[2] = dcval; + outptr[3] = dcval; + outptr[4] = dcval; + outptr[5] = dcval; + outptr[6] = dcval; + outptr[7] = dcval; + + wsptr += DCTSIZE; /* advance pointer to next row */ + continue; + } +#endif + + /* Even part */ + + tmp10 = ((DCTELEM)wsptr[0] + (DCTELEM)wsptr[4]); + tmp11 = ((DCTELEM)wsptr[0] - (DCTELEM)wsptr[4]); + + tmp13 = ((DCTELEM)wsptr[2] + (DCTELEM)wsptr[6]); + tmp12 = + MULTIPLY((DCTELEM)wsptr[2] - (DCTELEM)wsptr[6], FIX_1_414213562) - tmp13; + + tmp0 = tmp10 + tmp13; + tmp3 = tmp10 - tmp13; + tmp1 = tmp11 + tmp12; + tmp2 = tmp11 - tmp12; + + /* Odd part */ + + z13 = (DCTELEM)wsptr[5] + (DCTELEM)wsptr[3]; + z10 = (DCTELEM)wsptr[5] - (DCTELEM)wsptr[3]; + z11 = (DCTELEM)wsptr[1] + (DCTELEM)wsptr[7]; + z12 = (DCTELEM)wsptr[1] - (DCTELEM)wsptr[7]; + + tmp7 = z11 + z13; /* phase 5 */ + tmp11 = MULTIPLY(z11 - z13, FIX_1_414213562); /* 2*c4 */ + + z5 = MULTIPLY(z10 + z12, FIX_1_847759065); /* 2*c2 */ + tmp10 = MULTIPLY(z12, FIX_1_082392200) - z5; /* 2*(c2-c6) */ + tmp12 = MULTIPLY(z10, -FIX_2_613125930) + z5; /* -2*(c2+c6) */ + + tmp6 = tmp12 - tmp7; /* phase 2 */ + tmp5 = tmp11 - tmp6; + tmp4 = tmp10 + tmp5; + + /* Final output stage: scale down by a factor of 8 and range-limit */ + + outptr[0] = + range_limit[IDESCALE(tmp0 + tmp7, PASS1_BITS + 3) & RANGE_MASK]; + outptr[7] = + range_limit[IDESCALE(tmp0 - tmp7, PASS1_BITS + 3) & RANGE_MASK]; + outptr[1] = + range_limit[IDESCALE(tmp1 + tmp6, PASS1_BITS + 3) & RANGE_MASK]; + outptr[6] = + range_limit[IDESCALE(tmp1 - tmp6, PASS1_BITS + 3) & RANGE_MASK]; + outptr[2] = + range_limit[IDESCALE(tmp2 + tmp5, PASS1_BITS + 3) & RANGE_MASK]; + outptr[5] = + range_limit[IDESCALE(tmp2 - tmp5, PASS1_BITS + 3) & RANGE_MASK]; + outptr[4] = + range_limit[IDESCALE(tmp3 + tmp4, PASS1_BITS + 3) & RANGE_MASK]; + outptr[3] = + range_limit[IDESCALE(tmp3 - tmp4, PASS1_BITS + 3) & RANGE_MASK]; + + wsptr += DCTSIZE; /* advance pointer to next row */ + } +} + +#endif /* DCT_IFAST_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jidctint.c b/src/spice2x/external/libjpeg-turbo/jidctint.c new file mode 100644 index 0000000..c58592d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jidctint.c @@ -0,0 +1,2627 @@ +/* + * jidctint.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1998, Thomas G. Lane. + * Modification developed 2002-2018 by Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains a slower but more accurate integer implementation of the + * inverse DCT (Discrete Cosine Transform). In the IJG code, this routine + * must also perform dequantization of the input coefficients. + * + * A 2-D IDCT can be done by 1-D IDCT on each column followed by 1-D IDCT + * on each row (or vice versa, but it's more convenient to emit a row at + * a time). Direct algorithms are also available, but they are much more + * complex and seem not to be any faster when reduced to code. + * + * This implementation is based on an algorithm described in + * C. Loeffler, A. Ligtenberg and G. Moschytz, "Practical Fast 1-D DCT + * Algorithms with 11 Multiplications", Proc. Int'l. Conf. on Acoustics, + * Speech, and Signal Processing 1989 (ICASSP '89), pp. 988-991. + * The primary algorithm described there uses 11 multiplies and 29 adds. + * We use their alternate method with 12 multiplies and 32 adds. + * The advantage of this method is that no data path contains more than one + * multiplication; this allows a very simple and accurate implementation in + * scaled fixed-point arithmetic, with a minimal number of shifts. + * + * We also provide IDCT routines with various output sample block sizes for + * direct resolution reduction or enlargement without additional resampling: + * NxN (N=1...16) pixels for one 8x8 input DCT block. + * + * For N<8 we simply take the corresponding low-frequency coefficients of + * the 8x8 input DCT block and apply an NxN point IDCT on the sub-block + * to yield the downscaled outputs. + * This can be seen as direct low-pass downsampling from the DCT domain + * point of view rather than the usual spatial domain point of view, + * yielding significant computational savings and results at least + * as good as common bilinear (averaging) spatial downsampling. + * + * For N>8 we apply a partial NxN IDCT on the 8 input coefficients as + * lower frequencies and higher frequencies assumed to be zero. + * It turns out that the computational effort is similar to the 8x8 IDCT + * regarding the output size. + * Furthermore, the scaling and descaling is the same for all IDCT sizes. + * + * CAUTION: We rely on the FIX() macro except for the N=1,2,4,8 cases + * since there would be too many additional constants to pre-calculate. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ + +#ifdef DCT_ISLOW_SUPPORTED + + +/* + * This module is specialized to the case DCTSIZE = 8. + */ + +#if DCTSIZE != 8 + Sorry, this code only copes with 8x8 DCT blocks. /* deliberate syntax err */ +#endif + + +/* + * The poop on this scaling stuff is as follows: + * + * Each 1-D IDCT step produces outputs which are a factor of sqrt(N) + * larger than the true IDCT outputs. The final outputs are therefore + * a factor of N larger than desired; since N=8 this can be cured by + * a simple right shift at the end of the algorithm. The advantage of + * this arrangement is that we save two multiplications per 1-D IDCT, + * because the y0 and y4 inputs need not be divided by sqrt(N). + * + * We have to do addition and subtraction of the integer inputs, which + * is no problem, and multiplication by fractional constants, which is + * a problem to do in integer arithmetic. We multiply all the constants + * by CONST_SCALE and convert them to integer constants (thus retaining + * CONST_BITS bits of precision in the constants). After doing a + * multiplication we have to divide the product by CONST_SCALE, with proper + * rounding, to produce the correct output. This division can be done + * cheaply as a right shift of CONST_BITS bits. We postpone shifting + * as long as possible so that partial sums can be added together with + * full fractional precision. + * + * The outputs of the first pass are scaled up by PASS1_BITS bits so that + * they are represented to better-than-integral precision. These outputs + * require BITS_IN_JSAMPLE + PASS1_BITS + 3 bits; this fits in a 16-bit word + * with the recommended scaling. (To scale up 12-bit sample data further, an + * intermediate JLONG array would be needed.) + * + * To avoid overflow of the 32-bit intermediate results in pass 2, we must + * have BITS_IN_JSAMPLE + CONST_BITS + PASS1_BITS <= 26. Error analysis + * shows that the values given below are the most effective. + */ + +#if BITS_IN_JSAMPLE == 8 +#define CONST_BITS 13 +#define PASS1_BITS 2 +#else +#define CONST_BITS 13 +#define PASS1_BITS 1 /* lose a little precision to avoid overflow */ +#endif + +/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus + * causing a lot of useless floating-point operations at run time. + * To get around this we use the following pre-calculated constants. + * If you change CONST_BITS you may want to add appropriate values. + * (With a reasonable C compiler, you can just rely on the FIX() macro...) + */ + +#if CONST_BITS == 13 +#define FIX_0_298631336 ((JLONG)2446) /* FIX(0.298631336) */ +#define FIX_0_390180644 ((JLONG)3196) /* FIX(0.390180644) */ +#define FIX_0_541196100 ((JLONG)4433) /* FIX(0.541196100) */ +#define FIX_0_765366865 ((JLONG)6270) /* FIX(0.765366865) */ +#define FIX_0_899976223 ((JLONG)7373) /* FIX(0.899976223) */ +#define FIX_1_175875602 ((JLONG)9633) /* FIX(1.175875602) */ +#define FIX_1_501321110 ((JLONG)12299) /* FIX(1.501321110) */ +#define FIX_1_847759065 ((JLONG)15137) /* FIX(1.847759065) */ +#define FIX_1_961570560 ((JLONG)16069) /* FIX(1.961570560) */ +#define FIX_2_053119869 ((JLONG)16819) /* FIX(2.053119869) */ +#define FIX_2_562915447 ((JLONG)20995) /* FIX(2.562915447) */ +#define FIX_3_072711026 ((JLONG)25172) /* FIX(3.072711026) */ +#else +#define FIX_0_298631336 FIX(0.298631336) +#define FIX_0_390180644 FIX(0.390180644) +#define FIX_0_541196100 FIX(0.541196100) +#define FIX_0_765366865 FIX(0.765366865) +#define FIX_0_899976223 FIX(0.899976223) +#define FIX_1_175875602 FIX(1.175875602) +#define FIX_1_501321110 FIX(1.501321110) +#define FIX_1_847759065 FIX(1.847759065) +#define FIX_1_961570560 FIX(1.961570560) +#define FIX_2_053119869 FIX(2.053119869) +#define FIX_2_562915447 FIX(2.562915447) +#define FIX_3_072711026 FIX(3.072711026) +#endif + + +/* Multiply an JLONG variable by an JLONG constant to yield an JLONG result. + * For 8-bit samples with the recommended scaling, all the variable + * and constant values involved are no more than 16 bits wide, so a + * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. + * For 12-bit samples, a full 32-bit multiplication will be needed. + */ + +#if BITS_IN_JSAMPLE == 8 +#define MULTIPLY(var, const) MULTIPLY16C16(var, const) +#else +#define MULTIPLY(var, const) ((var) * (const)) +#endif + + +/* Dequantize a coefficient by multiplying it by the multiplier-table + * entry; produce an int result. In this module, both inputs and result + * are 16 bits or less, so either int or short multiply will work. + */ + +#define DEQUANTIZE(coef, quantval) (((ISLOW_MULT_TYPE)(coef)) * (quantval)) + + +/* + * Perform dequantization and inverse DCT on one block of coefficients. + */ + +GLOBAL(void) +_jpeg_idct_islow(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp1, tmp2, tmp3; + JLONG tmp10, tmp11, tmp12, tmp13; + JLONG z1, z2, z3, z4, z5; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[DCTSIZE2]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + /* Note results are scaled up by sqrt(8) compared to a true IDCT; */ + /* furthermore, we scale the results by 2**PASS1_BITS. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = DCTSIZE; ctr > 0; ctr--) { + /* Due to quantization, we will usually find that many of the input + * coefficients are zero, especially the AC terms. We can exploit this + * by short-circuiting the IDCT calculation for any column in which all + * the AC terms are zero. In that case each output is equal to the + * DC coefficient (with scale factor as needed). + * With typical images and quantization tables, half or more of the + * column DCT calculations can be simplified this way. + */ + + if (inptr[DCTSIZE * 1] == 0 && inptr[DCTSIZE * 2] == 0 && + inptr[DCTSIZE * 3] == 0 && inptr[DCTSIZE * 4] == 0 && + inptr[DCTSIZE * 5] == 0 && inptr[DCTSIZE * 6] == 0 && + inptr[DCTSIZE * 7] == 0) { + /* AC terms all zero */ + int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE * 0], + quantptr[DCTSIZE * 0]), PASS1_BITS); + + wsptr[DCTSIZE * 0] = dcval; + wsptr[DCTSIZE * 1] = dcval; + wsptr[DCTSIZE * 2] = dcval; + wsptr[DCTSIZE * 3] = dcval; + wsptr[DCTSIZE * 4] = dcval; + wsptr[DCTSIZE * 5] = dcval; + wsptr[DCTSIZE * 6] = dcval; + wsptr[DCTSIZE * 7] = dcval; + + inptr++; /* advance pointers to next column */ + quantptr++; + wsptr++; + continue; + } + + /* Even part: reverse the even part of the forward DCT. */ + /* The rotator is sqrt(2)*c(-6). */ + + z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + z1 = MULTIPLY(z2 + z3, FIX_0_541196100); + tmp2 = z1 + MULTIPLY(z3, -FIX_1_847759065); + tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865); + + z2 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + + tmp0 = LEFT_SHIFT(z2 + z3, CONST_BITS); + tmp1 = LEFT_SHIFT(z2 - z3, CONST_BITS); + + tmp10 = tmp0 + tmp3; + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + /* Odd part per figure 8; the matrix is unitary and hence its + * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively. + */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + tmp1 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + tmp2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + tmp3 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + + z1 = tmp0 + tmp3; + z2 = tmp1 + tmp2; + z3 = tmp0 + tmp2; + z4 = tmp1 + tmp3; + z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */ + + tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */ + tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */ + tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */ + tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */ + z1 = MULTIPLY(z1, -FIX_0_899976223); /* sqrt(2) * ( c7-c3) */ + z2 = MULTIPLY(z2, -FIX_2_562915447); /* sqrt(2) * (-c1-c3) */ + z3 = MULTIPLY(z3, -FIX_1_961570560); /* sqrt(2) * (-c3-c5) */ + z4 = MULTIPLY(z4, -FIX_0_390180644); /* sqrt(2) * ( c5-c3) */ + + z3 += z5; + z4 += z5; + + tmp0 += z1 + z3; + tmp1 += z2 + z4; + tmp2 += z2 + z3; + tmp3 += z1 + z4; + + /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */ + + wsptr[DCTSIZE * 0] = (int)DESCALE(tmp10 + tmp3, CONST_BITS - PASS1_BITS); + wsptr[DCTSIZE * 7] = (int)DESCALE(tmp10 - tmp3, CONST_BITS - PASS1_BITS); + wsptr[DCTSIZE * 1] = (int)DESCALE(tmp11 + tmp2, CONST_BITS - PASS1_BITS); + wsptr[DCTSIZE * 6] = (int)DESCALE(tmp11 - tmp2, CONST_BITS - PASS1_BITS); + wsptr[DCTSIZE * 2] = (int)DESCALE(tmp12 + tmp1, CONST_BITS - PASS1_BITS); + wsptr[DCTSIZE * 5] = (int)DESCALE(tmp12 - tmp1, CONST_BITS - PASS1_BITS); + wsptr[DCTSIZE * 3] = (int)DESCALE(tmp13 + tmp0, CONST_BITS - PASS1_BITS); + wsptr[DCTSIZE * 4] = (int)DESCALE(tmp13 - tmp0, CONST_BITS - PASS1_BITS); + + inptr++; /* advance pointers to next column */ + quantptr++; + wsptr++; + } + + /* Pass 2: process rows from work array, store into output array. */ + /* Note that we must descale the results by a factor of 8 == 2**3, */ + /* and also undo the PASS1_BITS scaling. */ + + wsptr = workspace; + for (ctr = 0; ctr < DCTSIZE; ctr++) { + outptr = output_buf[ctr] + output_col; + /* Rows of zeroes can be exploited in the same way as we did with columns. + * However, the column calculation has created many nonzero AC terms, so + * the simplification applies less often (typically 5% to 10% of the time). + * On machines with very fast multiplication, it's possible that the + * test takes more time than it's worth. In that case this section + * may be commented out. + */ + +#ifndef NO_ZERO_ROW_TEST + if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && wsptr[4] == 0 && + wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) { + /* AC terms all zero */ + _JSAMPLE dcval = range_limit[(int)DESCALE((JLONG)wsptr[0], + PASS1_BITS + 3) & RANGE_MASK]; + + outptr[0] = dcval; + outptr[1] = dcval; + outptr[2] = dcval; + outptr[3] = dcval; + outptr[4] = dcval; + outptr[5] = dcval; + outptr[6] = dcval; + outptr[7] = dcval; + + wsptr += DCTSIZE; /* advance pointer to next row */ + continue; + } +#endif + + /* Even part: reverse the even part of the forward DCT. */ + /* The rotator is sqrt(2)*c(-6). */ + + z2 = (JLONG)wsptr[2]; + z3 = (JLONG)wsptr[6]; + + z1 = MULTIPLY(z2 + z3, FIX_0_541196100); + tmp2 = z1 + MULTIPLY(z3, -FIX_1_847759065); + tmp3 = z1 + MULTIPLY(z2, FIX_0_765366865); + + tmp0 = LEFT_SHIFT((JLONG)wsptr[0] + (JLONG)wsptr[4], CONST_BITS); + tmp1 = LEFT_SHIFT((JLONG)wsptr[0] - (JLONG)wsptr[4], CONST_BITS); + + tmp10 = tmp0 + tmp3; + tmp13 = tmp0 - tmp3; + tmp11 = tmp1 + tmp2; + tmp12 = tmp1 - tmp2; + + /* Odd part per figure 8; the matrix is unitary and hence its + * transpose is its inverse. i0..i3 are y7,y5,y3,y1 respectively. + */ + + tmp0 = (JLONG)wsptr[7]; + tmp1 = (JLONG)wsptr[5]; + tmp2 = (JLONG)wsptr[3]; + tmp3 = (JLONG)wsptr[1]; + + z1 = tmp0 + tmp3; + z2 = tmp1 + tmp2; + z3 = tmp0 + tmp2; + z4 = tmp1 + tmp3; + z5 = MULTIPLY(z3 + z4, FIX_1_175875602); /* sqrt(2) * c3 */ + + tmp0 = MULTIPLY(tmp0, FIX_0_298631336); /* sqrt(2) * (-c1+c3+c5-c7) */ + tmp1 = MULTIPLY(tmp1, FIX_2_053119869); /* sqrt(2) * ( c1+c3-c5+c7) */ + tmp2 = MULTIPLY(tmp2, FIX_3_072711026); /* sqrt(2) * ( c1+c3+c5-c7) */ + tmp3 = MULTIPLY(tmp3, FIX_1_501321110); /* sqrt(2) * ( c1+c3-c5-c7) */ + z1 = MULTIPLY(z1, -FIX_0_899976223); /* sqrt(2) * ( c7-c3) */ + z2 = MULTIPLY(z2, -FIX_2_562915447); /* sqrt(2) * (-c1-c3) */ + z3 = MULTIPLY(z3, -FIX_1_961570560); /* sqrt(2) * (-c3-c5) */ + z4 = MULTIPLY(z4, -FIX_0_390180644); /* sqrt(2) * ( c5-c3) */ + + z3 += z5; + z4 += z5; + + tmp0 += z1 + z3; + tmp1 += z2 + z4; + tmp2 += z2 + z3; + tmp3 += z1 + z4; + + /* Final output stage: inputs are tmp10..tmp13, tmp0..tmp3 */ + + outptr[0] = range_limit[(int)DESCALE(tmp10 + tmp3, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)DESCALE(tmp10 - tmp3, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)DESCALE(tmp11 + tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)DESCALE(tmp11 - tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)DESCALE(tmp12 + tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)DESCALE(tmp12 - tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)DESCALE(tmp13 + tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)DESCALE(tmp13 - tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += DCTSIZE; /* advance pointer to next row */ + } +} + +#ifdef IDCT_SCALING_SUPPORTED + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a reduced-size 7x7 output block. + * + * Optimized algorithm with 12 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/14). + */ + +GLOBAL(void) +_jpeg_idct_7x7(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp1, tmp2, tmp10, tmp11, tmp12, tmp13; + JLONG z1, z2, z3; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[7 * 7]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 7; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + tmp13 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp13 = LEFT_SHIFT(tmp13, CONST_BITS); + /* Add fudge factor here for final descale. */ + tmp13 += ONE << (CONST_BITS - PASS1_BITS - 1); + + z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */ + tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */ + tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */ + tmp0 = z1 + z3; + z2 -= tmp0; + tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */ + tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */ + tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */ + tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */ + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + + tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */ + tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */ + tmp0 = tmp1 - tmp2; + tmp1 += tmp2; + tmp2 = MULTIPLY(z2 + z3, -FIX(1.378756276)); /* -c1 */ + tmp1 += tmp2; + z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */ + tmp0 += z2; + tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */ + + /* Final output stage */ + + wsptr[7 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS); + wsptr[7 * 6] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS); + wsptr[7 * 1] = (int)RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS - PASS1_BITS); + wsptr[7 * 5] = (int)RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS - PASS1_BITS); + wsptr[7 * 2] = (int)RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS - PASS1_BITS); + wsptr[7 * 4] = (int)RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS - PASS1_BITS); + wsptr[7 * 3] = (int)RIGHT_SHIFT(tmp13, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 7 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 7; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + tmp13 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + tmp13 = LEFT_SHIFT(tmp13, CONST_BITS); + + z1 = (JLONG)wsptr[2]; + z2 = (JLONG)wsptr[4]; + z3 = (JLONG)wsptr[6]; + + tmp10 = MULTIPLY(z2 - z3, FIX(0.881747734)); /* c4 */ + tmp12 = MULTIPLY(z1 - z2, FIX(0.314692123)); /* c6 */ + tmp11 = tmp10 + tmp12 + tmp13 - MULTIPLY(z2, FIX(1.841218003)); /* c2+c4-c6 */ + tmp0 = z1 + z3; + z2 -= tmp0; + tmp0 = MULTIPLY(tmp0, FIX(1.274162392)) + tmp13; /* c2 */ + tmp10 += tmp0 - MULTIPLY(z3, FIX(0.077722536)); /* c2-c4-c6 */ + tmp12 += tmp0 - MULTIPLY(z1, FIX(2.470602249)); /* c2+c4+c6 */ + tmp13 += MULTIPLY(z2, FIX(1.414213562)); /* c0 */ + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + + tmp1 = MULTIPLY(z1 + z2, FIX(0.935414347)); /* (c3+c1-c5)/2 */ + tmp2 = MULTIPLY(z1 - z2, FIX(0.170262339)); /* (c3+c5-c1)/2 */ + tmp0 = tmp1 - tmp2; + tmp1 += tmp2; + tmp2 = MULTIPLY(z2 + z3, -FIX(1.378756276)); /* -c1 */ + tmp1 += tmp2; + z2 = MULTIPLY(z1 + z3, FIX(0.613604268)); /* c5 */ + tmp0 += z2; + tmp2 += z2 + MULTIPLY(z3, FIX(1.870828693)); /* c3+c1-c5 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12 + tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp12 - tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 7; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a reduced-size 6x6 output block. + * + * Optimized algorithm with 3 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/12). + */ + +GLOBAL(void) +_jpeg_idct_6x6(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp1, tmp2, tmp10, tmp11, tmp12; + JLONG z1, z2, z3; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[6 * 6]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 6; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + /* Add fudge factor here for final descale. */ + tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1); + tmp2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */ + tmp1 = tmp0 + tmp10; + tmp11 = RIGHT_SHIFT(tmp0 - tmp10 - tmp10, CONST_BITS - PASS1_BITS); + tmp10 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */ + tmp10 = tmp1 + tmp0; + tmp12 = tmp1 - tmp0; + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */ + tmp0 = tmp1 + LEFT_SHIFT(z1 + z2, CONST_BITS); + tmp2 = tmp1 + LEFT_SHIFT(z3 - z2, CONST_BITS); + tmp1 = LEFT_SHIFT(z1 - z2 - z3, PASS1_BITS); + + /* Final output stage */ + + wsptr[6 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS); + wsptr[6 * 5] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS); + wsptr[6 * 1] = (int)(tmp11 + tmp1); + wsptr[6 * 4] = (int)(tmp11 - tmp1); + wsptr[6 * 2] = (int)RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS - PASS1_BITS); + wsptr[6 * 3] = (int)RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 6 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 6; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + tmp2 = (JLONG)wsptr[4]; + tmp10 = MULTIPLY(tmp2, FIX(0.707106781)); /* c4 */ + tmp1 = tmp0 + tmp10; + tmp11 = tmp0 - tmp10 - tmp10; + tmp10 = (JLONG)wsptr[2]; + tmp0 = MULTIPLY(tmp10, FIX(1.224744871)); /* c2 */ + tmp10 = tmp1 + tmp0; + tmp12 = tmp1 - tmp0; + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + tmp1 = MULTIPLY(z1 + z3, FIX(0.366025404)); /* c5 */ + tmp0 = tmp1 + LEFT_SHIFT(z1 + z2, CONST_BITS); + tmp2 = tmp1 + LEFT_SHIFT(z3 - z2, CONST_BITS); + tmp1 = LEFT_SHIFT(z1 - z2 - z3, CONST_BITS); + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12 + tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp12 - tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 6; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a reduced-size 5x5 output block. + * + * Optimized algorithm with 5 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/10). + */ + +GLOBAL(void) +_jpeg_idct_5x5(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp1, tmp10, tmp11, tmp12; + JLONG z1, z2, z3; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[5 * 5]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 5; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + tmp12 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp12 = LEFT_SHIFT(tmp12, CONST_BITS); + /* Add fudge factor here for final descale. */ + tmp12 += ONE << (CONST_BITS - PASS1_BITS - 1); + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + tmp1 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */ + z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */ + z3 = tmp12 + z2; + tmp10 = z3 + z1; + tmp11 = z3 - z1; + tmp12 -= LEFT_SHIFT(z2, 2); + + /* Odd part */ + + z2 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */ + tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */ + tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */ + + /* Final output stage */ + + wsptr[5 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS); + wsptr[5 * 4] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS); + wsptr[5 * 1] = (int)RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS - PASS1_BITS); + wsptr[5 * 3] = (int)RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS - PASS1_BITS); + wsptr[5 * 2] = (int)RIGHT_SHIFT(tmp12, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 5 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 5; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + tmp12 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + tmp12 = LEFT_SHIFT(tmp12, CONST_BITS); + tmp0 = (JLONG)wsptr[2]; + tmp1 = (JLONG)wsptr[4]; + z1 = MULTIPLY(tmp0 + tmp1, FIX(0.790569415)); /* (c2+c4)/2 */ + z2 = MULTIPLY(tmp0 - tmp1, FIX(0.353553391)); /* (c2-c4)/2 */ + z3 = tmp12 + z2; + tmp10 = z3 + z1; + tmp11 = z3 - z1; + tmp12 -= LEFT_SHIFT(z2, 2); + + /* Odd part */ + + z2 = (JLONG)wsptr[1]; + z3 = (JLONG)wsptr[3]; + + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c3 */ + tmp0 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c1-c3 */ + tmp1 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c1+c3 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 5; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a reduced-size 3x3 output block. + * + * Optimized algorithm with 2 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/6). + */ + +GLOBAL(void) +_jpeg_idct_3x3(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp2, tmp10, tmp12; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[3 * 3]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 3; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + /* Add fudge factor here for final descale. */ + tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1); + tmp2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */ + tmp10 = tmp0 + tmp12; + tmp2 = tmp0 - tmp12 - tmp12; + + /* Odd part */ + + tmp12 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */ + + /* Final output stage */ + + wsptr[3 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS); + wsptr[3 * 2] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS); + wsptr[3 * 1] = (int)RIGHT_SHIFT(tmp2, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 3 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 3; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + tmp2 = (JLONG)wsptr[2]; + tmp12 = MULTIPLY(tmp2, FIX(0.707106781)); /* c2 */ + tmp10 = tmp0 + tmp12; + tmp2 = tmp0 - tmp12 - tmp12; + + /* Odd part */ + + tmp12 = (JLONG)wsptr[1]; + tmp0 = MULTIPLY(tmp12, FIX(1.224744871)); /* c1 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 3; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a 9x9 output block. + * + * Optimized algorithm with 10 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/18). + */ + +GLOBAL(void) +_jpeg_idct_9x9(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13, tmp14; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 9]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + /* Add fudge factor here for final descale. */ + tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1); + + z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */ + tmp1 = tmp0 + tmp3; + tmp2 = tmp0 - tmp3 - tmp3; + + tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */ + tmp11 = tmp2 + tmp0; + tmp14 = tmp2 - tmp0 - tmp0; + + tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */ + tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */ + tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */ + + tmp10 = tmp1 + tmp0 - tmp3; + tmp12 = tmp1 - tmp0 + tmp2; + tmp13 = tmp1 - tmp2 + tmp3; + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + z2 = MULTIPLY(z2, -FIX(1.224744871)); /* -c3 */ + + tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */ + tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */ + tmp0 = tmp2 + tmp3 - z2; + tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */ + tmp2 += z2 - tmp1; + tmp3 += z2 + tmp1; + tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */ + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp10 + tmp0, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp10 - tmp0, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp11 + tmp1, CONST_BITS - PASS1_BITS); + wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp11 - tmp1, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp12 + tmp2, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp12 - tmp2, CONST_BITS - PASS1_BITS); + wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp13 + tmp3, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp13 - tmp3, CONST_BITS - PASS1_BITS); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp14, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 9 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 9; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + + z1 = (JLONG)wsptr[2]; + z2 = (JLONG)wsptr[4]; + z3 = (JLONG)wsptr[6]; + + tmp3 = MULTIPLY(z3, FIX(0.707106781)); /* c6 */ + tmp1 = tmp0 + tmp3; + tmp2 = tmp0 - tmp3 - tmp3; + + tmp0 = MULTIPLY(z1 - z2, FIX(0.707106781)); /* c6 */ + tmp11 = tmp2 + tmp0; + tmp14 = tmp2 - tmp0 - tmp0; + + tmp0 = MULTIPLY(z1 + z2, FIX(1.328926049)); /* c2 */ + tmp2 = MULTIPLY(z1, FIX(1.083350441)); /* c4 */ + tmp3 = MULTIPLY(z2, FIX(0.245575608)); /* c8 */ + + tmp10 = tmp1 + tmp0 - tmp3; + tmp12 = tmp1 - tmp0 + tmp2; + tmp13 = tmp1 - tmp2 + tmp3; + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + z4 = (JLONG)wsptr[7]; + + z2 = MULTIPLY(z2, -FIX(1.224744871)); /* -c3 */ + + tmp2 = MULTIPLY(z1 + z3, FIX(0.909038955)); /* c5 */ + tmp3 = MULTIPLY(z1 + z4, FIX(0.483689525)); /* c7 */ + tmp0 = tmp2 + tmp3 - z2; + tmp1 = MULTIPLY(z3 - z4, FIX(1.392728481)); /* c1 */ + tmp2 += z2 - tmp1; + tmp3 += z2 + tmp1; + tmp1 = MULTIPLY(z1 - z3 - z4, FIX(1.224744871)); /* c3 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp10 + tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp10 - tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp11 + tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp11 - tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp12 + tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp12 - tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp13 + tmp3, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp13 - tmp3, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a 10x10 output block. + * + * Optimized algorithm with 12 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/20). + */ + +GLOBAL(void) +_jpeg_idct_10x10(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp10, tmp11, tmp12, tmp13, tmp14; + JLONG tmp20, tmp21, tmp22, tmp23, tmp24; + JLONG z1, z2, z3, z4, z5; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 10]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + z3 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + z3 = LEFT_SHIFT(z3, CONST_BITS); + /* Add fudge factor here for final descale. */ + z3 += ONE << (CONST_BITS - PASS1_BITS - 1); + z4 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */ + z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */ + tmp10 = z3 + z1; + tmp11 = z3 - z2; + + tmp22 = RIGHT_SHIFT(z3 - LEFT_SHIFT(z1 - z2, 1), + CONST_BITS - PASS1_BITS); /* c0 = (c4-c8)*2 */ + + z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */ + tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */ + tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */ + + tmp20 = tmp10 + tmp12; + tmp24 = tmp10 - tmp12; + tmp21 = tmp11 + tmp13; + tmp23 = tmp11 - tmp13; + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + tmp11 = z2 + z4; + tmp13 = z2 - z4; + + tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */ + z5 = LEFT_SHIFT(z3, CONST_BITS); + + z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */ + z4 = z5 + tmp12; + + tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */ + tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */ + + z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */ + z4 = z5 - tmp12 - LEFT_SHIFT(tmp13, CONST_BITS - 1); + + tmp12 = LEFT_SHIFT(z1 - tmp13 - z3, PASS1_BITS); + + tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */ + tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */ + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)(tmp22 + tmp12); + wsptr[8 * 7] = (int)(tmp22 - tmp12); + wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 10 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 10; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + z3 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + z3 = LEFT_SHIFT(z3, CONST_BITS); + z4 = (JLONG)wsptr[4]; + z1 = MULTIPLY(z4, FIX(1.144122806)); /* c4 */ + z2 = MULTIPLY(z4, FIX(0.437016024)); /* c8 */ + tmp10 = z3 + z1; + tmp11 = z3 - z2; + + tmp22 = z3 - LEFT_SHIFT(z1 - z2, 1); /* c0 = (c4-c8)*2 */ + + z2 = (JLONG)wsptr[2]; + z3 = (JLONG)wsptr[6]; + + z1 = MULTIPLY(z2 + z3, FIX(0.831253876)); /* c6 */ + tmp12 = z1 + MULTIPLY(z2, FIX(0.513743148)); /* c2-c6 */ + tmp13 = z1 - MULTIPLY(z3, FIX(2.176250899)); /* c2+c6 */ + + tmp20 = tmp10 + tmp12; + tmp24 = tmp10 - tmp12; + tmp21 = tmp11 + tmp13; + tmp23 = tmp11 - tmp13; + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + z3 = LEFT_SHIFT(z3, CONST_BITS); + z4 = (JLONG)wsptr[7]; + + tmp11 = z2 + z4; + tmp13 = z2 - z4; + + tmp12 = MULTIPLY(tmp13, FIX(0.309016994)); /* (c3-c7)/2 */ + + z2 = MULTIPLY(tmp11, FIX(0.951056516)); /* (c3+c7)/2 */ + z4 = z3 + tmp12; + + tmp10 = MULTIPLY(z1, FIX(1.396802247)) + z2 + z4; /* c1 */ + tmp14 = MULTIPLY(z1, FIX(0.221231742)) - z2 + z4; /* c9 */ + + z2 = MULTIPLY(tmp11, FIX(0.587785252)); /* (c1-c9)/2 */ + z4 = z3 - tmp12 - LEFT_SHIFT(tmp13, CONST_BITS - 1); + + tmp12 = LEFT_SHIFT(z1 - tmp13, CONST_BITS) - z3; + + tmp11 = MULTIPLY(z1, FIX(1.260073511)) - z2 - z4; /* c3 */ + tmp13 = MULTIPLY(z1, FIX(0.642039522)) - z2 + z4; /* c7 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing an 11x11 output block. + * + * Optimized algorithm with 24 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/22). + */ + +GLOBAL(void) +_jpeg_idct_11x11(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp10, tmp11, tmp12, tmp13, tmp14; + JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 11]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + tmp10 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp10 = LEFT_SHIFT(tmp10, CONST_BITS); + /* Add fudge factor here for final descale. */ + tmp10 += ONE << (CONST_BITS - PASS1_BITS - 1); + + z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */ + tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */ + z4 = z1 + z3; + tmp24 = MULTIPLY(z4, -FIX(1.155664402)); /* -(c2-c10) */ + z4 -= z2; + tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */ + tmp21 = tmp20 + tmp23 + tmp25 - + MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */ + tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */ + tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */ + tmp24 += tmp25; + tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */ + tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */ + MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */ + tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */ + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + tmp11 = z1 + z2; + tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */ + tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */ + tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */ + tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */ + tmp10 = tmp11 + tmp12 + tmp13 - + MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */ + z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */ + tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */ + tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */ + z1 = MULTIPLY(z2 + z4, -FIX(1.798248910)); /* -(c1+c9) */ + tmp11 += z1; + tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */ + tmp14 += MULTIPLY(z2, -FIX(1.467221301)) + /* -(c5+c9) */ + MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */ + MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */ + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 11 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 11; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + tmp10 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + tmp10 = LEFT_SHIFT(tmp10, CONST_BITS); + + z1 = (JLONG)wsptr[2]; + z2 = (JLONG)wsptr[4]; + z3 = (JLONG)wsptr[6]; + + tmp20 = MULTIPLY(z2 - z3, FIX(2.546640132)); /* c2+c4 */ + tmp23 = MULTIPLY(z2 - z1, FIX(0.430815045)); /* c2-c6 */ + z4 = z1 + z3; + tmp24 = MULTIPLY(z4, -FIX(1.155664402)); /* -(c2-c10) */ + z4 -= z2; + tmp25 = tmp10 + MULTIPLY(z4, FIX(1.356927976)); /* c2 */ + tmp21 = tmp20 + tmp23 + tmp25 - + MULTIPLY(z2, FIX(1.821790775)); /* c2+c4+c10-c6 */ + tmp20 += tmp25 + MULTIPLY(z3, FIX(2.115825087)); /* c4+c6 */ + tmp23 += tmp25 - MULTIPLY(z1, FIX(1.513598477)); /* c6+c8 */ + tmp24 += tmp25; + tmp22 = tmp24 - MULTIPLY(z3, FIX(0.788749120)); /* c8+c10 */ + tmp24 += MULTIPLY(z2, FIX(1.944413522)) - /* c2+c8 */ + MULTIPLY(z1, FIX(1.390975730)); /* c4+c10 */ + tmp25 = tmp10 - MULTIPLY(z4, FIX(1.414213562)); /* c0 */ + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + z4 = (JLONG)wsptr[7]; + + tmp11 = z1 + z2; + tmp14 = MULTIPLY(tmp11 + z3 + z4, FIX(0.398430003)); /* c9 */ + tmp11 = MULTIPLY(tmp11, FIX(0.887983902)); /* c3-c9 */ + tmp12 = MULTIPLY(z1 + z3, FIX(0.670361295)); /* c5-c9 */ + tmp13 = tmp14 + MULTIPLY(z1 + z4, FIX(0.366151574)); /* c7-c9 */ + tmp10 = tmp11 + tmp12 + tmp13 - + MULTIPLY(z1, FIX(0.923107866)); /* c7+c5+c3-c1-2*c9 */ + z1 = tmp14 - MULTIPLY(z2 + z3, FIX(1.163011579)); /* c7+c9 */ + tmp11 += z1 + MULTIPLY(z2, FIX(2.073276588)); /* c1+c7+3*c9-c3 */ + tmp12 += z1 - MULTIPLY(z3, FIX(1.192193623)); /* c3+c5-c7-c9 */ + z1 = MULTIPLY(z2 + z4, -FIX(1.798248910)); /* -(c1+c9) */ + tmp11 += z1; + tmp13 += z1 + MULTIPLY(z4, FIX(2.102458632)); /* c1+c5+c9-c7 */ + tmp14 += MULTIPLY(z2, -FIX(1.467221301)) + /* -(c5+c9) */ + MULTIPLY(z3, FIX(1.001388905)) - /* c1-c9 */ + MULTIPLY(z4, FIX(1.684843907)); /* c3+c9 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a 12x12 output block. + * + * Optimized algorithm with 15 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/24). + */ + +GLOBAL(void) +_jpeg_idct_12x12(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15; + JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 12]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + z3 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + z3 = LEFT_SHIFT(z3, CONST_BITS); + /* Add fudge factor here for final descale. */ + z3 += ONE << (CONST_BITS - PASS1_BITS - 1); + + z4 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */ + + tmp10 = z3 + z4; + tmp11 = z3 - z4; + + z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */ + z1 = LEFT_SHIFT(z1, CONST_BITS); + z2 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + z2 = LEFT_SHIFT(z2, CONST_BITS); + + tmp12 = z1 - z2; + + tmp21 = z3 + tmp12; + tmp24 = z3 - tmp12; + + tmp12 = z4 + z2; + + tmp20 = tmp10 + tmp12; + tmp25 = tmp10 - tmp12; + + tmp12 = z4 - z1 - z2; + + tmp22 = tmp11 + tmp12; + tmp23 = tmp11 - tmp12; + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */ + tmp14 = MULTIPLY(z2, -FIX_0_541196100); /* -c9 */ + + tmp10 = z1 + z3; + tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */ + tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */ + tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */ + tmp13 = MULTIPLY(z3 + z4, -FIX(1.045510580)); /* -(c7+c11) */ + tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */ + tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */ + tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */ + MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */ + + z1 -= z4; + z2 -= z3; + z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */ + tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */ + tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */ + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 12 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 12; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + z3 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + z3 = LEFT_SHIFT(z3, CONST_BITS); + + z4 = (JLONG)wsptr[4]; + z4 = MULTIPLY(z4, FIX(1.224744871)); /* c4 */ + + tmp10 = z3 + z4; + tmp11 = z3 - z4; + + z1 = (JLONG)wsptr[2]; + z4 = MULTIPLY(z1, FIX(1.366025404)); /* c2 */ + z1 = LEFT_SHIFT(z1, CONST_BITS); + z2 = (JLONG)wsptr[6]; + z2 = LEFT_SHIFT(z2, CONST_BITS); + + tmp12 = z1 - z2; + + tmp21 = z3 + tmp12; + tmp24 = z3 - tmp12; + + tmp12 = z4 + z2; + + tmp20 = tmp10 + tmp12; + tmp25 = tmp10 - tmp12; + + tmp12 = z4 - z1 - z2; + + tmp22 = tmp11 + tmp12; + tmp23 = tmp11 - tmp12; + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + z4 = (JLONG)wsptr[7]; + + tmp11 = MULTIPLY(z2, FIX(1.306562965)); /* c3 */ + tmp14 = MULTIPLY(z2, -FIX_0_541196100); /* -c9 */ + + tmp10 = z1 + z3; + tmp15 = MULTIPLY(tmp10 + z4, FIX(0.860918669)); /* c7 */ + tmp12 = tmp15 + MULTIPLY(tmp10, FIX(0.261052384)); /* c5-c7 */ + tmp10 = tmp12 + tmp11 + MULTIPLY(z1, FIX(0.280143716)); /* c1-c5 */ + tmp13 = MULTIPLY(z3 + z4, -FIX(1.045510580)); /* -(c7+c11) */ + tmp12 += tmp13 + tmp14 - MULTIPLY(z3, FIX(1.478575242)); /* c1+c5-c7-c11 */ + tmp13 += tmp15 - tmp11 + MULTIPLY(z4, FIX(1.586706681)); /* c1+c11 */ + tmp15 += tmp14 - MULTIPLY(z1, FIX(0.676326758)) - /* c7-c11 */ + MULTIPLY(z4, FIX(1.982889723)); /* c5+c7 */ + + z1 -= z4; + z2 -= z3; + z3 = MULTIPLY(z1 + z2, FIX_0_541196100); /* c9 */ + tmp11 = z3 + MULTIPLY(z1, FIX_0_765366865); /* c3-c9 */ + tmp14 = z3 - MULTIPLY(z2, FIX_1_847759065); /* c3+c9 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a 13x13 output block. + * + * Optimized algorithm with 29 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/26). + */ + +GLOBAL(void) +_jpeg_idct_13x13(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15; + JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 13]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + z1 = LEFT_SHIFT(z1, CONST_BITS); + /* Add fudge factor here for final descale. */ + z1 += ONE << (CONST_BITS - PASS1_BITS - 1); + + z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + tmp10 = z3 + z4; + tmp11 = z3 - z4; + + tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */ + tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */ + + tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */ + tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */ + + tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */ + tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */ + + tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */ + tmp25 = MULTIPLY(z2, -FIX(1.252223920)) + tmp12 + tmp13; /* c4 */ + + tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */ + tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */ + + tmp23 = MULTIPLY(z2, -FIX(0.170464608)) - tmp12 - tmp13; /* c12 */ + tmp24 = MULTIPLY(z2, -FIX(0.803364869)) + tmp12 - tmp13; /* c8 */ + + tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */ + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */ + tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */ + tmp15 = z1 + z4; + tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */ + tmp10 = tmp11 + tmp12 + tmp13 - + MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */ + tmp14 = MULTIPLY(z2 + z3, -FIX(0.338443458)); /* -c11 */ + tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */ + tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */ + tmp14 = MULTIPLY(z2 + z4, -FIX(1.163874945)); /* -c5 */ + tmp11 += tmp14; + tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */ + tmp14 = MULTIPLY(z3 + z4, -FIX(0.657217813)); /* -c9 */ + tmp12 += tmp14; + tmp13 += tmp14; + tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */ + tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */ + MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */ + z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */ + tmp14 += z1; + tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */ + MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */ + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS); + wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 13 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 13; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + z1 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + z1 = LEFT_SHIFT(z1, CONST_BITS); + + z2 = (JLONG)wsptr[2]; + z3 = (JLONG)wsptr[4]; + z4 = (JLONG)wsptr[6]; + + tmp10 = z3 + z4; + tmp11 = z3 - z4; + + tmp12 = MULTIPLY(tmp10, FIX(1.155388986)); /* (c4+c6)/2 */ + tmp13 = MULTIPLY(tmp11, FIX(0.096834934)) + z1; /* (c4-c6)/2 */ + + tmp20 = MULTIPLY(z2, FIX(1.373119086)) + tmp12 + tmp13; /* c2 */ + tmp22 = MULTIPLY(z2, FIX(0.501487041)) - tmp12 + tmp13; /* c10 */ + + tmp12 = MULTIPLY(tmp10, FIX(0.316450131)); /* (c8-c12)/2 */ + tmp13 = MULTIPLY(tmp11, FIX(0.486914739)) + z1; /* (c8+c12)/2 */ + + tmp21 = MULTIPLY(z2, FIX(1.058554052)) - tmp12 + tmp13; /* c6 */ + tmp25 = MULTIPLY(z2, -FIX(1.252223920)) + tmp12 + tmp13; /* c4 */ + + tmp12 = MULTIPLY(tmp10, FIX(0.435816023)); /* (c2-c10)/2 */ + tmp13 = MULTIPLY(tmp11, FIX(0.937303064)) - z1; /* (c2+c10)/2 */ + + tmp23 = MULTIPLY(z2, -FIX(0.170464608)) - tmp12 - tmp13; /* c12 */ + tmp24 = MULTIPLY(z2, -FIX(0.803364869)) + tmp12 - tmp13; /* c8 */ + + tmp26 = MULTIPLY(tmp11 - z2, FIX(1.414213562)) + z1; /* c0 */ + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + z4 = (JLONG)wsptr[7]; + + tmp11 = MULTIPLY(z1 + z2, FIX(1.322312651)); /* c3 */ + tmp12 = MULTIPLY(z1 + z3, FIX(1.163874945)); /* c5 */ + tmp15 = z1 + z4; + tmp13 = MULTIPLY(tmp15, FIX(0.937797057)); /* c7 */ + tmp10 = tmp11 + tmp12 + tmp13 - + MULTIPLY(z1, FIX(2.020082300)); /* c7+c5+c3-c1 */ + tmp14 = MULTIPLY(z2 + z3, -FIX(0.338443458)); /* -c11 */ + tmp11 += tmp14 + MULTIPLY(z2, FIX(0.837223564)); /* c5+c9+c11-c3 */ + tmp12 += tmp14 - MULTIPLY(z3, FIX(1.572116027)); /* c1+c5-c9-c11 */ + tmp14 = MULTIPLY(z2 + z4, -FIX(1.163874945)); /* -c5 */ + tmp11 += tmp14; + tmp13 += tmp14 + MULTIPLY(z4, FIX(2.205608352)); /* c3+c5+c9-c7 */ + tmp14 = MULTIPLY(z3 + z4, -FIX(0.657217813)); /* -c9 */ + tmp12 += tmp14; + tmp13 += tmp14; + tmp15 = MULTIPLY(tmp15, FIX(0.338443458)); /* c11 */ + tmp14 = tmp15 + MULTIPLY(z1, FIX(0.318774355)) - /* c9-c11 */ + MULTIPLY(z2, FIX(0.466105296)); /* c1-c7 */ + z1 = MULTIPLY(z3 - z2, FIX(0.937797057)); /* c7 */ + tmp14 += z1; + tmp15 += z1 + MULTIPLY(z3, FIX(0.384515595)) - /* c3-c7 */ + MULTIPLY(z4, FIX(1.742345811)); /* c1+c11 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a 14x14 output block. + * + * Optimized algorithm with 20 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/28). + */ + +GLOBAL(void) +_jpeg_idct_14x14(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16; + JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 14]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + z1 = LEFT_SHIFT(z1, CONST_BITS); + /* Add fudge factor here for final descale. */ + z1 += ONE << (CONST_BITS - PASS1_BITS - 1); + z4 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */ + z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */ + z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */ + + tmp10 = z1 + z2; + tmp11 = z1 + z3; + tmp12 = z1 - z4; + + tmp23 = RIGHT_SHIFT(z1 - LEFT_SHIFT(z2 + z3 - z4, 1), + CONST_BITS - PASS1_BITS); /* c0 = (c4+c12-c8)*2 */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */ + + tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */ + tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */ + tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */ + MULTIPLY(z2, FIX(1.378756276)); /* c2 */ + + tmp20 = tmp10 + tmp13; + tmp26 = tmp10 - tmp13; + tmp21 = tmp11 + tmp14; + tmp25 = tmp11 - tmp14; + tmp22 = tmp12 + tmp15; + tmp24 = tmp12 - tmp15; + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + tmp13 = LEFT_SHIFT(z4, CONST_BITS); + + tmp14 = z1 + z3; + tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */ + tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */ + tmp10 = tmp11 + tmp12 + tmp13 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */ + tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */ + tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */ + z1 -= z2; + tmp15 = MULTIPLY(z1, FIX(0.467085129)) - tmp13; /* c11 */ + tmp16 += tmp15; + z1 += z4; + z4 = MULTIPLY(z2 + z3, -FIX(0.158341681)) - tmp13; /* -c13 */ + tmp11 += z4 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */ + tmp12 += z4 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */ + z4 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */ + tmp14 += z4 + tmp13 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */ + tmp15 += z4 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */ + + tmp13 = LEFT_SHIFT(z1 - z3, PASS1_BITS); + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 13] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 3] = (int)(tmp23 + tmp13); + wsptr[8 * 10] = (int)(tmp23 - tmp13); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS - PASS1_BITS); + wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 14 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 14; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + z1 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + z1 = LEFT_SHIFT(z1, CONST_BITS); + z4 = (JLONG)wsptr[4]; + z2 = MULTIPLY(z4, FIX(1.274162392)); /* c4 */ + z3 = MULTIPLY(z4, FIX(0.314692123)); /* c12 */ + z4 = MULTIPLY(z4, FIX(0.881747734)); /* c8 */ + + tmp10 = z1 + z2; + tmp11 = z1 + z3; + tmp12 = z1 - z4; + + tmp23 = z1 - LEFT_SHIFT(z2 + z3 - z4, 1); /* c0 = (c4+c12-c8)*2 */ + + z1 = (JLONG)wsptr[2]; + z2 = (JLONG)wsptr[6]; + + z3 = MULTIPLY(z1 + z2, FIX(1.105676686)); /* c6 */ + + tmp13 = z3 + MULTIPLY(z1, FIX(0.273079590)); /* c2-c6 */ + tmp14 = z3 - MULTIPLY(z2, FIX(1.719280954)); /* c6+c10 */ + tmp15 = MULTIPLY(z1, FIX(0.613604268)) - /* c10 */ + MULTIPLY(z2, FIX(1.378756276)); /* c2 */ + + tmp20 = tmp10 + tmp13; + tmp26 = tmp10 - tmp13; + tmp21 = tmp11 + tmp14; + tmp25 = tmp11 - tmp14; + tmp22 = tmp12 + tmp15; + tmp24 = tmp12 - tmp15; + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + z4 = (JLONG)wsptr[7]; + z4 = LEFT_SHIFT(z4, CONST_BITS); + + tmp14 = z1 + z3; + tmp11 = MULTIPLY(z1 + z2, FIX(1.334852607)); /* c3 */ + tmp12 = MULTIPLY(tmp14, FIX(1.197448846)); /* c5 */ + tmp10 = tmp11 + tmp12 + z4 - MULTIPLY(z1, FIX(1.126980169)); /* c3+c5-c1 */ + tmp14 = MULTIPLY(tmp14, FIX(0.752406978)); /* c9 */ + tmp16 = tmp14 - MULTIPLY(z1, FIX(1.061150426)); /* c9+c11-c13 */ + z1 -= z2; + tmp15 = MULTIPLY(z1, FIX(0.467085129)) - z4; /* c11 */ + tmp16 += tmp15; + tmp13 = MULTIPLY(z2 + z3, -FIX(0.158341681)) - z4; /* -c13 */ + tmp11 += tmp13 - MULTIPLY(z2, FIX(0.424103948)); /* c3-c9-c13 */ + tmp12 += tmp13 - MULTIPLY(z3, FIX(2.373959773)); /* c3+c5-c13 */ + tmp13 = MULTIPLY(z3 - z2, FIX(1.405321284)); /* c1 */ + tmp14 += tmp13 + z4 - MULTIPLY(z3, FIX(1.6906431334)); /* c1+c9-c11 */ + tmp15 += tmp13 + MULTIPLY(z2, FIX(0.674957567)); /* c1+c11-c5 */ + + tmp13 = LEFT_SHIFT(z1 - z3, CONST_BITS) + z4; + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[13] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26 + tmp16, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp26 - tmp16, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a 15x15 output block. + * + * Optimized algorithm with 22 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/30). + */ + +GLOBAL(void) +_jpeg_idct_15x15(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp10, tmp11, tmp12, tmp13, tmp14, tmp15, tmp16; + JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 15]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + z1 = LEFT_SHIFT(z1, CONST_BITS); + /* Add fudge factor here for final descale. */ + z1 += ONE << (CONST_BITS - PASS1_BITS - 1); + + z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */ + tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */ + + tmp12 = z1 - tmp10; + tmp13 = z1 + tmp11; + z1 -= LEFT_SHIFT(tmp11 - tmp10, 1); /* c0 = (c6-c12)*2 */ + + z4 = z2 - z3; + z3 += z2; + tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */ + tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */ + z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */ + + tmp20 = tmp13 + tmp10 + tmp11; + tmp23 = tmp12 - tmp10 + tmp11 + z2; + + tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */ + tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */ + + tmp25 = tmp13 - tmp10 - tmp11; + tmp26 = tmp12 + tmp10 - tmp11 - z2; + + tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */ + tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */ + + tmp21 = tmp12 + tmp10 + tmp11; + tmp24 = tmp13 - tmp10 + tmp11; + tmp11 += tmp11; + tmp22 = z1 + tmp11; /* c10 = c6-c12 */ + tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */ + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */ + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + tmp13 = z2 - z4; + tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */ + tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */ + tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */ + + tmp13 = MULTIPLY(z2, -FIX(0.831253876)); /* -c9 */ + tmp15 = MULTIPLY(z2, -FIX(1.344997024)); /* -c3 */ + z2 = z1 - z4; + tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */ + + tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */ + tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */ + tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */ + z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */ + tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */ + tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */ + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 14] = (int)RIGHT_SHIFT(tmp20 - tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 13] = (int)RIGHT_SHIFT(tmp21 - tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp22 - tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp23 - tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp24 - tmp14, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp15, CONST_BITS - PASS1_BITS); + wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp25 - tmp15, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26 + tmp16, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp26 - tmp16, CONST_BITS - PASS1_BITS); + wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp27, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 15 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 15; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + z1 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + z1 = LEFT_SHIFT(z1, CONST_BITS); + + z2 = (JLONG)wsptr[2]; + z3 = (JLONG)wsptr[4]; + z4 = (JLONG)wsptr[6]; + + tmp10 = MULTIPLY(z4, FIX(0.437016024)); /* c12 */ + tmp11 = MULTIPLY(z4, FIX(1.144122806)); /* c6 */ + + tmp12 = z1 - tmp10; + tmp13 = z1 + tmp11; + z1 -= LEFT_SHIFT(tmp11 - tmp10, 1); /* c0 = (c6-c12)*2 */ + + z4 = z2 - z3; + z3 += z2; + tmp10 = MULTIPLY(z3, FIX(1.337628990)); /* (c2+c4)/2 */ + tmp11 = MULTIPLY(z4, FIX(0.045680613)); /* (c2-c4)/2 */ + z2 = MULTIPLY(z2, FIX(1.439773946)); /* c4+c14 */ + + tmp20 = tmp13 + tmp10 + tmp11; + tmp23 = tmp12 - tmp10 + tmp11 + z2; + + tmp10 = MULTIPLY(z3, FIX(0.547059574)); /* (c8+c14)/2 */ + tmp11 = MULTIPLY(z4, FIX(0.399234004)); /* (c8-c14)/2 */ + + tmp25 = tmp13 - tmp10 - tmp11; + tmp26 = tmp12 + tmp10 - tmp11 - z2; + + tmp10 = MULTIPLY(z3, FIX(0.790569415)); /* (c6+c12)/2 */ + tmp11 = MULTIPLY(z4, FIX(0.353553391)); /* (c6-c12)/2 */ + + tmp21 = tmp12 + tmp10 + tmp11; + tmp24 = tmp13 - tmp10 + tmp11; + tmp11 += tmp11; + tmp22 = z1 + tmp11; /* c10 = c6-c12 */ + tmp27 = z1 - tmp11 - tmp11; /* c0 = (c6-c12)*2 */ + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z4 = (JLONG)wsptr[5]; + z3 = MULTIPLY(z4, FIX(1.224744871)); /* c5 */ + z4 = (JLONG)wsptr[7]; + + tmp13 = z2 - z4; + tmp15 = MULTIPLY(z1 + tmp13, FIX(0.831253876)); /* c9 */ + tmp11 = tmp15 + MULTIPLY(z1, FIX(0.513743148)); /* c3-c9 */ + tmp14 = tmp15 - MULTIPLY(tmp13, FIX(2.176250899)); /* c3+c9 */ + + tmp13 = MULTIPLY(z2, -FIX(0.831253876)); /* -c9 */ + tmp15 = MULTIPLY(z2, -FIX(1.344997024)); /* -c3 */ + z2 = z1 - z4; + tmp12 = z3 + MULTIPLY(z2, FIX(1.406466353)); /* c1 */ + + tmp10 = tmp12 + MULTIPLY(z4, FIX(2.457431844)) - tmp15; /* c1+c7 */ + tmp16 = tmp12 - MULTIPLY(z1, FIX(1.112434820)) + tmp13; /* c1-c13 */ + tmp12 = MULTIPLY(z2, FIX(1.224744871)) - z3; /* c5 */ + z2 = MULTIPLY(z1 + z4, FIX(0.575212477)); /* c11 */ + tmp13 += z2 + MULTIPLY(z1, FIX(0.475753014)) - z3; /* c7-c11 */ + tmp15 += z2 - MULTIPLY(z4, FIX(0.869244010)) + z3; /* c11+c13 */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[14] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[13] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp14, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp15, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26 + tmp16, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp26 - tmp16, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp27, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a 16x16 output block. + * + * Optimized algorithm with 28 multiplications in the 1-D kernel. + * cK represents sqrt(2) * cos(K*pi/32). + */ + +GLOBAL(void) +_jpeg_idct_16x16(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp1, tmp2, tmp3, tmp10, tmp11, tmp12, tmp13; + JLONG tmp20, tmp21, tmp22, tmp23, tmp24, tmp25, tmp26, tmp27; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[8 * 16]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = 0; ctr < 8; ctr++, inptr++, quantptr++, wsptr++) { + /* Even part */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + /* Add fudge factor here for final descale. */ + tmp0 += ONE << (CONST_BITS - PASS1_BITS - 1); + + z1 = DEQUANTIZE(inptr[DCTSIZE * 4], quantptr[DCTSIZE * 4]); + tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */ + tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */ + + tmp10 = tmp0 + tmp1; + tmp11 = tmp0 - tmp1; + tmp12 = tmp0 + tmp2; + tmp13 = tmp0 - tmp2; + + z1 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + z3 = z1 - z2; + z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */ + z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */ + + tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */ + tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */ + tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */ + tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */ + + tmp20 = tmp10 + tmp0; + tmp27 = tmp10 - tmp0; + tmp21 = tmp12 + tmp1; + tmp26 = tmp12 - tmp1; + tmp22 = tmp13 + tmp2; + tmp25 = tmp13 - tmp2; + tmp23 = tmp11 + tmp3; + tmp24 = tmp11 - tmp3; + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + + tmp11 = z1 + z3; + + tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */ + tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */ + tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */ + tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */ + tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */ + tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */ + tmp0 = tmp1 + tmp2 + tmp3 - + MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */ + tmp13 = tmp10 + tmp11 + tmp12 - + MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */ + z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */ + tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */ + tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */ + z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */ + tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */ + tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */ + z2 += z4; + z1 = MULTIPLY(z2, -FIX(0.666655658)); /* -c11 */ + tmp1 += z1; + tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */ + z2 = MULTIPLY(z2, -FIX(1.247225013)); /* -c5 */ + tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */ + tmp12 += z2; + z2 = MULTIPLY(z3 + z4, -FIX(1.353318001)); /* -c3 */ + tmp2 += z2; + tmp3 += z2; + z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */ + tmp10 += z2; + tmp11 += z2; + + /* Final output stage */ + + wsptr[8 * 0] = (int)RIGHT_SHIFT(tmp20 + tmp0, CONST_BITS - PASS1_BITS); + wsptr[8 * 15] = (int)RIGHT_SHIFT(tmp20 - tmp0, CONST_BITS - PASS1_BITS); + wsptr[8 * 1] = (int)RIGHT_SHIFT(tmp21 + tmp1, CONST_BITS - PASS1_BITS); + wsptr[8 * 14] = (int)RIGHT_SHIFT(tmp21 - tmp1, CONST_BITS - PASS1_BITS); + wsptr[8 * 2] = (int)RIGHT_SHIFT(tmp22 + tmp2, CONST_BITS - PASS1_BITS); + wsptr[8 * 13] = (int)RIGHT_SHIFT(tmp22 - tmp2, CONST_BITS - PASS1_BITS); + wsptr[8 * 3] = (int)RIGHT_SHIFT(tmp23 + tmp3, CONST_BITS - PASS1_BITS); + wsptr[8 * 12] = (int)RIGHT_SHIFT(tmp23 - tmp3, CONST_BITS - PASS1_BITS); + wsptr[8 * 4] = (int)RIGHT_SHIFT(tmp24 + tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 11] = (int)RIGHT_SHIFT(tmp24 - tmp10, CONST_BITS - PASS1_BITS); + wsptr[8 * 5] = (int)RIGHT_SHIFT(tmp25 + tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 10] = (int)RIGHT_SHIFT(tmp25 - tmp11, CONST_BITS - PASS1_BITS); + wsptr[8 * 6] = (int)RIGHT_SHIFT(tmp26 + tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 9] = (int)RIGHT_SHIFT(tmp26 - tmp12, CONST_BITS - PASS1_BITS); + wsptr[8 * 7] = (int)RIGHT_SHIFT(tmp27 + tmp13, CONST_BITS - PASS1_BITS); + wsptr[8 * 8] = (int)RIGHT_SHIFT(tmp27 - tmp13, CONST_BITS - PASS1_BITS); + } + + /* Pass 2: process 16 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 16; ctr++) { + outptr = output_buf[ctr] + output_col; + + /* Even part */ + + /* Add fudge factor here for final descale. */ + tmp0 = (JLONG)wsptr[0] + (ONE << (PASS1_BITS + 2)); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS); + + z1 = (JLONG)wsptr[4]; + tmp1 = MULTIPLY(z1, FIX(1.306562965)); /* c4[16] = c2[8] */ + tmp2 = MULTIPLY(z1, FIX_0_541196100); /* c12[16] = c6[8] */ + + tmp10 = tmp0 + tmp1; + tmp11 = tmp0 - tmp1; + tmp12 = tmp0 + tmp2; + tmp13 = tmp0 - tmp2; + + z1 = (JLONG)wsptr[2]; + z2 = (JLONG)wsptr[6]; + z3 = z1 - z2; + z4 = MULTIPLY(z3, FIX(0.275899379)); /* c14[16] = c7[8] */ + z3 = MULTIPLY(z3, FIX(1.387039845)); /* c2[16] = c1[8] */ + + tmp0 = z3 + MULTIPLY(z2, FIX_2_562915447); /* (c6+c2)[16] = (c3+c1)[8] */ + tmp1 = z4 + MULTIPLY(z1, FIX_0_899976223); /* (c6-c14)[16] = (c3-c7)[8] */ + tmp2 = z3 - MULTIPLY(z1, FIX(0.601344887)); /* (c2-c10)[16] = (c1-c5)[8] */ + tmp3 = z4 - MULTIPLY(z2, FIX(0.509795579)); /* (c10-c14)[16] = (c5-c7)[8] */ + + tmp20 = tmp10 + tmp0; + tmp27 = tmp10 - tmp0; + tmp21 = tmp12 + tmp1; + tmp26 = tmp12 - tmp1; + tmp22 = tmp13 + tmp2; + tmp25 = tmp13 - tmp2; + tmp23 = tmp11 + tmp3; + tmp24 = tmp11 - tmp3; + + /* Odd part */ + + z1 = (JLONG)wsptr[1]; + z2 = (JLONG)wsptr[3]; + z3 = (JLONG)wsptr[5]; + z4 = (JLONG)wsptr[7]; + + tmp11 = z1 + z3; + + tmp1 = MULTIPLY(z1 + z2, FIX(1.353318001)); /* c3 */ + tmp2 = MULTIPLY(tmp11, FIX(1.247225013)); /* c5 */ + tmp3 = MULTIPLY(z1 + z4, FIX(1.093201867)); /* c7 */ + tmp10 = MULTIPLY(z1 - z4, FIX(0.897167586)); /* c9 */ + tmp11 = MULTIPLY(tmp11, FIX(0.666655658)); /* c11 */ + tmp12 = MULTIPLY(z1 - z2, FIX(0.410524528)); /* c13 */ + tmp0 = tmp1 + tmp2 + tmp3 - + MULTIPLY(z1, FIX(2.286341144)); /* c7+c5+c3-c1 */ + tmp13 = tmp10 + tmp11 + tmp12 - + MULTIPLY(z1, FIX(1.835730603)); /* c9+c11+c13-c15 */ + z1 = MULTIPLY(z2 + z3, FIX(0.138617169)); /* c15 */ + tmp1 += z1 + MULTIPLY(z2, FIX(0.071888074)); /* c9+c11-c3-c15 */ + tmp2 += z1 - MULTIPLY(z3, FIX(1.125726048)); /* c5+c7+c15-c3 */ + z1 = MULTIPLY(z3 - z2, FIX(1.407403738)); /* c1 */ + tmp11 += z1 - MULTIPLY(z3, FIX(0.766367282)); /* c1+c11-c9-c13 */ + tmp12 += z1 + MULTIPLY(z2, FIX(1.971951411)); /* c1+c5+c13-c7 */ + z2 += z4; + z1 = MULTIPLY(z2, -FIX(0.666655658)); /* -c11 */ + tmp1 += z1; + tmp3 += z1 + MULTIPLY(z4, FIX(1.065388962)); /* c3+c11+c15-c7 */ + z2 = MULTIPLY(z2, -FIX(1.247225013)); /* -c5 */ + tmp10 += z2 + MULTIPLY(z4, FIX(3.141271809)); /* c1+c5+c9-c13 */ + tmp12 += z2; + z2 = MULTIPLY(z3 + z4, -FIX(1.353318001)); /* -c3 */ + tmp2 += z2; + tmp3 += z2; + z2 = MULTIPLY(z4 - z3, FIX(0.410524528)); /* c13 */ + tmp10 += z2; + tmp11 += z2; + + /* Final output stage */ + + outptr[0] = range_limit[(int)RIGHT_SHIFT(tmp20 + tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[15] = range_limit[(int)RIGHT_SHIFT(tmp20 - tmp0, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[1] = range_limit[(int)RIGHT_SHIFT(tmp21 + tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[14] = range_limit[(int)RIGHT_SHIFT(tmp21 - tmp1, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[2] = range_limit[(int)RIGHT_SHIFT(tmp22 + tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[13] = range_limit[(int)RIGHT_SHIFT(tmp22 - tmp2, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[3] = range_limit[(int)RIGHT_SHIFT(tmp23 + tmp3, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[12] = range_limit[(int)RIGHT_SHIFT(tmp23 - tmp3, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[4] = range_limit[(int)RIGHT_SHIFT(tmp24 + tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[11] = range_limit[(int)RIGHT_SHIFT(tmp24 - tmp10, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[5] = range_limit[(int)RIGHT_SHIFT(tmp25 + tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[10] = range_limit[(int)RIGHT_SHIFT(tmp25 - tmp11, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[6] = range_limit[(int)RIGHT_SHIFT(tmp26 + tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[9] = range_limit[(int)RIGHT_SHIFT(tmp26 - tmp12, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[7] = range_limit[(int)RIGHT_SHIFT(tmp27 + tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + outptr[8] = range_limit[(int)RIGHT_SHIFT(tmp27 - tmp13, + CONST_BITS + PASS1_BITS + 3) & + RANGE_MASK]; + + wsptr += 8; /* advance pointer to next row */ + } +} + +#endif /* IDCT_SCALING_SUPPORTED */ +#endif /* DCT_ISLOW_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jidctred.c b/src/spice2x/external/libjpeg-turbo/jidctred.c new file mode 100644 index 0000000..6521e3e --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jidctred.c @@ -0,0 +1,409 @@ +/* + * jidctred.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1994-1998, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2015, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains inverse-DCT routines that produce reduced-size output: + * either 4x4, 2x2, or 1x1 pixels from an 8x8 DCT block. + * + * The implementation is based on the Loeffler, Ligtenberg and Moschytz (LL&M) + * algorithm used in jidctint.c. We simply replace each 8-to-8 1-D IDCT step + * with an 8-to-4 step that produces the four averages of two adjacent outputs + * (or an 8-to-2 step producing two averages of four outputs, for 2x2 output). + * These steps were derived by computing the corresponding values at the end + * of the normal LL&M code, then simplifying as much as possible. + * + * 1x1 is trivial: just take the DC coefficient divided by 8. + * + * See jidctint.c for additional comments. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jdct.h" /* Private declarations for DCT subsystem */ + +#ifdef IDCT_SCALING_SUPPORTED + + +/* + * This module is specialized to the case DCTSIZE = 8. + */ + +#if DCTSIZE != 8 + Sorry, this code only copes with 8x8 DCTs. /* deliberate syntax err */ +#endif + + +/* Scaling is the same as in jidctint.c. */ + +#if BITS_IN_JSAMPLE == 8 +#define CONST_BITS 13 +#define PASS1_BITS 2 +#else +#define CONST_BITS 13 +#define PASS1_BITS 1 /* lose a little precision to avoid overflow */ +#endif + +/* Some C compilers fail to reduce "FIX(constant)" at compile time, thus + * causing a lot of useless floating-point operations at run time. + * To get around this we use the following pre-calculated constants. + * If you change CONST_BITS you may want to add appropriate values. + * (With a reasonable C compiler, you can just rely on the FIX() macro...) + */ + +#if CONST_BITS == 13 +#define FIX_0_211164243 ((JLONG)1730) /* FIX(0.211164243) */ +#define FIX_0_509795579 ((JLONG)4176) /* FIX(0.509795579) */ +#define FIX_0_601344887 ((JLONG)4926) /* FIX(0.601344887) */ +#define FIX_0_720959822 ((JLONG)5906) /* FIX(0.720959822) */ +#define FIX_0_765366865 ((JLONG)6270) /* FIX(0.765366865) */ +#define FIX_0_850430095 ((JLONG)6967) /* FIX(0.850430095) */ +#define FIX_0_899976223 ((JLONG)7373) /* FIX(0.899976223) */ +#define FIX_1_061594337 ((JLONG)8697) /* FIX(1.061594337) */ +#define FIX_1_272758580 ((JLONG)10426) /* FIX(1.272758580) */ +#define FIX_1_451774981 ((JLONG)11893) /* FIX(1.451774981) */ +#define FIX_1_847759065 ((JLONG)15137) /* FIX(1.847759065) */ +#define FIX_2_172734803 ((JLONG)17799) /* FIX(2.172734803) */ +#define FIX_2_562915447 ((JLONG)20995) /* FIX(2.562915447) */ +#define FIX_3_624509785 ((JLONG)29692) /* FIX(3.624509785) */ +#else +#define FIX_0_211164243 FIX(0.211164243) +#define FIX_0_509795579 FIX(0.509795579) +#define FIX_0_601344887 FIX(0.601344887) +#define FIX_0_720959822 FIX(0.720959822) +#define FIX_0_765366865 FIX(0.765366865) +#define FIX_0_850430095 FIX(0.850430095) +#define FIX_0_899976223 FIX(0.899976223) +#define FIX_1_061594337 FIX(1.061594337) +#define FIX_1_272758580 FIX(1.272758580) +#define FIX_1_451774981 FIX(1.451774981) +#define FIX_1_847759065 FIX(1.847759065) +#define FIX_2_172734803 FIX(2.172734803) +#define FIX_2_562915447 FIX(2.562915447) +#define FIX_3_624509785 FIX(3.624509785) +#endif + + +/* Multiply a JLONG variable by a JLONG constant to yield a JLONG result. + * For 8-bit samples with the recommended scaling, all the variable + * and constant values involved are no more than 16 bits wide, so a + * 16x16->32 bit multiply can be used instead of a full 32x32 multiply. + * For 12-bit samples, a full 32-bit multiplication will be needed. + */ + +#if BITS_IN_JSAMPLE == 8 +#define MULTIPLY(var, const) MULTIPLY16C16(var, const) +#else +#define MULTIPLY(var, const) ((var) * (const)) +#endif + + +/* Dequantize a coefficient by multiplying it by the multiplier-table + * entry; produce an int result. In this module, both inputs and result + * are 16 bits or less, so either int or short multiply will work. + */ + +#define DEQUANTIZE(coef, quantval) (((ISLOW_MULT_TYPE)(coef)) * (quantval)) + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a reduced-size 4x4 output block. + */ + +GLOBAL(void) +_jpeg_idct_4x4(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp2, tmp10, tmp12; + JLONG z1, z2, z3, z4; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[DCTSIZE * 4]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { + /* Don't bother to process column 4, because second pass won't use it */ + if (ctr == DCTSIZE - 4) + continue; + if (inptr[DCTSIZE * 1] == 0 && inptr[DCTSIZE * 2] == 0 && + inptr[DCTSIZE * 3] == 0 && inptr[DCTSIZE * 5] == 0 && + inptr[DCTSIZE * 6] == 0 && inptr[DCTSIZE * 7] == 0) { + /* AC terms all zero; we need not examine term 4 for 4x4 output */ + int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE * 0], + quantptr[DCTSIZE * 0]), PASS1_BITS); + + wsptr[DCTSIZE * 0] = dcval; + wsptr[DCTSIZE * 1] = dcval; + wsptr[DCTSIZE * 2] = dcval; + wsptr[DCTSIZE * 3] = dcval; + + continue; + } + + /* Even part */ + + tmp0 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp0 = LEFT_SHIFT(tmp0, CONST_BITS + 1); + + z2 = DEQUANTIZE(inptr[DCTSIZE * 2], quantptr[DCTSIZE * 2]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 6], quantptr[DCTSIZE * 6]); + + tmp2 = MULTIPLY(z2, FIX_1_847759065) + MULTIPLY(z3, -FIX_0_765366865); + + tmp10 = tmp0 + tmp2; + tmp12 = tmp0 - tmp2; + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + z2 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + z3 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + z4 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + + tmp0 = MULTIPLY(z1, -FIX_0_211164243) + /* sqrt(2) * ( c3-c1) */ + MULTIPLY(z2, FIX_1_451774981) + /* sqrt(2) * ( c3+c7) */ + MULTIPLY(z3, -FIX_2_172734803) + /* sqrt(2) * (-c1-c5) */ + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * ( c5+c7) */ + + tmp2 = MULTIPLY(z1, -FIX_0_509795579) + /* sqrt(2) * (c7-c5) */ + MULTIPLY(z2, -FIX_0_601344887) + /* sqrt(2) * (c5-c1) */ + MULTIPLY(z3, FIX_0_899976223) + /* sqrt(2) * (c3-c7) */ + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ + + /* Final output stage */ + + wsptr[DCTSIZE * 0] = + (int)DESCALE(tmp10 + tmp2, CONST_BITS - PASS1_BITS + 1); + wsptr[DCTSIZE * 3] = + (int)DESCALE(tmp10 - tmp2, CONST_BITS - PASS1_BITS + 1); + wsptr[DCTSIZE * 1] = + (int)DESCALE(tmp12 + tmp0, CONST_BITS - PASS1_BITS + 1); + wsptr[DCTSIZE * 2] = + (int)DESCALE(tmp12 - tmp0, CONST_BITS - PASS1_BITS + 1); + } + + /* Pass 2: process 4 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 4; ctr++) { + outptr = output_buf[ctr] + output_col; + /* It's not clear whether a zero row test is worthwhile here ... */ + +#ifndef NO_ZERO_ROW_TEST + if (wsptr[1] == 0 && wsptr[2] == 0 && wsptr[3] == 0 && + wsptr[5] == 0 && wsptr[6] == 0 && wsptr[7] == 0) { + /* AC terms all zero */ + _JSAMPLE dcval = range_limit[(int)DESCALE((JLONG)wsptr[0], + PASS1_BITS + 3) & RANGE_MASK]; + + outptr[0] = dcval; + outptr[1] = dcval; + outptr[2] = dcval; + outptr[3] = dcval; + + wsptr += DCTSIZE; /* advance pointer to next row */ + continue; + } +#endif + + /* Even part */ + + tmp0 = LEFT_SHIFT((JLONG)wsptr[0], CONST_BITS + 1); + + tmp2 = MULTIPLY((JLONG)wsptr[2], FIX_1_847759065) + + MULTIPLY((JLONG)wsptr[6], -FIX_0_765366865); + + tmp10 = tmp0 + tmp2; + tmp12 = tmp0 - tmp2; + + /* Odd part */ + + z1 = (JLONG)wsptr[7]; + z2 = (JLONG)wsptr[5]; + z3 = (JLONG)wsptr[3]; + z4 = (JLONG)wsptr[1]; + + tmp0 = MULTIPLY(z1, -FIX_0_211164243) + /* sqrt(2) * ( c3-c1) */ + MULTIPLY(z2, FIX_1_451774981) + /* sqrt(2) * ( c3+c7) */ + MULTIPLY(z3, -FIX_2_172734803) + /* sqrt(2) * (-c1-c5) */ + MULTIPLY(z4, FIX_1_061594337); /* sqrt(2) * ( c5+c7) */ + + tmp2 = MULTIPLY(z1, -FIX_0_509795579) + /* sqrt(2) * (c7-c5) */ + MULTIPLY(z2, -FIX_0_601344887) + /* sqrt(2) * (c5-c1) */ + MULTIPLY(z3, FIX_0_899976223) + /* sqrt(2) * (c3-c7) */ + MULTIPLY(z4, FIX_2_562915447); /* sqrt(2) * (c1+c3) */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)DESCALE(tmp10 + tmp2, + CONST_BITS + PASS1_BITS + 3 + 1) & + RANGE_MASK]; + outptr[3] = range_limit[(int)DESCALE(tmp10 - tmp2, + CONST_BITS + PASS1_BITS + 3 + 1) & + RANGE_MASK]; + outptr[1] = range_limit[(int)DESCALE(tmp12 + tmp0, + CONST_BITS + PASS1_BITS + 3 + 1) & + RANGE_MASK]; + outptr[2] = range_limit[(int)DESCALE(tmp12 - tmp0, + CONST_BITS + PASS1_BITS + 3 + 1) & + RANGE_MASK]; + + wsptr += DCTSIZE; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a reduced-size 2x2 output block. + */ + +GLOBAL(void) +_jpeg_idct_2x2(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + JLONG tmp0, tmp10, z1; + JCOEFPTR inptr; + ISLOW_MULT_TYPE *quantptr; + int *wsptr; + _JSAMPROW outptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + int ctr; + int workspace[DCTSIZE * 2]; /* buffers data between passes */ + SHIFT_TEMPS + + /* Pass 1: process columns from input, store into work array. */ + + inptr = coef_block; + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + wsptr = workspace; + for (ctr = DCTSIZE; ctr > 0; inptr++, quantptr++, wsptr++, ctr--) { + /* Don't bother to process columns 2,4,6 */ + if (ctr == DCTSIZE - 2 || ctr == DCTSIZE - 4 || ctr == DCTSIZE - 6) + continue; + if (inptr[DCTSIZE * 1] == 0 && inptr[DCTSIZE * 3] == 0 && + inptr[DCTSIZE * 5] == 0 && inptr[DCTSIZE * 7] == 0) { + /* AC terms all zero; we need not examine terms 2,4,6 for 2x2 output */ + int dcval = LEFT_SHIFT(DEQUANTIZE(inptr[DCTSIZE * 0], + quantptr[DCTSIZE * 0]), PASS1_BITS); + + wsptr[DCTSIZE * 0] = dcval; + wsptr[DCTSIZE * 1] = dcval; + + continue; + } + + /* Even part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 0], quantptr[DCTSIZE * 0]); + tmp10 = LEFT_SHIFT(z1, CONST_BITS + 2); + + /* Odd part */ + + z1 = DEQUANTIZE(inptr[DCTSIZE * 7], quantptr[DCTSIZE * 7]); + tmp0 = MULTIPLY(z1, -FIX_0_720959822); /* sqrt(2) * ( c7-c5+c3-c1) */ + z1 = DEQUANTIZE(inptr[DCTSIZE * 5], quantptr[DCTSIZE * 5]); + tmp0 += MULTIPLY(z1, FIX_0_850430095); /* sqrt(2) * (-c1+c3+c5+c7) */ + z1 = DEQUANTIZE(inptr[DCTSIZE * 3], quantptr[DCTSIZE * 3]); + tmp0 += MULTIPLY(z1, -FIX_1_272758580); /* sqrt(2) * (-c1+c3-c5-c7) */ + z1 = DEQUANTIZE(inptr[DCTSIZE * 1], quantptr[DCTSIZE * 1]); + tmp0 += MULTIPLY(z1, FIX_3_624509785); /* sqrt(2) * ( c1+c3+c5+c7) */ + + /* Final output stage */ + + wsptr[DCTSIZE * 0] = + (int)DESCALE(tmp10 + tmp0, CONST_BITS - PASS1_BITS + 2); + wsptr[DCTSIZE * 1] = + (int)DESCALE(tmp10 - tmp0, CONST_BITS - PASS1_BITS + 2); + } + + /* Pass 2: process 2 rows from work array, store into output array. */ + + wsptr = workspace; + for (ctr = 0; ctr < 2; ctr++) { + outptr = output_buf[ctr] + output_col; + /* It's not clear whether a zero row test is worthwhile here ... */ + +#ifndef NO_ZERO_ROW_TEST + if (wsptr[1] == 0 && wsptr[3] == 0 && wsptr[5] == 0 && wsptr[7] == 0) { + /* AC terms all zero */ + _JSAMPLE dcval = range_limit[(int)DESCALE((JLONG)wsptr[0], + PASS1_BITS + 3) & RANGE_MASK]; + + outptr[0] = dcval; + outptr[1] = dcval; + + wsptr += DCTSIZE; /* advance pointer to next row */ + continue; + } +#endif + + /* Even part */ + + tmp10 = LEFT_SHIFT((JLONG)wsptr[0], CONST_BITS + 2); + + /* Odd part */ + + tmp0 = MULTIPLY((JLONG)wsptr[7], -FIX_0_720959822) + /* sqrt(2) * ( c7-c5+c3-c1) */ + MULTIPLY((JLONG)wsptr[5], FIX_0_850430095) + /* sqrt(2) * (-c1+c3+c5+c7) */ + MULTIPLY((JLONG)wsptr[3], -FIX_1_272758580) + /* sqrt(2) * (-c1+c3-c5-c7) */ + MULTIPLY((JLONG)wsptr[1], FIX_3_624509785); /* sqrt(2) * ( c1+c3+c5+c7) */ + + /* Final output stage */ + + outptr[0] = range_limit[(int)DESCALE(tmp10 + tmp0, + CONST_BITS + PASS1_BITS + 3 + 2) & + RANGE_MASK]; + outptr[1] = range_limit[(int)DESCALE(tmp10 - tmp0, + CONST_BITS + PASS1_BITS + 3 + 2) & + RANGE_MASK]; + + wsptr += DCTSIZE; /* advance pointer to next row */ + } +} + + +/* + * Perform dequantization and inverse DCT on one block of coefficients, + * producing a reduced-size 1x1 output block. + */ + +GLOBAL(void) +_jpeg_idct_1x1(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, _JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + int dcval; + ISLOW_MULT_TYPE *quantptr; + _JSAMPLE *range_limit = IDCT_range_limit(cinfo); + SHIFT_TEMPS + + /* We hardly need an inverse DCT routine for this: just take the + * average pixel value, which is one-eighth of the DC coefficient. + */ + quantptr = (ISLOW_MULT_TYPE *)compptr->dct_table; + dcval = DEQUANTIZE(coef_block[0], quantptr[0]); + dcval = (int)DESCALE((JLONG)dcval, 3); + + output_buf[0][output_col] = range_limit[dcval & RANGE_MASK]; +} + +#endif /* IDCT_SCALING_SUPPORTED */ diff --git a/src/spice2x/external/libjpeg-turbo/jinclude.h b/src/spice2x/external/libjpeg-turbo/jinclude.h new file mode 100644 index 0000000..56e7a4b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jinclude.h @@ -0,0 +1,147 @@ +/* + * jinclude.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1994, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022-2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file exists to provide a single place to fix any problems with + * including the wrong system include files. (Common problems are taken + * care of by the standard jconfig symbols, but on really weird systems + * you may have to edit this file.) + * + * NOTE: this file is NOT intended to be included by applications using the + * JPEG library. Most applications need only include jpeglib.h. + */ + +#ifndef __JINCLUDE_H__ +#define __JINCLUDE_H__ + +/* Include auto-config file to find out which system include files we need. */ + +#include "jconfig.h" /* auto configuration options */ +#include "jconfigint.h" +#define JCONFIG_INCLUDED /* so that jpeglib.h doesn't do it again */ + +/* + * Note that the core JPEG library does not require ; + * only the default error handler and data source/destination modules do. + * But we must pull it in because of the references to FILE in jpeglib.h. + * You can remove those references if you want to compile without . + */ + +#include +#include +#include +#include + +/* + * These macros/inline functions facilitate using Microsoft's "safe string" + * functions with Visual Studio builds without the need to scatter #ifdefs + * throughout the code base. + */ + + +#ifdef _MSC_VER + +#define SNPRINTF(str, n, format, ...) \ + _snprintf_s(str, n, _TRUNCATE, format, ##__VA_ARGS__) + +#else + +#define SNPRINTF snprintf + +#endif + + +#ifndef NO_GETENV + +#ifdef _MSC_VER + +static INLINE int GETENV_S(char *buffer, size_t buffer_size, const char *name) +{ + size_t required_size; + + return (int)getenv_s(&required_size, buffer, buffer_size, name); +} + +#else /* _MSC_VER */ + +#include + +/* This provides a similar interface to the Microsoft/C11 getenv_s() function, + * but other than parameter validation, it has no advantages over getenv(). + */ + +static INLINE int GETENV_S(char *buffer, size_t buffer_size, const char *name) +{ + char *env; + + if (!buffer) { + if (buffer_size == 0) + return 0; + else + return (errno = EINVAL); + } + if (buffer_size == 0) + return (errno = EINVAL); + if (!name) { + *buffer = 0; + return 0; + } + + env = getenv(name); + if (!env) + { + *buffer = 0; + return 0; + } + + if (strlen(env) + 1 > buffer_size) { + *buffer = 0; + return ERANGE; + } + + strncpy(buffer, env, buffer_size); + + return 0; +} + +#endif /* _MSC_VER */ + +#endif /* NO_GETENV */ + + +#ifndef NO_PUTENV + +#ifdef _WIN32 + +#define PUTENV_S(name, value) _putenv_s(name, value) + +#else + +#include + +/* This provides a similar interface to the Microsoft _putenv_s() function, but + * other than parameter validation, it has no advantages over setenv(). + */ + +static INLINE int PUTENV_S(const char *name, const char *value) +{ + if (!name || !value) + return (errno = EINVAL); + + setenv(name, value, 1); + + return errno; +} + +#endif /* _WIN32 */ + +#endif /* NO_PUTENV */ + + +#endif /* JINCLUDE_H */ diff --git a/src/spice2x/external/libjpeg-turbo/jlossls.h b/src/spice2x/external/libjpeg-turbo/jlossls.h new file mode 100644 index 0000000..ce41704 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jlossls.h @@ -0,0 +1,101 @@ +/* + * jlossls.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1998, Thomas G. Lane. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This include file contains common declarations for the lossless JPEG + * codec modules. + */ + +#ifndef JLOSSLS_H +#define JLOSSLS_H + +#if defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) + +#define JPEG_INTERNALS +#include "jpeglib.h" +#include "jsamplecomp.h" + + +#define ALLOC_DARRAY(pool_id, diffsperrow, numrows) \ + (JDIFFARRAY)(*cinfo->mem->alloc_sarray) \ + ((j_common_ptr)cinfo, pool_id, \ + (diffsperrow) * sizeof(JDIFF) / sizeof(_JSAMPLE), numrows) + + +/* + * Table H.1: Predictors for lossless coding. + */ + +#define PREDICTOR1 Ra +#define PREDICTOR2 Rb +#define PREDICTOR3 Rc +#define PREDICTOR4 (int)((JLONG)Ra + (JLONG)Rb - (JLONG)Rc) +#define PREDICTOR5 (int)((JLONG)Ra + RIGHT_SHIFT((JLONG)Rb - (JLONG)Rc, 1)) +#define PREDICTOR6 (int)((JLONG)Rb + RIGHT_SHIFT((JLONG)Ra - (JLONG)Rc, 1)) +#define PREDICTOR7 (int)RIGHT_SHIFT((JLONG)Ra + (JLONG)Rb, 1) + +#endif + + +#ifdef C_LOSSLESS_SUPPORTED + +typedef void (*predict_difference_method_ptr) (j_compress_ptr cinfo, int ci, + _JSAMPROW input_buf, + _JSAMPROW prev_row, + JDIFFROW diff_buf, + JDIMENSION width); + +/* Lossless compressor */ +typedef struct { + struct jpeg_forward_dct pub; /* public fields */ + + /* It is useful to allow each component to have a separate diff method. */ + predict_difference_method_ptr predict_difference[MAX_COMPONENTS]; + + /* MCU rows left in the restart interval for each component */ + unsigned int restart_rows_to_go[MAX_COMPONENTS]; + + /* Sample scaling */ + void (*scaler_scale) (j_compress_ptr cinfo, _JSAMPROW input_buf, + _JSAMPROW output_buf, JDIMENSION width); +} jpeg_lossless_compressor; + +typedef jpeg_lossless_compressor *lossless_comp_ptr; + +#endif /* C_LOSSLESS_SUPPORTED */ + + +#ifdef D_LOSSLESS_SUPPORTED + +typedef void (*predict_undifference_method_ptr) (j_decompress_ptr cinfo, + int comp_index, + JDIFFROW diff_buf, + JDIFFROW prev_row, + JDIFFROW undiff_buf, + JDIMENSION width); + +/* Lossless decompressor */ +typedef struct { + struct jpeg_inverse_dct pub; /* public fields */ + + /* It is useful to allow each component to have a separate undiff method. */ + predict_undifference_method_ptr predict_undifference[MAX_COMPONENTS]; + + /* Sample scaling */ + void (*scaler_scale) (j_decompress_ptr cinfo, JDIFFROW diff_buf, + _JSAMPROW output_buf, JDIMENSION width); +} jpeg_lossless_decompressor; + +typedef jpeg_lossless_decompressor *lossless_decomp_ptr; + +#endif /* D_LOSSLESS_SUPPORTED */ + +#endif /* JLOSSLS_H */ diff --git a/src/spice2x/external/libjpeg-turbo/jmemmgr.c b/src/spice2x/external/libjpeg-turbo/jmemmgr.c new file mode 100644 index 0000000..dca8f5c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jmemmgr.c @@ -0,0 +1,1289 @@ +/* + * jmemmgr.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2016, 2021-2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains the JPEG system-independent memory management + * routines. This code is usable across a wide variety of machines; most + * of the system dependencies have been isolated in a separate file. + * The major functions provided here are: + * * pool-based allocation and freeing of memory; + * * policy decisions about how to divide available memory among the + * virtual arrays; + * * control logic for swapping virtual arrays between main memory and + * backing storage. + * The separate system-dependent file provides the actual backing-storage + * access code, and it contains the policy decision about how much total + * main memory to use. + * This file is system-dependent in the sense that some of its functions + * are unnecessary in some systems. For example, if there is enough virtual + * memory so that backing storage will never be used, much of the virtual + * array control logic could be removed. (Of course, if you have that much + * memory then you shouldn't care about a little bit of unused code...) + */ + +#define JPEG_INTERNALS +#define AM_MEMORY_MANAGER /* we define jvirt_Xarray_control structs */ +#include "jinclude.h" +#include "jpeglib.h" +#include "jmemsys.h" /* import the system-dependent declarations */ +#if !defined(_MSC_VER) || _MSC_VER > 1600 +#include +#endif +#include + + +LOCAL(size_t) +round_up_pow2(size_t a, size_t b) +/* a rounded up to the next multiple of b, i.e. ceil(a/b)*b */ +/* Assumes a >= 0, b > 0, and b is a power of 2 */ +{ + return ((a + b - 1) & (~(b - 1))); +} + + +/* + * Some important notes: + * The allocation routines provided here must never return NULL. + * They should exit to error_exit if unsuccessful. + * + * It's not a good idea to try to merge the sarray and barray routines, + * even though they are textually almost the same, because samples are + * usually stored as bytes while coefficients are shorts or ints. Thus, + * in machines where byte pointers have a different representation from + * word pointers, the resulting machine code could not be the same. + */ + + +/* + * Many machines require storage alignment: longs must start on 4-byte + * boundaries, doubles on 8-byte boundaries, etc. On such machines, malloc() + * always returns pointers that are multiples of the worst-case alignment + * requirement, and we had better do so too. + * There isn't any really portable way to determine the worst-case alignment + * requirement. This module assumes that the alignment requirement is + * multiples of ALIGN_SIZE. + * By default, we define ALIGN_SIZE as the maximum of sizeof(double) and + * sizeof(void *). This is necessary on some workstations (where doubles + * really do need 8-byte alignment) and will work fine on nearly everything. + * We use the maximum of sizeof(double) and sizeof(void *) since sizeof(double) + * may be insufficient, for example, on CHERI-enabled platforms with 16-byte + * pointers and a 16-byte alignment requirement. If your machine has lesser + * alignment needs, you can save a few bytes by making ALIGN_SIZE smaller. + * The only place I know of where this will NOT work is certain Macintosh + * 680x0 compilers that define double as a 10-byte IEEE extended float. + * Doing 10-byte alignment is counterproductive because longwords won't be + * aligned well. Put "#define ALIGN_SIZE 4" in jconfig.h if you have + * such a compiler. + */ + +#ifndef ALIGN_SIZE /* so can override from jconfig.h */ +#ifndef WITH_SIMD +#define ALIGN_SIZE MAX(sizeof(void *), sizeof(double)) +#else +#define ALIGN_SIZE 32 /* Most of the SIMD instructions we support require + 16-byte (128-bit) alignment, but AVX2 requires + 32-byte alignment. */ +#endif +#endif + +/* + * We allocate objects from "pools", where each pool is gotten with a single + * request to jpeg_get_small() or jpeg_get_large(). There is no per-object + * overhead within a pool, except for alignment padding. Each pool has a + * header with a link to the next pool of the same class. + * Small and large pool headers are identical. + */ + +typedef struct small_pool_struct *small_pool_ptr; + +typedef struct small_pool_struct { + small_pool_ptr next; /* next in list of pools */ + size_t bytes_used; /* how many bytes already used within pool */ + size_t bytes_left; /* bytes still available in this pool */ +} small_pool_hdr; + +typedef struct large_pool_struct *large_pool_ptr; + +typedef struct large_pool_struct { + large_pool_ptr next; /* next in list of pools */ + size_t bytes_used; /* how many bytes already used within pool */ + size_t bytes_left; /* bytes still available in this pool */ +} large_pool_hdr; + +/* + * Here is the full definition of a memory manager object. + */ + +typedef struct { + struct jpeg_memory_mgr pub; /* public fields */ + + /* Each pool identifier (lifetime class) names a linked list of pools. */ + small_pool_ptr small_list[JPOOL_NUMPOOLS]; + large_pool_ptr large_list[JPOOL_NUMPOOLS]; + + /* Since we only have one lifetime class of virtual arrays, only one + * linked list is necessary (for each datatype). Note that the virtual + * array control blocks being linked together are actually stored somewhere + * in the small-pool list. + */ + jvirt_sarray_ptr virt_sarray_list; + jvirt_barray_ptr virt_barray_list; + + /* This counts total space obtained from jpeg_get_small/large */ + size_t total_space_allocated; + + /* alloc_sarray and alloc_barray set this value for use by virtual + * array routines. + */ + JDIMENSION last_rowsperchunk; /* from most recent alloc_sarray/barray */ +} my_memory_mgr; + +typedef my_memory_mgr *my_mem_ptr; + + +/* + * The control blocks for virtual arrays. + * Note that these blocks are allocated in the "small" pool area. + * System-dependent info for the associated backing store (if any) is hidden + * inside the backing_store_info struct. + */ + +struct jvirt_sarray_control { + JSAMPARRAY mem_buffer; /* => the in-memory buffer (if + cinfo->data_precision is 12, then this is + actually a J12SAMPARRAY) */ + JDIMENSION rows_in_array; /* total virtual array height */ + JDIMENSION samplesperrow; /* width of array (and of memory buffer) */ + JDIMENSION maxaccess; /* max rows accessed by access_virt_sarray */ + JDIMENSION rows_in_mem; /* height of memory buffer */ + JDIMENSION rowsperchunk; /* allocation chunk size in mem_buffer */ + JDIMENSION cur_start_row; /* first logical row # in the buffer */ + JDIMENSION first_undef_row; /* row # of first uninitialized row */ + boolean pre_zero; /* pre-zero mode requested? */ + boolean dirty; /* do current buffer contents need written? */ + boolean b_s_open; /* is backing-store data valid? */ + jvirt_sarray_ptr next; /* link to next virtual sarray control block */ + backing_store_info b_s_info; /* System-dependent control info */ +}; + +struct jvirt_barray_control { + JBLOCKARRAY mem_buffer; /* => the in-memory buffer */ + JDIMENSION rows_in_array; /* total virtual array height */ + JDIMENSION blocksperrow; /* width of array (and of memory buffer) */ + JDIMENSION maxaccess; /* max rows accessed by access_virt_barray */ + JDIMENSION rows_in_mem; /* height of memory buffer */ + JDIMENSION rowsperchunk; /* allocation chunk size in mem_buffer */ + JDIMENSION cur_start_row; /* first logical row # in the buffer */ + JDIMENSION first_undef_row; /* row # of first uninitialized row */ + boolean pre_zero; /* pre-zero mode requested? */ + boolean dirty; /* do current buffer contents need written? */ + boolean b_s_open; /* is backing-store data valid? */ + jvirt_barray_ptr next; /* link to next virtual barray control block */ + backing_store_info b_s_info; /* System-dependent control info */ +}; + + +#ifdef MEM_STATS /* optional extra stuff for statistics */ + +LOCAL(void) +print_mem_stats(j_common_ptr cinfo, int pool_id) +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + small_pool_ptr shdr_ptr; + large_pool_ptr lhdr_ptr; + + /* Since this is only a debugging stub, we can cheat a little by using + * fprintf directly rather than going through the trace message code. + * This is helpful because message parm array can't handle longs. + */ + fprintf(stderr, "Freeing pool %d, total space = %ld\n", + pool_id, mem->total_space_allocated); + + for (lhdr_ptr = mem->large_list[pool_id]; lhdr_ptr != NULL; + lhdr_ptr = lhdr_ptr->next) { + fprintf(stderr, " Large chunk used %ld\n", (long)lhdr_ptr->bytes_used); + } + + for (shdr_ptr = mem->small_list[pool_id]; shdr_ptr != NULL; + shdr_ptr = shdr_ptr->next) { + fprintf(stderr, " Small chunk used %ld free %ld\n", + (long)shdr_ptr->bytes_used, (long)shdr_ptr->bytes_left); + } +} + +#endif /* MEM_STATS */ + + +LOCAL(void) +out_of_memory(j_common_ptr cinfo, int which) +/* Report an out-of-memory error and stop execution */ +/* If we compiled MEM_STATS support, report alloc requests before dying */ +{ +#ifdef MEM_STATS + cinfo->err->trace_level = 2; /* force self_destruct to report stats */ +#endif + ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, which); +} + + +/* + * Allocation of "small" objects. + * + * For these, we use pooled storage. When a new pool must be created, + * we try to get enough space for the current request plus a "slop" factor, + * where the slop will be the amount of leftover space in the new pool. + * The speed vs. space tradeoff is largely determined by the slop values. + * A different slop value is provided for each pool class (lifetime), + * and we also distinguish the first pool of a class from later ones. + * NOTE: the values given work fairly well on both 16- and 32-bit-int + * machines, but may be too small if longs are 64 bits or more. + * + * Since we do not know what alignment malloc() gives us, we have to + * allocate ALIGN_SIZE-1 extra space per pool to have room for alignment + * adjustment. + */ + +static const size_t first_pool_slop[JPOOL_NUMPOOLS] = { + 1600, /* first PERMANENT pool */ + 16000 /* first IMAGE pool */ +}; + +static const size_t extra_pool_slop[JPOOL_NUMPOOLS] = { + 0, /* additional PERMANENT pools */ + 5000 /* additional IMAGE pools */ +}; + +#define MIN_SLOP 50 /* greater than 0 to avoid futile looping */ + + +METHODDEF(void *) +alloc_small(j_common_ptr cinfo, int pool_id, size_t sizeofobject) +/* Allocate a "small" object */ +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + small_pool_ptr hdr_ptr, prev_hdr_ptr; + char *data_ptr; + size_t min_request, slop; + + /* + * Round up the requested size to a multiple of ALIGN_SIZE in order + * to assure alignment for the next object allocated in the same pool + * and so that algorithms can straddle outside the proper area up + * to the next alignment. + */ + if (sizeofobject > MAX_ALLOC_CHUNK) { + /* This prevents overflow/wrap-around in round_up_pow2() if sizeofobject + is close to SIZE_MAX. */ + out_of_memory(cinfo, 7); + } + sizeofobject = round_up_pow2(sizeofobject, ALIGN_SIZE); + + /* Check for unsatisfiable request (do now to ensure no overflow below) */ + if ((sizeof(small_pool_hdr) + sizeofobject + ALIGN_SIZE - 1) > + MAX_ALLOC_CHUNK) + out_of_memory(cinfo, 1); /* request exceeds malloc's ability */ + + /* See if space is available in any existing pool */ + if (pool_id < 0 || pool_id >= JPOOL_NUMPOOLS) + ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */ + prev_hdr_ptr = NULL; + hdr_ptr = mem->small_list[pool_id]; + while (hdr_ptr != NULL) { + if (hdr_ptr->bytes_left >= sizeofobject) + break; /* found pool with enough space */ + prev_hdr_ptr = hdr_ptr; + hdr_ptr = hdr_ptr->next; + } + + /* Time to make a new pool? */ + if (hdr_ptr == NULL) { + /* min_request is what we need now, slop is what will be leftover */ + min_request = sizeof(small_pool_hdr) + sizeofobject + ALIGN_SIZE - 1; + if (prev_hdr_ptr == NULL) /* first pool in class? */ + slop = first_pool_slop[pool_id]; + else + slop = extra_pool_slop[pool_id]; + /* Don't ask for more than MAX_ALLOC_CHUNK */ + if (slop > (size_t)(MAX_ALLOC_CHUNK - min_request)) + slop = (size_t)(MAX_ALLOC_CHUNK - min_request); + /* Try to get space, if fail reduce slop and try again */ + for (;;) { + hdr_ptr = (small_pool_ptr)jpeg_get_small(cinfo, min_request + slop); + if (hdr_ptr != NULL) + break; + slop /= 2; + if (slop < MIN_SLOP) /* give up when it gets real small */ + out_of_memory(cinfo, 2); /* jpeg_get_small failed */ + } + mem->total_space_allocated += min_request + slop; + /* Success, initialize the new pool header and add to end of list */ + hdr_ptr->next = NULL; + hdr_ptr->bytes_used = 0; + hdr_ptr->bytes_left = sizeofobject + slop; + if (prev_hdr_ptr == NULL) /* first pool in class? */ + mem->small_list[pool_id] = hdr_ptr; + else + prev_hdr_ptr->next = hdr_ptr; + } + + /* OK, allocate the object from the current pool */ + data_ptr = (char *)hdr_ptr; /* point to first data byte in pool... */ + data_ptr += sizeof(small_pool_hdr); /* ...by skipping the header... */ + if ((size_t)data_ptr % ALIGN_SIZE) /* ...and adjust for alignment */ + data_ptr += ALIGN_SIZE - (size_t)data_ptr % ALIGN_SIZE; + data_ptr += hdr_ptr->bytes_used; /* point to place for object */ + hdr_ptr->bytes_used += sizeofobject; + hdr_ptr->bytes_left -= sizeofobject; + + return (void *)data_ptr; +} + + +/* + * Allocation of "large" objects. + * + * The external semantics of these are the same as "small" objects. However, + * the pool management heuristics are quite different. We assume that each + * request is large enough that it may as well be passed directly to + * jpeg_get_large; the pool management just links everything together + * so that we can free it all on demand. + * Note: the major use of "large" objects is in + * JSAMPARRAY/J12SAMPARRAY/J16SAMPARRAY and JBLOCKARRAY structures. The + * routines that create these structures (see below) deliberately bunch rows + * together to ensure a large request size. + */ + +METHODDEF(void *) +alloc_large(j_common_ptr cinfo, int pool_id, size_t sizeofobject) +/* Allocate a "large" object */ +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + large_pool_ptr hdr_ptr; + char *data_ptr; + + /* + * Round up the requested size to a multiple of ALIGN_SIZE so that + * algorithms can straddle outside the proper area up to the next + * alignment. + */ + if (sizeofobject > MAX_ALLOC_CHUNK) { + /* This prevents overflow/wrap-around in round_up_pow2() if sizeofobject + is close to SIZE_MAX. */ + out_of_memory(cinfo, 8); + } + sizeofobject = round_up_pow2(sizeofobject, ALIGN_SIZE); + + /* Check for unsatisfiable request (do now to ensure no overflow below) */ + if ((sizeof(large_pool_hdr) + sizeofobject + ALIGN_SIZE - 1) > + MAX_ALLOC_CHUNK) + out_of_memory(cinfo, 3); /* request exceeds malloc's ability */ + + /* Always make a new pool */ + if (pool_id < 0 || pool_id >= JPOOL_NUMPOOLS) + ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */ + + hdr_ptr = (large_pool_ptr)jpeg_get_large(cinfo, sizeofobject + + sizeof(large_pool_hdr) + + ALIGN_SIZE - 1); + if (hdr_ptr == NULL) + out_of_memory(cinfo, 4); /* jpeg_get_large failed */ + mem->total_space_allocated += sizeofobject + sizeof(large_pool_hdr) + + ALIGN_SIZE - 1; + + /* Success, initialize the new pool header and add to list */ + hdr_ptr->next = mem->large_list[pool_id]; + /* We maintain space counts in each pool header for statistical purposes, + * even though they are not needed for allocation. + */ + hdr_ptr->bytes_used = sizeofobject; + hdr_ptr->bytes_left = 0; + mem->large_list[pool_id] = hdr_ptr; + + data_ptr = (char *)hdr_ptr; /* point to first data byte in pool... */ + data_ptr += sizeof(small_pool_hdr); /* ...by skipping the header... */ + if ((size_t)data_ptr % ALIGN_SIZE) /* ...and adjust for alignment */ + data_ptr += ALIGN_SIZE - (size_t)data_ptr % ALIGN_SIZE; + + return (void *)data_ptr; +} + + +/* + * Creation of 2-D sample arrays. + * + * To minimize allocation overhead and to allow I/O of large contiguous + * blocks, we allocate the sample rows in groups of as many rows as possible + * without exceeding MAX_ALLOC_CHUNK total bytes per allocation request. + * NB: the virtual array control routines, later in this file, know about + * this chunking of rows. The rowsperchunk value is left in the mem manager + * object so that it can be saved away if this sarray is the workspace for + * a virtual array. + * + * Since we are often upsampling with a factor 2, we align the size (not + * the start) to 2 * ALIGN_SIZE so that the upsampling routines don't have + * to be as careful about size. + */ + +METHODDEF(JSAMPARRAY) +alloc_sarray(j_common_ptr cinfo, int pool_id, JDIMENSION samplesperrow, + JDIMENSION numrows) +/* Allocate a 2-D sample array */ +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + JSAMPARRAY result; + JSAMPROW workspace; + JDIMENSION rowsperchunk, currow, i; + long ltemp; + J12SAMPARRAY result12; + J12SAMPROW workspace12; +#if defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) + J16SAMPARRAY result16; + J16SAMPROW workspace16; +#endif + int data_precision = cinfo->is_decompressor ? + ((j_decompress_ptr)cinfo)->data_precision : + ((j_compress_ptr)cinfo)->data_precision; + size_t sample_size = data_precision == 16 ? + sizeof(J16SAMPLE) : (data_precision == 12 ? + sizeof(J12SAMPLE) : + sizeof(JSAMPLE)); + + /* Make sure each row is properly aligned */ + if ((ALIGN_SIZE % sample_size) != 0) + out_of_memory(cinfo, 5); /* safety check */ + + if (samplesperrow > MAX_ALLOC_CHUNK) { + /* This prevents overflow/wrap-around in round_up_pow2() if sizeofobject + is close to SIZE_MAX. */ + out_of_memory(cinfo, 9); + } + samplesperrow = (JDIMENSION)round_up_pow2(samplesperrow, (2 * ALIGN_SIZE) / + sample_size); + + /* Calculate max # of rows allowed in one allocation chunk */ + ltemp = (MAX_ALLOC_CHUNK - sizeof(large_pool_hdr)) / + ((long)samplesperrow * (long)sample_size); + if (ltemp <= 0) + ERREXIT(cinfo, JERR_WIDTH_OVERFLOW); + if (ltemp < (long)numrows) + rowsperchunk = (JDIMENSION)ltemp; + else + rowsperchunk = numrows; + mem->last_rowsperchunk = rowsperchunk; + + if (data_precision == 16) { +#if defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) + /* Get space for row pointers (small object) */ + result16 = (J16SAMPARRAY)alloc_small(cinfo, pool_id, + (size_t)(numrows * + sizeof(J16SAMPROW))); + + /* Get the rows themselves (large objects) */ + currow = 0; + while (currow < numrows) { + rowsperchunk = MIN(rowsperchunk, numrows - currow); + workspace16 = (J16SAMPROW)alloc_large(cinfo, pool_id, + (size_t)((size_t)rowsperchunk * (size_t)samplesperrow * sample_size)); + for (i = rowsperchunk; i > 0; i--) { + result16[currow++] = workspace16; + workspace16 += samplesperrow; + } + } + + return (JSAMPARRAY)result16; +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, data_precision); + return NULL; +#endif + } else if (data_precision == 12) { + /* Get space for row pointers (small object) */ + result12 = (J12SAMPARRAY)alloc_small(cinfo, pool_id, + (size_t)(numrows * + sizeof(J12SAMPROW))); + + /* Get the rows themselves (large objects) */ + currow = 0; + while (currow < numrows) { + rowsperchunk = MIN(rowsperchunk, numrows - currow); + workspace12 = (J12SAMPROW)alloc_large(cinfo, pool_id, + (size_t)((size_t)rowsperchunk * (size_t)samplesperrow * sample_size)); + for (i = rowsperchunk; i > 0; i--) { + result12[currow++] = workspace12; + workspace12 += samplesperrow; + } + } + + return (JSAMPARRAY)result12; + } else { + /* Get space for row pointers (small object) */ + result = (JSAMPARRAY)alloc_small(cinfo, pool_id, + (size_t)(numrows * sizeof(JSAMPROW))); + + /* Get the rows themselves (large objects) */ + currow = 0; + while (currow < numrows) { + rowsperchunk = MIN(rowsperchunk, numrows - currow); + workspace = (JSAMPROW)alloc_large(cinfo, pool_id, + (size_t)((size_t)rowsperchunk * (size_t)samplesperrow * sample_size)); + for (i = rowsperchunk; i > 0; i--) { + result[currow++] = workspace; + workspace += samplesperrow; + } + } + + return result; + } +} + + +/* + * Creation of 2-D coefficient-block arrays. + * This is essentially the same as the code for sample arrays, above. + */ + +METHODDEF(JBLOCKARRAY) +alloc_barray(j_common_ptr cinfo, int pool_id, JDIMENSION blocksperrow, + JDIMENSION numrows) +/* Allocate a 2-D coefficient-block array */ +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + JBLOCKARRAY result; + JBLOCKROW workspace; + JDIMENSION rowsperchunk, currow, i; + long ltemp; + + /* Make sure each row is properly aligned */ + if ((sizeof(JBLOCK) % ALIGN_SIZE) != 0) + out_of_memory(cinfo, 6); /* safety check */ + + /* Calculate max # of rows allowed in one allocation chunk */ + ltemp = (MAX_ALLOC_CHUNK - sizeof(large_pool_hdr)) / + ((long)blocksperrow * sizeof(JBLOCK)); + if (ltemp <= 0) + ERREXIT(cinfo, JERR_WIDTH_OVERFLOW); + if (ltemp < (long)numrows) + rowsperchunk = (JDIMENSION)ltemp; + else + rowsperchunk = numrows; + mem->last_rowsperchunk = rowsperchunk; + + /* Get space for row pointers (small object) */ + result = (JBLOCKARRAY)alloc_small(cinfo, pool_id, + (size_t)(numrows * sizeof(JBLOCKROW))); + + /* Get the rows themselves (large objects) */ + currow = 0; + while (currow < numrows) { + rowsperchunk = MIN(rowsperchunk, numrows - currow); + workspace = (JBLOCKROW)alloc_large(cinfo, pool_id, + (size_t)((size_t)rowsperchunk * (size_t)blocksperrow * + sizeof(JBLOCK))); + for (i = rowsperchunk; i > 0; i--) { + result[currow++] = workspace; + workspace += blocksperrow; + } + } + + return result; +} + + +/* + * About virtual array management: + * + * The above "normal" array routines are only used to allocate strip buffers + * (as wide as the image, but just a few rows high). Full-image-sized buffers + * are handled as "virtual" arrays. The array is still accessed a strip at a + * time, but the memory manager must save the whole array for repeated + * accesses. The intended implementation is that there is a strip buffer in + * memory (as high as is possible given the desired memory limit), plus a + * backing file that holds the rest of the array. + * + * The request_virt_array routines are told the total size of the image and + * the maximum number of rows that will be accessed at once. The in-memory + * buffer must be at least as large as the maxaccess value. + * + * The request routines create control blocks but not the in-memory buffers. + * That is postponed until realize_virt_arrays is called. At that time the + * total amount of space needed is known (approximately, anyway), so free + * memory can be divided up fairly. + * + * The access_virt_array routines are responsible for making a specific strip + * area accessible (after reading or writing the backing file, if necessary). + * Note that the access routines are told whether the caller intends to modify + * the accessed strip; during a read-only pass this saves having to rewrite + * data to disk. The access routines are also responsible for pre-zeroing + * any newly accessed rows, if pre-zeroing was requested. + * + * In current usage, the access requests are usually for nonoverlapping + * strips; that is, successive access start_row numbers differ by exactly + * num_rows = maxaccess. This means we can get good performance with simple + * buffer dump/reload logic, by making the in-memory buffer be a multiple + * of the access height; then there will never be accesses across bufferload + * boundaries. The code will still work with overlapping access requests, + * but it doesn't handle bufferload overlaps very efficiently. + */ + + +METHODDEF(jvirt_sarray_ptr) +request_virt_sarray(j_common_ptr cinfo, int pool_id, boolean pre_zero, + JDIMENSION samplesperrow, JDIMENSION numrows, + JDIMENSION maxaccess) +/* Request a virtual 2-D sample array */ +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + jvirt_sarray_ptr result; + + /* Only IMAGE-lifetime virtual arrays are currently supported */ + if (pool_id != JPOOL_IMAGE) + ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */ + + /* get control block */ + result = (jvirt_sarray_ptr)alloc_small(cinfo, pool_id, + sizeof(struct jvirt_sarray_control)); + + result->mem_buffer = NULL; /* marks array not yet realized */ + result->rows_in_array = numrows; + result->samplesperrow = samplesperrow; + result->maxaccess = maxaccess; + result->pre_zero = pre_zero; + result->b_s_open = FALSE; /* no associated backing-store object */ + result->next = mem->virt_sarray_list; /* add to list of virtual arrays */ + mem->virt_sarray_list = result; + + return result; +} + + +METHODDEF(jvirt_barray_ptr) +request_virt_barray(j_common_ptr cinfo, int pool_id, boolean pre_zero, + JDIMENSION blocksperrow, JDIMENSION numrows, + JDIMENSION maxaccess) +/* Request a virtual 2-D coefficient-block array */ +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + jvirt_barray_ptr result; + + /* Only IMAGE-lifetime virtual arrays are currently supported */ + if (pool_id != JPOOL_IMAGE) + ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */ + + /* get control block */ + result = (jvirt_barray_ptr)alloc_small(cinfo, pool_id, + sizeof(struct jvirt_barray_control)); + + result->mem_buffer = NULL; /* marks array not yet realized */ + result->rows_in_array = numrows; + result->blocksperrow = blocksperrow; + result->maxaccess = maxaccess; + result->pre_zero = pre_zero; + result->b_s_open = FALSE; /* no associated backing-store object */ + result->next = mem->virt_barray_list; /* add to list of virtual arrays */ + mem->virt_barray_list = result; + + return result; +} + + +METHODDEF(void) +realize_virt_arrays(j_common_ptr cinfo) +/* Allocate the in-memory buffers for any unrealized virtual arrays */ +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + size_t space_per_minheight, maximum_space, avail_mem; + size_t minheights, max_minheights; + jvirt_sarray_ptr sptr; + jvirt_barray_ptr bptr; + int data_precision = cinfo->is_decompressor ? + ((j_decompress_ptr)cinfo)->data_precision : + ((j_compress_ptr)cinfo)->data_precision; + size_t sample_size = data_precision == 16 ? + sizeof(J16SAMPLE) : (data_precision == 12 ? + sizeof(J12SAMPLE) : + sizeof(JSAMPLE)); + + /* Compute the minimum space needed (maxaccess rows in each buffer) + * and the maximum space needed (full image height in each buffer). + * These may be of use to the system-dependent jpeg_mem_available routine. + */ + space_per_minheight = 0; + maximum_space = 0; + for (sptr = mem->virt_sarray_list; sptr != NULL; sptr = sptr->next) { + if (sptr->mem_buffer == NULL) { /* if not realized yet */ + size_t new_space = (long)sptr->rows_in_array * + (long)sptr->samplesperrow * sample_size; + + space_per_minheight += (long)sptr->maxaccess * + (long)sptr->samplesperrow * sample_size; + if (SIZE_MAX - maximum_space < new_space) + out_of_memory(cinfo, 10); + maximum_space += new_space; + } + } + for (bptr = mem->virt_barray_list; bptr != NULL; bptr = bptr->next) { + if (bptr->mem_buffer == NULL) { /* if not realized yet */ + size_t new_space = (long)bptr->rows_in_array * + (long)bptr->blocksperrow * sizeof(JBLOCK); + + space_per_minheight += (long)bptr->maxaccess * + (long)bptr->blocksperrow * sizeof(JBLOCK); + if (SIZE_MAX - maximum_space < new_space) + out_of_memory(cinfo, 11); + maximum_space += new_space; + } + } + + if (space_per_minheight <= 0) + return; /* no unrealized arrays, no work */ + + /* Determine amount of memory to actually use; this is system-dependent. */ + avail_mem = jpeg_mem_available(cinfo, space_per_minheight, maximum_space, + mem->total_space_allocated); + + /* If the maximum space needed is available, make all the buffers full + * height; otherwise parcel it out with the same number of minheights + * in each buffer. + */ + if (avail_mem >= maximum_space) + max_minheights = 1000000000L; + else { + max_minheights = avail_mem / space_per_minheight; + /* If there doesn't seem to be enough space, try to get the minimum + * anyway. This allows a "stub" implementation of jpeg_mem_available(). + */ + if (max_minheights <= 0) + max_minheights = 1; + } + + /* Allocate the in-memory buffers and initialize backing store as needed. */ + + for (sptr = mem->virt_sarray_list; sptr != NULL; sptr = sptr->next) { + if (sptr->mem_buffer == NULL) { /* if not realized yet */ + minheights = ((long)sptr->rows_in_array - 1L) / sptr->maxaccess + 1L; + if (minheights <= max_minheights) { + /* This buffer fits in memory */ + sptr->rows_in_mem = sptr->rows_in_array; + } else { + /* It doesn't fit in memory, create backing store. */ + sptr->rows_in_mem = (JDIMENSION)(max_minheights * sptr->maxaccess); + jpeg_open_backing_store(cinfo, &sptr->b_s_info, + (long)sptr->rows_in_array * + (long)sptr->samplesperrow * + (long)sample_size); + sptr->b_s_open = TRUE; + } + sptr->mem_buffer = alloc_sarray(cinfo, JPOOL_IMAGE, + sptr->samplesperrow, sptr->rows_in_mem); + sptr->rowsperchunk = mem->last_rowsperchunk; + sptr->cur_start_row = 0; + sptr->first_undef_row = 0; + sptr->dirty = FALSE; + } + } + + for (bptr = mem->virt_barray_list; bptr != NULL; bptr = bptr->next) { + if (bptr->mem_buffer == NULL) { /* if not realized yet */ + minheights = ((long)bptr->rows_in_array - 1L) / bptr->maxaccess + 1L; + if (minheights <= max_minheights) { + /* This buffer fits in memory */ + bptr->rows_in_mem = bptr->rows_in_array; + } else { + /* It doesn't fit in memory, create backing store. */ + bptr->rows_in_mem = (JDIMENSION)(max_minheights * bptr->maxaccess); + jpeg_open_backing_store(cinfo, &bptr->b_s_info, + (long)bptr->rows_in_array * + (long)bptr->blocksperrow * + (long)sizeof(JBLOCK)); + bptr->b_s_open = TRUE; + } + bptr->mem_buffer = alloc_barray(cinfo, JPOOL_IMAGE, + bptr->blocksperrow, bptr->rows_in_mem); + bptr->rowsperchunk = mem->last_rowsperchunk; + bptr->cur_start_row = 0; + bptr->first_undef_row = 0; + bptr->dirty = FALSE; + } + } +} + + +LOCAL(void) +do_sarray_io(j_common_ptr cinfo, jvirt_sarray_ptr ptr, boolean writing) +/* Do backing store read or write of a virtual sample array */ +{ + long bytesperrow, file_offset, byte_count, rows, thisrow, i; + int data_precision = cinfo->is_decompressor ? + ((j_decompress_ptr)cinfo)->data_precision : + ((j_compress_ptr)cinfo)->data_precision; + size_t sample_size = data_precision == 16 ? + sizeof(J16SAMPLE) : (data_precision == 12 ? + sizeof(J12SAMPLE) : + sizeof(JSAMPLE)); + + bytesperrow = (long)ptr->samplesperrow * (long)sample_size; + file_offset = ptr->cur_start_row * bytesperrow; + /* Loop to read or write each allocation chunk in mem_buffer */ + for (i = 0; i < (long)ptr->rows_in_mem; i += ptr->rowsperchunk) { + /* One chunk, but check for short chunk at end of buffer */ + rows = MIN((long)ptr->rowsperchunk, (long)ptr->rows_in_mem - i); + /* Transfer no more than is currently defined */ + thisrow = (long)ptr->cur_start_row + i; + rows = MIN(rows, (long)ptr->first_undef_row - thisrow); + /* Transfer no more than fits in file */ + rows = MIN(rows, (long)ptr->rows_in_array - thisrow); + if (rows <= 0) /* this chunk might be past end of file! */ + break; + byte_count = rows * bytesperrow; + if (data_precision == 16) { +#if defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) + J16SAMPARRAY mem_buffer16 = (J16SAMPARRAY)ptr->mem_buffer; + + if (writing) + (*ptr->b_s_info.write_backing_store) (cinfo, &ptr->b_s_info, + (void *)mem_buffer16[i], + file_offset, byte_count); + else + (*ptr->b_s_info.read_backing_store) (cinfo, &ptr->b_s_info, + (void *)mem_buffer16[i], + file_offset, byte_count); +#else + ERREXIT1(cinfo, JERR_BAD_PRECISION, data_precision); +#endif + } else if (data_precision == 12) { + J12SAMPARRAY mem_buffer12 = (J12SAMPARRAY)ptr->mem_buffer; + + if (writing) + (*ptr->b_s_info.write_backing_store) (cinfo, &ptr->b_s_info, + (void *)mem_buffer12[i], + file_offset, byte_count); + else + (*ptr->b_s_info.read_backing_store) (cinfo, &ptr->b_s_info, + (void *)mem_buffer12[i], + file_offset, byte_count); + } else { + if (writing) + (*ptr->b_s_info.write_backing_store) (cinfo, &ptr->b_s_info, + (void *)ptr->mem_buffer[i], + file_offset, byte_count); + else + (*ptr->b_s_info.read_backing_store) (cinfo, &ptr->b_s_info, + (void *)ptr->mem_buffer[i], + file_offset, byte_count); + } + file_offset += byte_count; + } +} + + +LOCAL(void) +do_barray_io(j_common_ptr cinfo, jvirt_barray_ptr ptr, boolean writing) +/* Do backing store read or write of a virtual coefficient-block array */ +{ + long bytesperrow, file_offset, byte_count, rows, thisrow, i; + + bytesperrow = (long)ptr->blocksperrow * sizeof(JBLOCK); + file_offset = ptr->cur_start_row * bytesperrow; + /* Loop to read or write each allocation chunk in mem_buffer */ + for (i = 0; i < (long)ptr->rows_in_mem; i += ptr->rowsperchunk) { + /* One chunk, but check for short chunk at end of buffer */ + rows = MIN((long)ptr->rowsperchunk, (long)ptr->rows_in_mem - i); + /* Transfer no more than is currently defined */ + thisrow = (long)ptr->cur_start_row + i; + rows = MIN(rows, (long)ptr->first_undef_row - thisrow); + /* Transfer no more than fits in file */ + rows = MIN(rows, (long)ptr->rows_in_array - thisrow); + if (rows <= 0) /* this chunk might be past end of file! */ + break; + byte_count = rows * bytesperrow; + if (writing) + (*ptr->b_s_info.write_backing_store) (cinfo, &ptr->b_s_info, + (void *)ptr->mem_buffer[i], + file_offset, byte_count); + else + (*ptr->b_s_info.read_backing_store) (cinfo, &ptr->b_s_info, + (void *)ptr->mem_buffer[i], + file_offset, byte_count); + file_offset += byte_count; + } +} + + +METHODDEF(JSAMPARRAY) +access_virt_sarray(j_common_ptr cinfo, jvirt_sarray_ptr ptr, + JDIMENSION start_row, JDIMENSION num_rows, boolean writable) +/* Access the part of a virtual sample array starting at start_row */ +/* and extending for num_rows rows. writable is true if */ +/* caller intends to modify the accessed area. */ +{ + JDIMENSION end_row = start_row + num_rows; + JDIMENSION undef_row; + int data_precision = cinfo->is_decompressor ? + ((j_decompress_ptr)cinfo)->data_precision : + ((j_compress_ptr)cinfo)->data_precision; + size_t sample_size = data_precision == 16 ? + sizeof(J16SAMPLE) : (data_precision == 12 ? + sizeof(J12SAMPLE) : + sizeof(JSAMPLE)); + + /* debugging check */ + if (end_row > ptr->rows_in_array || num_rows > ptr->maxaccess || + ptr->mem_buffer == NULL) + ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS); + + /* Make the desired part of the virtual array accessible */ + if (start_row < ptr->cur_start_row || + end_row > ptr->cur_start_row + ptr->rows_in_mem) { + if (!ptr->b_s_open) + ERREXIT(cinfo, JERR_VIRTUAL_BUG); + /* Flush old buffer contents if necessary */ + if (ptr->dirty) { + do_sarray_io(cinfo, ptr, TRUE); + ptr->dirty = FALSE; + } + /* Decide what part of virtual array to access. + * Algorithm: if target address > current window, assume forward scan, + * load starting at target address. If target address < current window, + * assume backward scan, load so that target area is top of window. + * Note that when switching from forward write to forward read, will have + * start_row = 0, so the limiting case applies and we load from 0 anyway. + */ + if (start_row > ptr->cur_start_row) { + ptr->cur_start_row = start_row; + } else { + /* use long arithmetic here to avoid overflow & unsigned problems */ + long ltemp; + + ltemp = (long)end_row - (long)ptr->rows_in_mem; + if (ltemp < 0) + ltemp = 0; /* don't fall off front end of file */ + ptr->cur_start_row = (JDIMENSION)ltemp; + } + /* Read in the selected part of the array. + * During the initial write pass, we will do no actual read + * because the selected part is all undefined. + */ + do_sarray_io(cinfo, ptr, FALSE); + } + /* Ensure the accessed part of the array is defined; prezero if needed. + * To improve locality of access, we only prezero the part of the array + * that the caller is about to access, not the entire in-memory array. + */ + if (ptr->first_undef_row < end_row) { + if (ptr->first_undef_row < start_row) { + if (writable) /* writer skipped over a section of array */ + ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS); + undef_row = start_row; /* but reader is allowed to read ahead */ + } else { + undef_row = ptr->first_undef_row; + } + if (writable) + ptr->first_undef_row = end_row; + if (ptr->pre_zero) { + size_t bytesperrow = (size_t)ptr->samplesperrow * sample_size; + undef_row -= ptr->cur_start_row; /* make indexes relative to buffer */ + end_row -= ptr->cur_start_row; + while (undef_row < end_row) { + jzero_far((void *)ptr->mem_buffer[undef_row], bytesperrow); + undef_row++; + } + } else { + if (!writable) /* reader looking at undefined data */ + ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS); + } + } + /* Flag the buffer dirty if caller will write in it */ + if (writable) + ptr->dirty = TRUE; + /* Return address of proper part of the buffer */ + return ptr->mem_buffer + (start_row - ptr->cur_start_row); +} + + +METHODDEF(JBLOCKARRAY) +access_virt_barray(j_common_ptr cinfo, jvirt_barray_ptr ptr, + JDIMENSION start_row, JDIMENSION num_rows, boolean writable) +/* Access the part of a virtual block array starting at start_row */ +/* and extending for num_rows rows. writable is true if */ +/* caller intends to modify the accessed area. */ +{ + JDIMENSION end_row = start_row + num_rows; + JDIMENSION undef_row; + + /* debugging check */ + if (end_row > ptr->rows_in_array || num_rows > ptr->maxaccess || + ptr->mem_buffer == NULL) + ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS); + + /* Make the desired part of the virtual array accessible */ + if (start_row < ptr->cur_start_row || + end_row > ptr->cur_start_row + ptr->rows_in_mem) { + if (!ptr->b_s_open) + ERREXIT(cinfo, JERR_VIRTUAL_BUG); + /* Flush old buffer contents if necessary */ + if (ptr->dirty) { + do_barray_io(cinfo, ptr, TRUE); + ptr->dirty = FALSE; + } + /* Decide what part of virtual array to access. + * Algorithm: if target address > current window, assume forward scan, + * load starting at target address. If target address < current window, + * assume backward scan, load so that target area is top of window. + * Note that when switching from forward write to forward read, will have + * start_row = 0, so the limiting case applies and we load from 0 anyway. + */ + if (start_row > ptr->cur_start_row) { + ptr->cur_start_row = start_row; + } else { + /* use long arithmetic here to avoid overflow & unsigned problems */ + long ltemp; + + ltemp = (long)end_row - (long)ptr->rows_in_mem; + if (ltemp < 0) + ltemp = 0; /* don't fall off front end of file */ + ptr->cur_start_row = (JDIMENSION)ltemp; + } + /* Read in the selected part of the array. + * During the initial write pass, we will do no actual read + * because the selected part is all undefined. + */ + do_barray_io(cinfo, ptr, FALSE); + } + /* Ensure the accessed part of the array is defined; prezero if needed. + * To improve locality of access, we only prezero the part of the array + * that the caller is about to access, not the entire in-memory array. + */ + if (ptr->first_undef_row < end_row) { + if (ptr->first_undef_row < start_row) { + if (writable) /* writer skipped over a section of array */ + ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS); + undef_row = start_row; /* but reader is allowed to read ahead */ + } else { + undef_row = ptr->first_undef_row; + } + if (writable) + ptr->first_undef_row = end_row; + if (ptr->pre_zero) { + size_t bytesperrow = (size_t)ptr->blocksperrow * sizeof(JBLOCK); + undef_row -= ptr->cur_start_row; /* make indexes relative to buffer */ + end_row -= ptr->cur_start_row; + while (undef_row < end_row) { + jzero_far((void *)ptr->mem_buffer[undef_row], bytesperrow); + undef_row++; + } + } else { + if (!writable) /* reader looking at undefined data */ + ERREXIT(cinfo, JERR_BAD_VIRTUAL_ACCESS); + } + } + /* Flag the buffer dirty if caller will write in it */ + if (writable) + ptr->dirty = TRUE; + /* Return address of proper part of the buffer */ + return ptr->mem_buffer + (start_row - ptr->cur_start_row); +} + + +/* + * Release all objects belonging to a specified pool. + */ + +METHODDEF(void) +free_pool(j_common_ptr cinfo, int pool_id) +{ + my_mem_ptr mem = (my_mem_ptr)cinfo->mem; + small_pool_ptr shdr_ptr; + large_pool_ptr lhdr_ptr; + size_t space_freed; + + if (pool_id < 0 || pool_id >= JPOOL_NUMPOOLS) + ERREXIT1(cinfo, JERR_BAD_POOL_ID, pool_id); /* safety check */ + +#ifdef MEM_STATS + if (cinfo->err->trace_level > 1) + print_mem_stats(cinfo, pool_id); /* print pool's memory usage statistics */ +#endif + + /* If freeing IMAGE pool, close any virtual arrays first */ + if (pool_id == JPOOL_IMAGE) { + jvirt_sarray_ptr sptr; + jvirt_barray_ptr bptr; + + for (sptr = mem->virt_sarray_list; sptr != NULL; sptr = sptr->next) { + if (sptr->b_s_open) { /* there may be no backing store */ + sptr->b_s_open = FALSE; /* prevent recursive close if error */ + (*sptr->b_s_info.close_backing_store) (cinfo, &sptr->b_s_info); + } + } + mem->virt_sarray_list = NULL; + for (bptr = mem->virt_barray_list; bptr != NULL; bptr = bptr->next) { + if (bptr->b_s_open) { /* there may be no backing store */ + bptr->b_s_open = FALSE; /* prevent recursive close if error */ + (*bptr->b_s_info.close_backing_store) (cinfo, &bptr->b_s_info); + } + } + mem->virt_barray_list = NULL; + } + + /* Release large objects */ + lhdr_ptr = mem->large_list[pool_id]; + mem->large_list[pool_id] = NULL; + + while (lhdr_ptr != NULL) { + large_pool_ptr next_lhdr_ptr = lhdr_ptr->next; + space_freed = lhdr_ptr->bytes_used + + lhdr_ptr->bytes_left + + sizeof(large_pool_hdr) + ALIGN_SIZE - 1; + jpeg_free_large(cinfo, (void *)lhdr_ptr, space_freed); + mem->total_space_allocated -= space_freed; + lhdr_ptr = next_lhdr_ptr; + } + + /* Release small objects */ + shdr_ptr = mem->small_list[pool_id]; + mem->small_list[pool_id] = NULL; + + while (shdr_ptr != NULL) { + small_pool_ptr next_shdr_ptr = shdr_ptr->next; + space_freed = shdr_ptr->bytes_used + shdr_ptr->bytes_left + + sizeof(small_pool_hdr) + ALIGN_SIZE - 1; + jpeg_free_small(cinfo, (void *)shdr_ptr, space_freed); + mem->total_space_allocated -= space_freed; + shdr_ptr = next_shdr_ptr; + } +} + + +/* + * Close up shop entirely. + * Note that this cannot be called unless cinfo->mem is non-NULL. + */ + +METHODDEF(void) +self_destruct(j_common_ptr cinfo) +{ + int pool; + + /* Close all backing store, release all memory. + * Releasing pools in reverse order might help avoid fragmentation + * with some (brain-damaged) malloc libraries. + */ + for (pool = JPOOL_NUMPOOLS - 1; pool >= JPOOL_PERMANENT; pool--) { + free_pool(cinfo, pool); + } + + /* Release the memory manager control block too. */ + jpeg_free_small(cinfo, (void *)cinfo->mem, sizeof(my_memory_mgr)); + cinfo->mem = NULL; /* ensures I will be called only once */ + + jpeg_mem_term(cinfo); /* system-dependent cleanup */ +} + + +/* + * Memory manager initialization. + * When this is called, only the error manager pointer is valid in cinfo! + */ + +GLOBAL(void) +jinit_memory_mgr(j_common_ptr cinfo) +{ + my_mem_ptr mem; + long max_to_use; + int pool; + size_t test_mac; + + cinfo->mem = NULL; /* for safety if init fails */ + + /* Check for configuration errors. + * sizeof(ALIGN_TYPE) should be a power of 2; otherwise, it probably + * doesn't reflect any real hardware alignment requirement. + * The test is a little tricky: for X>0, X and X-1 have no one-bits + * in common if and only if X is a power of 2, ie has only one one-bit. + * Some compilers may give an "unreachable code" warning here; ignore it. + */ + if ((ALIGN_SIZE & (ALIGN_SIZE - 1)) != 0) + ERREXIT(cinfo, JERR_BAD_ALIGN_TYPE); + /* MAX_ALLOC_CHUNK must be representable as type size_t, and must be + * a multiple of ALIGN_SIZE. + * Again, an "unreachable code" warning may be ignored here. + * But a "constant too large" warning means you need to fix MAX_ALLOC_CHUNK. + */ + test_mac = (size_t)MAX_ALLOC_CHUNK; + if ((long)test_mac != MAX_ALLOC_CHUNK || + (MAX_ALLOC_CHUNK % ALIGN_SIZE) != 0) + ERREXIT(cinfo, JERR_BAD_ALLOC_CHUNK); + + max_to_use = jpeg_mem_init(cinfo); /* system-dependent initialization */ + + /* Attempt to allocate memory manager's control block */ + mem = (my_mem_ptr)jpeg_get_small(cinfo, sizeof(my_memory_mgr)); + + if (mem == NULL) { + jpeg_mem_term(cinfo); /* system-dependent cleanup */ + ERREXIT1(cinfo, JERR_OUT_OF_MEMORY, 0); + } + + /* OK, fill in the method pointers */ + mem->pub.alloc_small = alloc_small; + mem->pub.alloc_large = alloc_large; + mem->pub.alloc_sarray = alloc_sarray; + mem->pub.alloc_barray = alloc_barray; + mem->pub.request_virt_sarray = request_virt_sarray; + mem->pub.request_virt_barray = request_virt_barray; + mem->pub.realize_virt_arrays = realize_virt_arrays; + mem->pub.access_virt_sarray = access_virt_sarray; + mem->pub.access_virt_barray = access_virt_barray; + mem->pub.free_pool = free_pool; + mem->pub.self_destruct = self_destruct; + + /* Make MAX_ALLOC_CHUNK accessible to other modules */ + mem->pub.max_alloc_chunk = MAX_ALLOC_CHUNK; + + /* Initialize working state */ + mem->pub.max_memory_to_use = max_to_use; + + for (pool = JPOOL_NUMPOOLS - 1; pool >= JPOOL_PERMANENT; pool--) { + mem->small_list[pool] = NULL; + mem->large_list[pool] = NULL; + } + mem->virt_sarray_list = NULL; + mem->virt_barray_list = NULL; + + mem->total_space_allocated = sizeof(my_memory_mgr); + + /* Declare ourselves open for business */ + cinfo->mem = &mem->pub; + + /* Check for an environment variable JPEGMEM; if found, override the + * default max_memory setting from jpeg_mem_init. Note that the + * surrounding application may again override this value. + * If your system doesn't support getenv(), define NO_GETENV to disable + * this feature. + */ +#ifndef NO_GETENV + { + char memenv[30] = { 0 }; + + if (!GETENV_S(memenv, 30, "JPEGMEM") && strlen(memenv) > 0) { + char ch = 'x'; + +#ifdef _MSC_VER + if (sscanf_s(memenv, "%ld%c", &max_to_use, &ch, 1) > 0) { +#else + if (sscanf(memenv, "%ld%c", &max_to_use, &ch) > 0) { +#endif + if (ch == 'm' || ch == 'M') + max_to_use *= 1000L; + mem->pub.max_memory_to_use = max_to_use * 1000L; + } + } + } +#endif + +} diff --git a/src/spice2x/external/libjpeg-turbo/jmemnobs.c b/src/spice2x/external/libjpeg-turbo/jmemnobs.c new file mode 100644 index 0000000..692775f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jmemnobs.c @@ -0,0 +1,110 @@ +/* + * jmemnobs.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1992-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2017-2018, 2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file provides a really simple implementation of the system- + * dependent portion of the JPEG memory manager. This implementation + * assumes that no backing-store files are needed: all required space + * can be obtained from malloc(). + * This is very portable in the sense that it'll compile on almost anything, + * but you'd better have lots of main memory (or virtual memory) if you want + * to process big images. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jmemsys.h" /* import the system-dependent declarations */ + + +/* + * Memory allocation and freeing are controlled by the regular library + * routines malloc() and free(). + */ + +GLOBAL(void *) +jpeg_get_small(j_common_ptr cinfo, size_t sizeofobject) +{ + return (void *)MALLOC(sizeofobject); +} + +GLOBAL(void) +jpeg_free_small(j_common_ptr cinfo, void *object, size_t sizeofobject) +{ + free(object); +} + + +/* + * "Large" objects are treated the same as "small" ones. + */ + +GLOBAL(void *) +jpeg_get_large(j_common_ptr cinfo, size_t sizeofobject) +{ + return (void *)MALLOC(sizeofobject); +} + +GLOBAL(void) +jpeg_free_large(j_common_ptr cinfo, void *object, size_t sizeofobject) +{ + free(object); +} + + +/* + * This routine computes the total memory space available for allocation. + */ + +GLOBAL(size_t) +jpeg_mem_available(j_common_ptr cinfo, size_t min_bytes_needed, + size_t max_bytes_needed, size_t already_allocated) +{ + if (cinfo->mem->max_memory_to_use) { + if ((size_t)cinfo->mem->max_memory_to_use > already_allocated) + return cinfo->mem->max_memory_to_use - already_allocated; + else + return 0; + } else { + /* Here we always say, "we got all you want bud!" */ + return max_bytes_needed; + } +} + + +/* + * Backing store (temporary file) management. + * Since jpeg_mem_available always promised the moon, + * this should never be called and we can just error out. + */ + +GLOBAL(void) +jpeg_open_backing_store(j_common_ptr cinfo, backing_store_ptr info, + long total_bytes_needed) +{ + ERREXIT(cinfo, JERR_NO_BACKING_STORE); +} + + +/* + * These routines take care of any system-dependent initialization and + * cleanup required. Here, there isn't any. + */ + +GLOBAL(long) +jpeg_mem_init(j_common_ptr cinfo) +{ + return 0; /* just set max_memory_to_use to 0 */ +} + +GLOBAL(void) +jpeg_mem_term(j_common_ptr cinfo) +{ + /* no work */ +} diff --git a/src/spice2x/external/libjpeg-turbo/jmemsys.h b/src/spice2x/external/libjpeg-turbo/jmemsys.h new file mode 100644 index 0000000..ac09ef4 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jmemsys.h @@ -0,0 +1,147 @@ +/* + * jmemsys.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1992-1997, Thomas G. Lane. + * It was modified by The libjpeg-turbo Project to include only code and + * information relevant to libjpeg-turbo. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This include file defines the interface between the system-independent + * and system-dependent portions of the JPEG memory manager. No other + * modules need include it. (The system-independent portion is jmemmgr.c; + * there are several different versions of the system-dependent portion.) + * + * This file works as-is for the system-dependent memory managers supplied + * in the IJG distribution. You may need to modify it if you write a + * custom memory manager. If system-dependent changes are needed in + * this file, the best method is to #ifdef them based on a configuration + * symbol supplied in jconfig.h. + */ + + +/* + * These two functions are used to allocate and release small chunks of + * memory. (Typically the total amount requested through jpeg_get_small is + * no more than 20K or so; this will be requested in chunks of a few K each.) + * Behavior should be the same as for the standard library functions malloc + * and free; in particular, jpeg_get_small must return NULL on failure. + * On most systems, these ARE malloc and free. jpeg_free_small is passed the + * size of the object being freed, just in case it's needed. + */ + +EXTERN(void *) jpeg_get_small(j_common_ptr cinfo, size_t sizeofobject); +EXTERN(void) jpeg_free_small(j_common_ptr cinfo, void *object, + size_t sizeofobject); + +/* + * These two functions are used to allocate and release large chunks of + * memory (up to the total free space designated by jpeg_mem_available). + * These are identical to the jpeg_get/free_small routines; but we keep them + * separate anyway, in case a different allocation strategy is desirable for + * large chunks. + */ + +EXTERN(void *) jpeg_get_large(j_common_ptr cinfo, size_t sizeofobject); +EXTERN(void) jpeg_free_large(j_common_ptr cinfo, void *object, + size_t sizeofobject); + +/* + * The macro MAX_ALLOC_CHUNK designates the maximum number of bytes that may + * be requested in a single call to jpeg_get_large (and jpeg_get_small for that + * matter, but that case should never come into play). This macro was needed + * to model the 64Kb-segment-size limit of far addressing on 80x86 machines. + * On machines with flat address spaces, any large constant may be used. + * + * NB: jmemmgr.c expects that MAX_ALLOC_CHUNK will be representable as type + * size_t and will be a multiple of sizeof(align_type). + */ + +#ifndef MAX_ALLOC_CHUNK /* may be overridden in jconfig.h */ +#define MAX_ALLOC_CHUNK 1000000000L +#endif + +/* + * This routine computes the total space still available for allocation by + * jpeg_get_large. If more space than this is needed, backing store will be + * used. NOTE: any memory already allocated must not be counted. + * + * There is a minimum space requirement, corresponding to the minimum + * feasible buffer sizes; jmemmgr.c will request that much space even if + * jpeg_mem_available returns zero. The maximum space needed, enough to hold + * all working storage in memory, is also passed in case it is useful. + * Finally, the total space already allocated is passed. If no better + * method is available, cinfo->mem->max_memory_to_use - already_allocated + * is often a suitable calculation. + * + * It is OK for jpeg_mem_available to underestimate the space available + * (that'll just lead to more backing-store access than is really necessary). + * However, an overestimate will lead to failure. Hence it's wise to subtract + * a slop factor from the true available space. 5% should be enough. + * + * On machines with lots of virtual memory, any large constant may be returned. + * Conversely, zero may be returned to always use the minimum amount of memory. + */ + +EXTERN(size_t) jpeg_mem_available(j_common_ptr cinfo, size_t min_bytes_needed, + size_t max_bytes_needed, + size_t already_allocated); + + +/* + * This structure holds whatever state is needed to access a single + * backing-store object. The read/write/close method pointers are called + * by jmemmgr.c to manipulate the backing-store object; all other fields + * are private to the system-dependent backing store routines. + */ + +#define TEMP_NAME_LENGTH 64 /* max length of a temporary file's name */ + + +typedef struct backing_store_struct *backing_store_ptr; + +typedef struct backing_store_struct { + /* Methods for reading/writing/closing this backing-store object */ + void (*read_backing_store) (j_common_ptr cinfo, backing_store_ptr info, + void *buffer_address, long file_offset, + long byte_count); + void (*write_backing_store) (j_common_ptr cinfo, backing_store_ptr info, + void *buffer_address, long file_offset, + long byte_count); + void (*close_backing_store) (j_common_ptr cinfo, backing_store_ptr info); + + /* Private fields for system-dependent backing-store management */ + /* For a typical implementation with temp files, we need: */ + FILE *temp_file; /* stdio reference to temp file */ + char temp_name[TEMP_NAME_LENGTH]; /* name of temp file */ +} backing_store_info; + + +/* + * Initial opening of a backing-store object. This must fill in the + * read/write/close pointers in the object. The read/write routines + * may take an error exit if the specified maximum file size is exceeded. + * (If jpeg_mem_available always returns a large value, this routine can + * just take an error exit.) + */ + +EXTERN(void) jpeg_open_backing_store(j_common_ptr cinfo, + backing_store_ptr info, + long total_bytes_needed); + + +/* + * These routines take care of any system-dependent initialization and + * cleanup required. jpeg_mem_init will be called before anything is + * allocated (and, therefore, nothing in cinfo is of use except the error + * manager pointer). It should return a suitable default value for + * max_memory_to_use; this may subsequently be overridden by the surrounding + * application. (Note that max_memory_to_use is only important if + * jpeg_mem_available chooses to consult it ... no one else will.) + * jpeg_mem_term may assume that all requested memory has been freed and that + * all opened backing-store objects have been closed. + */ + +EXTERN(long) jpeg_mem_init(j_common_ptr cinfo); +EXTERN(void) jpeg_mem_term(j_common_ptr cinfo); diff --git a/src/spice2x/external/libjpeg-turbo/jmorecfg.h b/src/spice2x/external/libjpeg-turbo/jmorecfg.h new file mode 100644 index 0000000..89c7842 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jmorecfg.h @@ -0,0 +1,385 @@ +/* + * jmorecfg.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Modified 1997-2009 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2009, 2011, 2014-2015, 2018, 2020, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains additional configuration options that customize the + * JPEG software for special applications or support machine-dependent + * optimizations. Most users will not need to touch this file. + */ + + +/* + * Maximum number of components (color channels) allowed in JPEG image. + * To meet the letter of Rec. ITU-T T.81 | ISO/IEC 10918-1, set this to 255. + * However, darn few applications need more than 4 channels (maybe 5 for CMYK + + * alpha mask). We recommend 10 as a reasonable compromise; use 4 if you are + * really short on memory. (Each allowed component costs a hundred or so + * bytes of storage, whether actually used in an image or not.) + */ + +#define MAX_COMPONENTS 10 /* maximum number of image components */ + + +/* + * Basic data types. + * You may need to change these if you have a machine with unusual data + * type sizes; for example, "char" not 8 bits, "short" not 16 bits, + * or "long" not 32 bits. We don't care whether "int" is 16 or 32 bits, + * but it had better be at least 16. + */ + +/* Representation of a single sample (pixel element value). + * We frequently allocate large arrays of these, so it's important to keep + * them small. But if you have memory to burn and access to char or short + * arrays is very slow on your hardware, you might want to change these. + */ + +/* JSAMPLE should be the smallest type that will hold the values 0..255. */ + +typedef unsigned char JSAMPLE; +#define GETJSAMPLE(value) ((int)(value)) + +#define MAXJSAMPLE 255 +#define CENTERJSAMPLE 128 + + +/* J12SAMPLE should be the smallest type that will hold the values 0..4095. */ + +typedef short J12SAMPLE; + +#define MAXJ12SAMPLE 4095 +#define CENTERJ12SAMPLE 2048 + + +/* J16SAMPLE should be the smallest type that will hold the values 0..65535. */ + +typedef unsigned short J16SAMPLE; + +#define MAXJ16SAMPLE 65535 +#define CENTERJ16SAMPLE 32768 + + +/* Representation of a DCT frequency coefficient. + * This should be a signed value of at least 16 bits; "short" is usually OK. + * Again, we allocate large arrays of these, but you can change to int + * if you have memory to burn and "short" is really slow. + */ + +typedef short JCOEF; + + +/* Compressed datastreams are represented as arrays of JOCTET. + * These must be EXACTLY 8 bits wide, at least once they are written to + * external storage. Note that when using the stdio data source/destination + * managers, this is also the data type passed to fread/fwrite. + */ + +typedef unsigned char JOCTET; +#define GETJOCTET(value) (value) + + +/* These typedefs are used for various table entries and so forth. + * They must be at least as wide as specified; but making them too big + * won't cost a huge amount of memory, so we don't provide special + * extraction code like we did for JSAMPLE. (In other words, these + * typedefs live at a different point on the speed/space tradeoff curve.) + */ + +/* UINT8 must hold at least the values 0..255. */ + +typedef unsigned char UINT8; + +/* UINT16 must hold at least the values 0..65535. */ + +typedef unsigned short UINT16; + +/* INT16 must hold at least the values -32768..32767. */ + +#ifndef XMD_H /* X11/xmd.h correctly defines INT16 */ +typedef short INT16; +#endif + +/* INT32 must hold at least signed 32-bit values. + * + * NOTE: The INT32 typedef dates back to libjpeg v5 (1994.) Integers were + * sometimes 16-bit back then (MS-DOS), which is why INT32 is typedef'd to + * long. It also wasn't common (or at least as common) in 1994 for INT32 to be + * defined by platform headers. Since then, however, INT32 is defined in + * several other common places: + * + * Xmd.h (X11 header) typedefs INT32 to int on 64-bit platforms and long on + * 32-bit platforms (i.e always a 32-bit signed type.) + * + * basetsd.h (Win32 header) typedefs INT32 to int (always a 32-bit signed type + * on modern platforms.) + * + * qglobal.h (Qt header) typedefs INT32 to int (always a 32-bit signed type on + * modern platforms.) + * + * This is a recipe for conflict, since "long" and "int" aren't always + * compatible types. Since the definition of INT32 has technically been part + * of the libjpeg API for more than 20 years, we can't remove it, but we do not + * use it internally any longer. We instead define a separate type (JLONG) + * for internal use, which ensures that internal behavior will always be the + * same regardless of any external headers that may be included. + */ + +#ifndef XMD_H /* X11/xmd.h correctly defines INT32 */ +#ifndef _BASETSD_H_ /* Microsoft defines it in basetsd.h */ +#ifndef _BASETSD_H /* MinGW is slightly different */ +#ifndef QGLOBAL_H /* Qt defines it in qglobal.h */ +typedef long INT32; +#endif +#endif +#endif +#endif + +/* Datatype used for image dimensions. The JPEG standard only supports + * images up to 64K*64K due to 16-bit fields in SOF markers. Therefore + * "unsigned int" is sufficient on all machines. However, if you need to + * handle larger images and you don't mind deviating from the spec, you + * can change this datatype. (Note that changing this datatype will + * potentially require modifying the SIMD code. The x86-64 SIMD extensions, + * in particular, assume a 32-bit JDIMENSION.) + */ + +typedef unsigned int JDIMENSION; + +#define JPEG_MAX_DIMENSION 65500L /* a tad under 64K to prevent overflows */ + + +/* These macros are used in all function definitions and extern declarations. + * You could modify them if you need to change function linkage conventions; + * in particular, you'll need to do that to make the library a Windows DLL. + * Another application is to make all functions global for use with debuggers + * or code profilers that require it. + */ + +/* a function called through method pointers: */ +#define METHODDEF(type) static type +/* a function used only in its module: */ +#define LOCAL(type) static type +/* a function referenced thru EXTERNs: */ +#define GLOBAL(type) type +/* a reference to a GLOBAL function: */ +#define EXTERN(type) extern type + + +/* Originally, this macro was used as a way of defining function prototypes + * for both modern compilers as well as older compilers that did not support + * prototype parameters. libjpeg-turbo has never supported these older, + * non-ANSI compilers, but the macro is still included because there is some + * software out there that uses it. + */ + +#define JMETHOD(type, methodname, arglist) type (*methodname) arglist + + +/* libjpeg-turbo no longer supports platforms that have far symbols (MS-DOS), + * but again, some software relies on this macro. + */ + +#undef FAR +#define FAR + + +/* + * On a few systems, type boolean and/or its values FALSE, TRUE may appear + * in standard header files. Or you may have conflicts with application- + * specific header files that you want to include together with these files. + * Defining HAVE_BOOLEAN before including jpeglib.h should make it work. + */ + +#ifndef HAVE_BOOLEAN +typedef int boolean; +#endif +#ifndef FALSE /* in case these macros already exist */ +#define FALSE 0 /* values of boolean */ +#endif +#ifndef TRUE +#define TRUE 1 +#endif + + +/* + * The remaining options affect code selection within the JPEG library, + * but they don't need to be visible to most applications using the library. + * To minimize application namespace pollution, the symbols won't be + * defined unless JPEG_INTERNALS or JPEG_INTERNAL_OPTIONS has been defined. + */ + +#ifdef JPEG_INTERNALS +#define JPEG_INTERNAL_OPTIONS +#endif + +#ifdef JPEG_INTERNAL_OPTIONS + + +/* + * These defines indicate whether to include various optional functions. + * Undefining some of these symbols will produce a smaller but less capable + * library. Note that you can leave certain source files out of the + * compilation/linking process if you've #undef'd the corresponding symbols. + * (You may HAVE to do that if your compiler doesn't like null source files.) + */ + +/* Capability options common to encoder and decoder: */ + +#define DCT_ISLOW_SUPPORTED /* accurate integer method */ +#define DCT_IFAST_SUPPORTED /* less accurate int method [legacy feature] */ +#define DCT_FLOAT_SUPPORTED /* floating-point method [legacy feature] */ + +/* Encoder capability options: */ + +#define C_MULTISCAN_FILES_SUPPORTED /* Multiple-scan JPEG files? */ +#define C_PROGRESSIVE_SUPPORTED /* Progressive JPEG? (Requires MULTISCAN)*/ +#define C_LOSSLESS_SUPPORTED /* Lossless JPEG? */ +#define ENTROPY_OPT_SUPPORTED /* Optimization of entropy coding parms? */ +/* Note: if you selected 12-bit data precision, it is dangerous to turn off + * ENTROPY_OPT_SUPPORTED. The standard Huffman tables are only good for 8-bit + * precision, so jchuff.c normally uses entropy optimization to compute + * usable tables for higher precision. If you don't want to do optimization, + * you'll have to supply different default Huffman tables. + * The exact same statements apply for progressive and lossless JPEG: + * the default tables don't work for progressive mode or lossless mode. + * (This may get fixed, however.) + */ +#define INPUT_SMOOTHING_SUPPORTED /* Input image smoothing option? */ + +/* Decoder capability options: */ + +#define D_MULTISCAN_FILES_SUPPORTED /* Multiple-scan JPEG files? */ +#define D_PROGRESSIVE_SUPPORTED /* Progressive JPEG? (Requires MULTISCAN)*/ +#define D_LOSSLESS_SUPPORTED /* Lossless JPEG? */ +#define SAVE_MARKERS_SUPPORTED /* jpeg_save_markers() needed? */ +#define BLOCK_SMOOTHING_SUPPORTED /* Block smoothing? (Progressive only) */ +#define IDCT_SCALING_SUPPORTED /* Output rescaling via IDCT? */ +#undef UPSAMPLE_SCALING_SUPPORTED /* Output rescaling at upsample stage? */ +#define UPSAMPLE_MERGING_SUPPORTED /* Fast path for sloppy upsampling? */ +#define QUANT_1PASS_SUPPORTED /* 1-pass color quantization? */ +#define QUANT_2PASS_SUPPORTED /* 2-pass color quantization? */ + +/* more capability options later, no doubt */ + + +/* + * The RGB_RED, RGB_GREEN, RGB_BLUE, and RGB_PIXELSIZE macros are a vestigial + * feature of libjpeg. The idea was that, if an application developer needed + * to compress from/decompress to a BGR/BGRX/RGBX/XBGR/XRGB buffer, they could + * change these macros, rebuild libjpeg, and link their application statically + * with it. In reality, few people ever did this, because there were some + * severe restrictions involved (cjpeg and djpeg no longer worked properly, + * compressing/decompressing RGB JPEGs no longer worked properly, and the color + * quantizer wouldn't work with pixel sizes other than 3.) Furthermore, since + * all of the O/S-supplied versions of libjpeg were built with the default + * values of RGB_RED, RGB_GREEN, RGB_BLUE, and RGB_PIXELSIZE, many applications + * have come to regard these values as immutable. + * + * The libjpeg-turbo colorspace extensions provide a much cleaner way of + * compressing from/decompressing to buffers with arbitrary component orders + * and pixel sizes. Thus, we do not support changing the values of RGB_RED, + * RGB_GREEN, RGB_BLUE, or RGB_PIXELSIZE. In addition to the restrictions + * listed above, changing these values will also break the SIMD extensions and + * the regression tests. + */ + +#define RGB_RED 0 /* Offset of Red in an RGB scanline element */ +#define RGB_GREEN 1 /* Offset of Green */ +#define RGB_BLUE 2 /* Offset of Blue */ +#define RGB_PIXELSIZE 3 /* JSAMPLEs per RGB scanline element */ + +#define JPEG_NUMCS 17 + +#define EXT_RGB_RED 0 +#define EXT_RGB_GREEN 1 +#define EXT_RGB_BLUE 2 +#define EXT_RGB_PIXELSIZE 3 + +#define EXT_RGBX_RED 0 +#define EXT_RGBX_GREEN 1 +#define EXT_RGBX_BLUE 2 +#define EXT_RGBX_PIXELSIZE 4 + +#define EXT_BGR_RED 2 +#define EXT_BGR_GREEN 1 +#define EXT_BGR_BLUE 0 +#define EXT_BGR_PIXELSIZE 3 + +#define EXT_BGRX_RED 2 +#define EXT_BGRX_GREEN 1 +#define EXT_BGRX_BLUE 0 +#define EXT_BGRX_PIXELSIZE 4 + +#define EXT_XBGR_RED 3 +#define EXT_XBGR_GREEN 2 +#define EXT_XBGR_BLUE 1 +#define EXT_XBGR_PIXELSIZE 4 + +#define EXT_XRGB_RED 1 +#define EXT_XRGB_GREEN 2 +#define EXT_XRGB_BLUE 3 +#define EXT_XRGB_PIXELSIZE 4 + +static const int rgb_red[JPEG_NUMCS] = { + -1, -1, RGB_RED, -1, -1, -1, EXT_RGB_RED, EXT_RGBX_RED, + EXT_BGR_RED, EXT_BGRX_RED, EXT_XBGR_RED, EXT_XRGB_RED, + EXT_RGBX_RED, EXT_BGRX_RED, EXT_XBGR_RED, EXT_XRGB_RED, + -1 +}; + +static const int rgb_green[JPEG_NUMCS] = { + -1, -1, RGB_GREEN, -1, -1, -1, EXT_RGB_GREEN, EXT_RGBX_GREEN, + EXT_BGR_GREEN, EXT_BGRX_GREEN, EXT_XBGR_GREEN, EXT_XRGB_GREEN, + EXT_RGBX_GREEN, EXT_BGRX_GREEN, EXT_XBGR_GREEN, EXT_XRGB_GREEN, + -1 +}; + +static const int rgb_blue[JPEG_NUMCS] = { + -1, -1, RGB_BLUE, -1, -1, -1, EXT_RGB_BLUE, EXT_RGBX_BLUE, + EXT_BGR_BLUE, EXT_BGRX_BLUE, EXT_XBGR_BLUE, EXT_XRGB_BLUE, + EXT_RGBX_BLUE, EXT_BGRX_BLUE, EXT_XBGR_BLUE, EXT_XRGB_BLUE, + -1 +}; + +static const int rgb_pixelsize[JPEG_NUMCS] = { + -1, -1, RGB_PIXELSIZE, -1, -1, -1, EXT_RGB_PIXELSIZE, EXT_RGBX_PIXELSIZE, + EXT_BGR_PIXELSIZE, EXT_BGRX_PIXELSIZE, EXT_XBGR_PIXELSIZE, EXT_XRGB_PIXELSIZE, + EXT_RGBX_PIXELSIZE, EXT_BGRX_PIXELSIZE, EXT_XBGR_PIXELSIZE, EXT_XRGB_PIXELSIZE, + -1 +}; + +/* Definitions for speed-related optimizations. */ + +/* On some machines (notably 68000 series) "int" is 32 bits, but multiplying + * two 16-bit shorts is faster than multiplying two ints. Define MULTIPLIER + * as short on such a machine. MULTIPLIER must be at least 16 bits wide. + */ + +#ifndef MULTIPLIER +#ifndef WITH_SIMD +#define MULTIPLIER int /* type for fastest integer multiply */ +#else +#define MULTIPLIER short /* prefer 16-bit with SIMD for parellelism */ +#endif +#endif + + +/* FAST_FLOAT should be either float or double, whichever is done faster + * by your compiler. (Note that this type is only used in the floating point + * DCT routines, so it only matters if you've defined DCT_FLOAT_SUPPORTED.) + */ + +#ifndef FAST_FLOAT +#define FAST_FLOAT float +#endif + +#endif /* JPEG_INTERNAL_OPTIONS */ diff --git a/src/spice2x/external/libjpeg-turbo/jpeg_nbits.c b/src/spice2x/external/libjpeg-turbo/jpeg_nbits.c new file mode 100644 index 0000000..c8ee6b0 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jpeg_nbits.c @@ -0,0 +1,4134 @@ +/* + * Copyright (C) 2024, D. R. Commander. + * + * For conditions of distribution and use, see the accompanying README.ijg + * file. + */ + +#include "jpeg_nbits.h" +#include "jconfigint.h" + + +#ifndef USE_CLZ_INTRINSIC + +#define INCLUDE_JPEG_NBITS_TABLE + +/* When building for x86[-64] with the SIMD extensions enabled, the C Huffman + * encoders can reuse jpeg_nbits_table from the SSE2 baseline Huffman encoder. + */ +#if (defined(__x86_64__) || defined(__i386__) || defined(_M_IX86) || \ + defined(_M_X64)) && defined(WITH_SIMD) +#undef INCLUDE_JPEG_NBITS_TABLE +#endif + +#endif + + +#ifdef INCLUDE_JPEG_NBITS_TABLE + +const unsigned char HIDDEN jpeg_nbits_table[65536] = { + 0, 1, 2, 2, 3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, + 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, 5, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, + 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, 7, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, 8, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, 9, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, 10, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, 11, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, 13, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, + 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16, 16 +}; + +#else + +/* Suppress compiler warnings about empty translation unit. */ + +typedef int dummy_jpeg_nbits_table; + +#endif diff --git a/src/spice2x/external/libjpeg-turbo/jpeg_nbits.h b/src/spice2x/external/libjpeg-turbo/jpeg_nbits.h new file mode 100644 index 0000000..6481a12 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jpeg_nbits.h @@ -0,0 +1,43 @@ +/* + * Copyright (C) 2014, 2021, 2024, D. R. Commander. + * Copyright (C) 2014, Olle Liljenzin. + * Copyright (C) 2020, Arm Limited. + * + * For conditions of distribution and use, see the accompanying README.ijg + * file. + */ + +/* + * NOTE: If USE_CLZ_INTRINSIC is defined, then clz/bsr instructions will be + * used for bit counting rather than the lookup table. This will reduce the + * memory footprint by 64k, which is important for some mobile applications + * that create many isolated instances of libjpeg-turbo (web browsers, for + * instance.) This may improve performance on some mobile platforms as well. + * This feature is enabled by default only on Arm processors, because some x86 + * chips have a slow implementation of bsr, and the use of clz/bsr cannot be + * shown to have a significant performance impact even on the x86 chips that + * have a fast implementation of it. When building for Armv6, you can + * explicitly disable the use of clz/bsr by adding -mthumb to the compiler + * flags (this defines __thumb__). + */ + +/* NOTE: Both GCC and Clang define __GNUC__ */ +#if (defined(__GNUC__) && (defined(__arm__) || defined(__aarch64__))) || \ + defined(_M_ARM) || defined(_M_ARM64) +#if !defined(__thumb__) || defined(__thumb2__) +#define USE_CLZ_INTRINSIC +#endif +#endif + +#ifdef USE_CLZ_INTRINSIC +#if defined(_MSC_VER) && !defined(__clang__) +#define JPEG_NBITS_NONZERO(x) (32 - _CountLeadingZeros(x)) +#else +#define JPEG_NBITS_NONZERO(x) (32 - __builtin_clz(x)) +#endif +#define JPEG_NBITS(x) (x ? JPEG_NBITS_NONZERO(x) : 0) +#else +extern const unsigned char jpeg_nbits_table[65536]; +#define JPEG_NBITS(x) (jpeg_nbits_table[x]) +#define JPEG_NBITS_NONZERO(x) JPEG_NBITS(x) +#endif diff --git a/src/spice2x/external/libjpeg-turbo/jpegapicomp.h b/src/spice2x/external/libjpeg-turbo/jpegapicomp.h new file mode 100644 index 0000000..bb3912e --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jpegapicomp.h @@ -0,0 +1,32 @@ +/* + * jpegapicomp.h + * + * Copyright (C) 2010, 2020, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * JPEG compatibility macros + * These declarations are considered internal to the JPEG library; most + * applications using the library shouldn't need to include this file. + */ + +#if JPEG_LIB_VERSION >= 70 +#define _DCT_scaled_size DCT_h_scaled_size +#define _DCT_h_scaled_size DCT_h_scaled_size +#define _DCT_v_scaled_size DCT_v_scaled_size +#define _min_DCT_scaled_size min_DCT_h_scaled_size +#define _min_DCT_h_scaled_size min_DCT_h_scaled_size +#define _min_DCT_v_scaled_size min_DCT_v_scaled_size +#define _jpeg_width jpeg_width +#define _jpeg_height jpeg_height +#define JERR_ARITH_NOTIMPL JERR_NOT_COMPILED +#else +#define _DCT_scaled_size DCT_scaled_size +#define _DCT_h_scaled_size DCT_scaled_size +#define _DCT_v_scaled_size DCT_scaled_size +#define _min_DCT_scaled_size min_DCT_scaled_size +#define _min_DCT_h_scaled_size min_DCT_scaled_size +#define _min_DCT_v_scaled_size min_DCT_scaled_size +#define _jpeg_width image_width +#define _jpeg_height image_height +#endif diff --git a/src/spice2x/external/libjpeg-turbo/jpegint.h b/src/spice2x/external/libjpeg-turbo/jpegint.h new file mode 100644 index 0000000..a90493f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jpegint.h @@ -0,0 +1,600 @@ +/* + * jpegint.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1997, Thomas G. Lane. + * Modified 1997-2009 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2015-2017, 2019, 2021-2022, 2024, D. R. Commander. + * Copyright (C) 2015, Google, Inc. + * Copyright (C) 2021, Alex Richardson. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file provides common declarations for the various JPEG modules. + * These declarations are considered internal to the JPEG library; most + * applications using the library shouldn't need to include this file. + */ + + +/* Representation of a spatial difference value. + * This should be a signed value of at least 16 bits; int is usually OK. + */ + +typedef int JDIFF; + +typedef JDIFF FAR *JDIFFROW; /* pointer to one row of difference values */ +typedef JDIFFROW *JDIFFARRAY; /* ptr to some rows (a 2-D diff array) */ +typedef JDIFFARRAY *JDIFFIMAGE; /* a 3-D diff array: top index is color */ + + +/* Declarations for both compression & decompression */ + +typedef enum { /* Operating modes for buffer controllers */ + JBUF_PASS_THRU, /* Plain stripwise operation */ + /* Remaining modes require a full-image buffer to have been created */ + JBUF_SAVE_SOURCE, /* Run source subobject only, save output */ + JBUF_CRANK_DEST, /* Run dest subobject only, using saved data */ + JBUF_SAVE_AND_PASS /* Run both subobjects, save output */ +} J_BUF_MODE; + +/* Values of global_state field (jdapi.c has some dependencies on ordering!) */ +#define CSTATE_START 100 /* after create_compress */ +#define CSTATE_SCANNING 101 /* start_compress done, write_scanlines OK */ +#define CSTATE_RAW_OK 102 /* start_compress done, write_raw_data OK */ +#define CSTATE_WRCOEFS 103 /* jpeg_write_coefficients done */ +#define DSTATE_START 200 /* after create_decompress */ +#define DSTATE_INHEADER 201 /* reading header markers, no SOS yet */ +#define DSTATE_READY 202 /* found SOS, ready for start_decompress */ +#define DSTATE_PRELOAD 203 /* reading multiscan file in start_decompress*/ +#define DSTATE_PRESCAN 204 /* performing dummy pass for 2-pass quant */ +#define DSTATE_SCANNING 205 /* start_decompress done, read_scanlines OK */ +#define DSTATE_RAW_OK 206 /* start_decompress done, read_raw_data OK */ +#define DSTATE_BUFIMAGE 207 /* expecting jpeg_start_output */ +#define DSTATE_BUFPOST 208 /* looking for SOS/EOI in jpeg_finish_output */ +#define DSTATE_RDCOEFS 209 /* reading file in jpeg_read_coefficients */ +#define DSTATE_STOPPING 210 /* looking for EOI in jpeg_finish_decompress */ + + +/* JLONG must hold at least signed 32-bit values. */ +typedef long JLONG; + +/* JUINTPTR must hold pointer values. */ +#ifdef __UINTPTR_TYPE__ +/* + * __UINTPTR_TYPE__ is GNU-specific and available in GCC 4.6+ and Clang 3.0+. + * Fortunately, that is sufficient to support the few architectures for which + * sizeof(void *) != sizeof(size_t). The only other options would require C99 + * or Clang-specific builtins. + */ +typedef __UINTPTR_TYPE__ JUINTPTR; +#else +typedef size_t JUINTPTR; +#endif + +#define IsExtRGB(cs) \ + (cs == JCS_RGB || (cs >= JCS_EXT_RGB && cs <= JCS_EXT_ARGB)) + +/* + * Left shift macro that handles a negative operand without causing any + * sanitizer warnings + */ + +#define LEFT_SHIFT(a, b) ((JLONG)((unsigned long)(a) << (b))) + + +/* Declarations for compression modules */ + +/* Master control module */ +struct jpeg_comp_master { + void (*prepare_for_pass) (j_compress_ptr cinfo); + void (*pass_startup) (j_compress_ptr cinfo); + void (*finish_pass) (j_compress_ptr cinfo); + + /* State variables made visible to other modules */ + boolean call_pass_startup; /* True if pass_startup must be called */ + boolean is_last_pass; /* True during last pass */ + boolean lossless; /* True if lossless mode is enabled */ +}; + +/* Main buffer control (downsampled-data buffer) */ +struct jpeg_c_main_controller { + void (*start_pass) (j_compress_ptr cinfo, J_BUF_MODE pass_mode); + void (*process_data) (j_compress_ptr cinfo, JSAMPARRAY input_buf, + JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail); + void (*process_data_12) (j_compress_ptr cinfo, J12SAMPARRAY input_buf, + JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail); +#ifdef C_LOSSLESS_SUPPORTED + void (*process_data_16) (j_compress_ptr cinfo, J16SAMPARRAY input_buf, + JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail); +#endif +}; + +/* Compression preprocessing (downsampling input buffer control) */ +struct jpeg_c_prep_controller { + void (*start_pass) (j_compress_ptr cinfo, J_BUF_MODE pass_mode); + void (*pre_process_data) (j_compress_ptr cinfo, JSAMPARRAY input_buf, + JDIMENSION *in_row_ctr, JDIMENSION in_rows_avail, + JSAMPIMAGE output_buf, + JDIMENSION *out_row_group_ctr, + JDIMENSION out_row_groups_avail); + void (*pre_process_data_12) (j_compress_ptr cinfo, J12SAMPARRAY input_buf, + JDIMENSION *in_row_ctr, + JDIMENSION in_rows_avail, + J12SAMPIMAGE output_buf, + JDIMENSION *out_row_group_ctr, + JDIMENSION out_row_groups_avail); +#ifdef C_LOSSLESS_SUPPORTED + void (*pre_process_data_16) (j_compress_ptr cinfo, J16SAMPARRAY input_buf, + JDIMENSION *in_row_ctr, + JDIMENSION in_rows_avail, + J16SAMPIMAGE output_buf, + JDIMENSION *out_row_group_ctr, + JDIMENSION out_row_groups_avail); +#endif +}; + +/* Lossy mode: Coefficient buffer control + * Lossless mode: Difference buffer control + */ +struct jpeg_c_coef_controller { + void (*start_pass) (j_compress_ptr cinfo, J_BUF_MODE pass_mode); + boolean (*compress_data) (j_compress_ptr cinfo, JSAMPIMAGE input_buf); + boolean (*compress_data_12) (j_compress_ptr cinfo, J12SAMPIMAGE input_buf); +#ifdef C_LOSSLESS_SUPPORTED + boolean (*compress_data_16) (j_compress_ptr cinfo, J16SAMPIMAGE input_buf); +#endif +}; + +/* Colorspace conversion */ +struct jpeg_color_converter { + void (*start_pass) (j_compress_ptr cinfo); + void (*color_convert) (j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows); + void (*color_convert_12) (j_compress_ptr cinfo, J12SAMPARRAY input_buf, + J12SAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows); +#ifdef C_LOSSLESS_SUPPORTED + void (*color_convert_16) (j_compress_ptr cinfo, J16SAMPARRAY input_buf, + J16SAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows); +#endif +}; + +/* Downsampling */ +struct jpeg_downsampler { + void (*start_pass) (j_compress_ptr cinfo); + void (*downsample) (j_compress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION in_row_index, JSAMPIMAGE output_buf, + JDIMENSION out_row_group_index); + void (*downsample_12) (j_compress_ptr cinfo, J12SAMPIMAGE input_buf, + JDIMENSION in_row_index, J12SAMPIMAGE output_buf, + JDIMENSION out_row_group_index); +#ifdef C_LOSSLESS_SUPPORTED + void (*downsample_16) (j_compress_ptr cinfo, J16SAMPIMAGE input_buf, + JDIMENSION in_row_index, J16SAMPIMAGE output_buf, + JDIMENSION out_row_group_index); +#endif + + boolean need_context_rows; /* TRUE if need rows above & below */ +}; + +/* Lossy mode: Forward DCT (also controls coefficient quantization) + * Lossless mode: Prediction, sample differencing, and point transform + */ +struct jpeg_forward_dct { + void (*start_pass) (j_compress_ptr cinfo); + + /* Lossy mode */ + /* perhaps this should be an array??? */ + void (*forward_DCT) (j_compress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY sample_data, JBLOCKROW coef_blocks, + JDIMENSION start_row, JDIMENSION start_col, + JDIMENSION num_blocks); + void (*forward_DCT_12) (j_compress_ptr cinfo, jpeg_component_info *compptr, + J12SAMPARRAY sample_data, JBLOCKROW coef_blocks, + JDIMENSION start_row, JDIMENSION start_col, + JDIMENSION num_blocks); +}; + +/* Entropy encoding */ +struct jpeg_entropy_encoder { + void (*start_pass) (j_compress_ptr cinfo, boolean gather_statistics); + + /* Lossy mode */ + boolean (*encode_mcu) (j_compress_ptr cinfo, JBLOCKROW *MCU_data); + /* Lossless mode */ + JDIMENSION (*encode_mcus) (j_compress_ptr cinfo, JDIFFIMAGE diff_buf, + JDIMENSION MCU_row_num, JDIMENSION MCU_col_num, + JDIMENSION nMCU); + + void (*finish_pass) (j_compress_ptr cinfo); +}; + +/* Marker writing */ +struct jpeg_marker_writer { + void (*write_file_header) (j_compress_ptr cinfo); + void (*write_frame_header) (j_compress_ptr cinfo); + void (*write_scan_header) (j_compress_ptr cinfo); + void (*write_file_trailer) (j_compress_ptr cinfo); + void (*write_tables_only) (j_compress_ptr cinfo); + /* These routines are exported to allow insertion of extra markers */ + /* Probably only COM and APPn markers should be written this way */ + void (*write_marker_header) (j_compress_ptr cinfo, int marker, + unsigned int datalen); + void (*write_marker_byte) (j_compress_ptr cinfo, int val); +}; + + +/* Declarations for decompression modules */ + +/* Master control module */ +struct jpeg_decomp_master { + void (*prepare_for_output_pass) (j_decompress_ptr cinfo); + void (*finish_output_pass) (j_decompress_ptr cinfo); + + /* State variables made visible to other modules */ + boolean is_dummy_pass; /* True during 1st pass for 2-pass quant */ + boolean lossless; /* True if decompressing a lossless image */ + + /* Partial decompression variables */ + JDIMENSION first_iMCU_col; + JDIMENSION last_iMCU_col; + JDIMENSION first_MCU_col[MAX_COMPONENTS]; + JDIMENSION last_MCU_col[MAX_COMPONENTS]; + boolean jinit_upsampler_no_alloc; + + /* Last iMCU row that was successfully decoded */ + JDIMENSION last_good_iMCU_row; + + /* Tail of list of saved markers */ + jpeg_saved_marker_ptr marker_list_end; +}; + +/* Input control module */ +struct jpeg_input_controller { + int (*consume_input) (j_decompress_ptr cinfo); + void (*reset_input_controller) (j_decompress_ptr cinfo); + void (*start_input_pass) (j_decompress_ptr cinfo); + void (*finish_input_pass) (j_decompress_ptr cinfo); + + /* State variables made visible to other modules */ + boolean has_multiple_scans; /* True if file has multiple scans */ + boolean eoi_reached; /* True when EOI has been consumed */ +}; + +/* Main buffer control (downsampled-data buffer) */ +struct jpeg_d_main_controller { + void (*start_pass) (j_decompress_ptr cinfo, J_BUF_MODE pass_mode); + void (*process_data) (j_decompress_ptr cinfo, JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); + void (*process_data_12) (j_decompress_ptr cinfo, J12SAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); +#ifdef D_LOSSLESS_SUPPORTED + void (*process_data_16) (j_decompress_ptr cinfo, J16SAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); +#endif +}; + +/* Lossy mode: Coefficient buffer control + * Lossless mode: Difference buffer control + */ +struct jpeg_d_coef_controller { + void (*start_input_pass) (j_decompress_ptr cinfo); + int (*consume_data) (j_decompress_ptr cinfo); + void (*start_output_pass) (j_decompress_ptr cinfo); + int (*decompress_data) (j_decompress_ptr cinfo, JSAMPIMAGE output_buf); + int (*decompress_data_12) (j_decompress_ptr cinfo, J12SAMPIMAGE output_buf); +#ifdef D_LOSSLESS_SUPPORTED + int (*decompress_data_16) (j_decompress_ptr cinfo, J16SAMPIMAGE output_buf); +#endif + + /* These variables keep track of the current location of the input side. */ + /* cinfo->input_iMCU_row is also used for this. */ + JDIMENSION MCU_ctr; /* counts MCUs processed in current row */ + int MCU_vert_offset; /* counts MCU rows within iMCU row */ + int MCU_rows_per_iMCU_row; /* number of such rows needed */ + + /* The output side's location is represented by cinfo->output_iMCU_row. */ + + /* Lossy mode */ + /* Pointer to array of coefficient virtual arrays, or NULL if none */ + jvirt_barray_ptr *coef_arrays; +}; + +/* Decompression postprocessing (color quantization buffer control) */ +struct jpeg_d_post_controller { + void (*start_pass) (j_decompress_ptr cinfo, J_BUF_MODE pass_mode); + void (*post_process_data) (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, + JSAMPARRAY output_buf, JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); + void (*post_process_data_12) (j_decompress_ptr cinfo, J12SAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, + J12SAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +#ifdef D_LOSSLESS_SUPPORTED + void (*post_process_data_16) (j_decompress_ptr cinfo, J16SAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, + J16SAMPARRAY output_buf, + JDIMENSION *out_row_ctr, + JDIMENSION out_rows_avail); +#endif +}; + +/* Marker reading & parsing */ +struct jpeg_marker_reader { + void (*reset_marker_reader) (j_decompress_ptr cinfo); + /* Read markers until SOS or EOI. + * Returns same codes as are defined for jpeg_consume_input: + * JPEG_SUSPENDED, JPEG_REACHED_SOS, or JPEG_REACHED_EOI. + */ + int (*read_markers) (j_decompress_ptr cinfo); + /* Read a restart marker --- exported for use by entropy decoder only */ + jpeg_marker_parser_method read_restart_marker; + + /* State of marker reader --- nominally internal, but applications + * supplying COM or APPn handlers might like to know the state. + */ + boolean saw_SOI; /* found SOI? */ + boolean saw_SOF; /* found SOF? */ + int next_restart_num; /* next restart number expected (0-7) */ + unsigned int discarded_bytes; /* # of bytes skipped looking for a marker */ +}; + +/* Entropy decoding */ +struct jpeg_entropy_decoder { + void (*start_pass) (j_decompress_ptr cinfo); + + /* Lossy mode */ + boolean (*decode_mcu) (j_decompress_ptr cinfo, JBLOCKROW *MCU_data); + /* Lossless mode */ + JDIMENSION (*decode_mcus) (j_decompress_ptr cinfo, JDIFFIMAGE diff_buf, + JDIMENSION MCU_row_num, JDIMENSION MCU_col_num, + JDIMENSION nMCU); + boolean (*process_restart) (j_decompress_ptr cinfo); + + /* This is here to share code between baseline and progressive decoders; */ + /* other modules probably should not use it */ + boolean insufficient_data; /* set TRUE after emitting warning */ +}; + +/* Lossy mode: Inverse DCT (also performs dequantization) + * Lossless mode: Prediction, sample undifferencing, point transform, and + * sample size scaling + */ +typedef void (*inverse_DCT_method_ptr) (j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, + JSAMPARRAY output_buf, + JDIMENSION output_col); +typedef void (*inverse_DCT_12_method_ptr) (j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, + J12SAMPARRAY output_buf, + JDIMENSION output_col); + +struct jpeg_inverse_dct { + void (*start_pass) (j_decompress_ptr cinfo); + + /* Lossy mode */ + /* It is useful to allow each component to have a separate IDCT method. */ + inverse_DCT_method_ptr inverse_DCT[MAX_COMPONENTS]; + inverse_DCT_12_method_ptr inverse_DCT_12[MAX_COMPONENTS]; +}; + +/* Upsampling (note that upsampler must also call color converter) */ +struct jpeg_upsampler { + void (*start_pass) (j_decompress_ptr cinfo); + void (*upsample) (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, JSAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); + void (*upsample_12) (j_decompress_ptr cinfo, J12SAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, J12SAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); +#ifdef D_LOSSLESS_SUPPORTED + void (*upsample_16) (j_decompress_ptr cinfo, J16SAMPIMAGE input_buf, + JDIMENSION *in_row_group_ctr, + JDIMENSION in_row_groups_avail, J16SAMPARRAY output_buf, + JDIMENSION *out_row_ctr, JDIMENSION out_rows_avail); +#endif + + boolean need_context_rows; /* TRUE if need rows above & below */ +}; + +/* Colorspace conversion */ +struct jpeg_color_deconverter { + void (*start_pass) (j_decompress_ptr cinfo); + void (*color_convert) (j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION input_row, JSAMPARRAY output_buf, + int num_rows); + void (*color_convert_12) (j_decompress_ptr cinfo, J12SAMPIMAGE input_buf, + JDIMENSION input_row, J12SAMPARRAY output_buf, + int num_rows); +#ifdef D_LOSSLESS_SUPPORTED + void (*color_convert_16) (j_decompress_ptr cinfo, J16SAMPIMAGE input_buf, + JDIMENSION input_row, J16SAMPARRAY output_buf, + int num_rows); +#endif +}; + +/* Color quantization or color precision reduction */ +struct jpeg_color_quantizer { + void (*start_pass) (j_decompress_ptr cinfo, boolean is_pre_scan); + void (*color_quantize) (j_decompress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPARRAY output_buf, int num_rows); + void (*color_quantize_12) (j_decompress_ptr cinfo, J12SAMPARRAY input_buf, + J12SAMPARRAY output_buf, int num_rows); + void (*finish_pass) (j_decompress_ptr cinfo); + void (*new_color_map) (j_decompress_ptr cinfo); +}; + + +/* Miscellaneous useful macros */ + +#undef MAX +#define MAX(a, b) ((a) > (b) ? (a) : (b)) +#undef MIN +#define MIN(a, b) ((a) < (b) ? (a) : (b)) + +#ifdef ZERO_BUFFERS +#define MALLOC(size) calloc(1, size) +#else +#define MALLOC(size) malloc(size) +#endif + + +/* We assume that right shift corresponds to signed division by 2 with + * rounding towards minus infinity. This is correct for typical "arithmetic + * shift" instructions that shift in copies of the sign bit. But some + * C compilers implement >> with an unsigned shift. For these machines you + * must define RIGHT_SHIFT_IS_UNSIGNED. + * RIGHT_SHIFT provides a proper signed right shift of a JLONG quantity. + * It is only applied with constant shift counts. SHIFT_TEMPS must be + * included in the variables of any routine using RIGHT_SHIFT. + */ + +#ifdef RIGHT_SHIFT_IS_UNSIGNED +#define SHIFT_TEMPS JLONG shift_temp; +#define RIGHT_SHIFT(x, shft) \ + ((shift_temp = (x)) < 0 ? \ + (shift_temp >> (shft)) | ((~((JLONG)0)) << (32 - (shft))) : \ + (shift_temp >> (shft))) +#else +#define SHIFT_TEMPS +#define RIGHT_SHIFT(x, shft) ((x) >> (shft)) +#endif + + +/* Compression module initialization routines */ +EXTERN(void) jinit_compress_master(j_compress_ptr cinfo); +EXTERN(void) jinit_c_master_control(j_compress_ptr cinfo, + boolean transcode_only); +EXTERN(void) jinit_c_main_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_c_main_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_c_prep_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_c_prep_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_c_coef_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_c_coef_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_color_converter(j_compress_ptr cinfo); +EXTERN(void) j12init_color_converter(j_compress_ptr cinfo); +EXTERN(void) jinit_downsampler(j_compress_ptr cinfo); +EXTERN(void) j12init_downsampler(j_compress_ptr cinfo); +EXTERN(void) jinit_forward_dct(j_compress_ptr cinfo); +EXTERN(void) j12init_forward_dct(j_compress_ptr cinfo); +EXTERN(void) jinit_huff_encoder(j_compress_ptr cinfo); +EXTERN(void) jinit_phuff_encoder(j_compress_ptr cinfo); +EXTERN(void) jinit_arith_encoder(j_compress_ptr cinfo); +EXTERN(void) jinit_marker_writer(j_compress_ptr cinfo); +#ifdef C_LOSSLESS_SUPPORTED +EXTERN(void) j16init_c_main_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j16init_c_prep_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j16init_color_converter(j_compress_ptr cinfo); +EXTERN(void) j16init_downsampler(j_compress_ptr cinfo); +EXTERN(void) jinit_c_diff_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_c_diff_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j16init_c_diff_controller(j_compress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_lhuff_encoder(j_compress_ptr cinfo); +EXTERN(void) jinit_lossless_compressor(j_compress_ptr cinfo); +EXTERN(void) j12init_lossless_compressor(j_compress_ptr cinfo); +EXTERN(void) j16init_lossless_compressor(j_compress_ptr cinfo); +#endif + +/* Decompression module initialization routines */ +EXTERN(void) jinit_master_decompress(j_decompress_ptr cinfo); +EXTERN(void) jinit_d_main_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_d_main_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_d_coef_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_d_coef_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_d_post_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_d_post_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_input_controller(j_decompress_ptr cinfo); +EXTERN(void) jinit_marker_reader(j_decompress_ptr cinfo); +EXTERN(void) jinit_huff_decoder(j_decompress_ptr cinfo); +EXTERN(void) jinit_phuff_decoder(j_decompress_ptr cinfo); +EXTERN(void) jinit_arith_decoder(j_decompress_ptr cinfo); +EXTERN(void) jinit_inverse_dct(j_decompress_ptr cinfo); +EXTERN(void) j12init_inverse_dct(j_decompress_ptr cinfo); +EXTERN(void) jinit_upsampler(j_decompress_ptr cinfo); +EXTERN(void) j12init_upsampler(j_decompress_ptr cinfo); +EXTERN(void) jinit_color_deconverter(j_decompress_ptr cinfo); +EXTERN(void) j12init_color_deconverter(j_decompress_ptr cinfo); +EXTERN(void) jinit_1pass_quantizer(j_decompress_ptr cinfo); +EXTERN(void) j12init_1pass_quantizer(j_decompress_ptr cinfo); +EXTERN(void) jinit_2pass_quantizer(j_decompress_ptr cinfo); +EXTERN(void) j12init_2pass_quantizer(j_decompress_ptr cinfo); +EXTERN(void) jinit_merged_upsampler(j_decompress_ptr cinfo); +EXTERN(void) j12init_merged_upsampler(j_decompress_ptr cinfo); +#ifdef D_LOSSLESS_SUPPORTED +EXTERN(void) j16init_d_main_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j16init_d_post_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j16init_upsampler(j_decompress_ptr cinfo); +EXTERN(void) j16init_color_deconverter(j_decompress_ptr cinfo); +EXTERN(void) jinit_d_diff_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j12init_d_diff_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) j16init_d_diff_controller(j_decompress_ptr cinfo, + boolean need_full_buffer); +EXTERN(void) jinit_lhuff_decoder(j_decompress_ptr cinfo); +EXTERN(void) jinit_lossless_decompressor(j_decompress_ptr cinfo); +EXTERN(void) j12init_lossless_decompressor(j_decompress_ptr cinfo); +EXTERN(void) j16init_lossless_decompressor(j_decompress_ptr cinfo); +#endif + +/* Memory manager initialization */ +EXTERN(void) jinit_memory_mgr(j_common_ptr cinfo); + +/* Utility routines in jutils.c */ +EXTERN(long) jdiv_round_up(long a, long b); +EXTERN(long) jround_up(long a, long b); +EXTERN(void) jcopy_sample_rows(JSAMPARRAY input_array, int source_row, + JSAMPARRAY output_array, int dest_row, + int num_rows, JDIMENSION num_cols); +EXTERN(void) j12copy_sample_rows(J12SAMPARRAY input_array, int source_row, + J12SAMPARRAY output_array, int dest_row, + int num_rows, JDIMENSION num_cols); +#if defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) +EXTERN(void) j16copy_sample_rows(J16SAMPARRAY input_array, int source_row, + J16SAMPARRAY output_array, int dest_row, + int num_rows, JDIMENSION num_cols); +#endif +EXTERN(void) jcopy_block_row(JBLOCKROW input_row, JBLOCKROW output_row, + JDIMENSION num_blocks); +EXTERN(void) jzero_far(void *target, size_t bytestozero); +/* Constant tables in jutils.c */ +#if 0 /* This table is not actually needed in v6a */ +extern const int jpeg_zigzag_order[]; /* natural coef order to zigzag order */ +#endif +extern const int jpeg_natural_order[]; /* zigzag coef order to natural order */ + +/* Arithmetic coding probability estimation tables in jaricom.c */ +extern const JLONG jpeg_aritab[]; diff --git a/src/spice2x/external/libjpeg-turbo/jpeglib.h b/src/spice2x/external/libjpeg-turbo/jpeglib.h new file mode 100644 index 0000000..17e7833 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jpeglib.h @@ -0,0 +1,1218 @@ +/* + * jpeglib.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1998, Thomas G. Lane. + * Modified 2002-2009 by Guido Vollbeding. + * Lossless JPEG Modifications: + * Copyright (C) 1999, Ken Murchison. + * libjpeg-turbo Modifications: + * Copyright (C) 2009-2011, 2013-2014, 2016-2017, 2020, 2022-2024, + D. R. Commander. + * Copyright (C) 2015, Google, Inc. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file defines the application interface for the JPEG library. + * Most applications using the library need only include this file, + * and perhaps jerror.h if they want to know the exact error codes. + */ + +/* NOTE: This header file does not include stdio.h, despite the fact that it + * uses FILE and size_t. That is by design, since the libjpeg API predates the + * widespread adoption of ANSI/ISO C. Referring to libjpeg.txt, it is a + * documented requirement that calling programs "include system headers that + * define at least the typedefs FILE and size_t" before including jpeglib.h. + * Technically speaking, changing that requirement by including stdio.h here + * would break backward API compatibility. Please do not file bug reports, + * feature requests, or pull requests regarding this. + */ + +#ifndef JPEGLIB_H +#define JPEGLIB_H + +/* + * First we include the configuration files that record how this + * installation of the JPEG library is set up. jconfig.h can be + * generated automatically for many systems. jmorecfg.h contains + * manual configuration options that most people need not worry about. + */ + +#ifndef JCONFIG_INCLUDED /* in case jinclude.h already did */ +#include "jconfig.h" /* widely used configuration options */ +#endif +#include "jmorecfg.h" /* seldom changed options */ + + +#ifdef __cplusplus +#ifndef DONT_USE_EXTERN_C +extern "C" { +#endif +#endif + + +/* Various constants determining the sizes of things. + * All of these are specified by the JPEG standard, so don't change them + * if you want to be compatible. + */ + +/* NOTE: In lossless mode, an MCU contains one or more samples rather than one + * or more 8x8 DCT blocks, so the term "data unit" is used to generically + * describe a sample in lossless mode or an 8x8 DCT block in lossy mode. To + * preserve backward API/ABI compatibility, the field and macro names retain + * the "block" terminology. + */ + +#define DCTSIZE 8 /* The basic DCT block is 8x8 samples */ +#define DCTSIZE2 64 /* DCTSIZE squared; # of elements in a block */ +#define NUM_QUANT_TBLS 4 /* Quantization tables are numbered 0..3 */ +#define NUM_HUFF_TBLS 4 /* Huffman tables are numbered 0..3 */ +#define NUM_ARITH_TBLS 16 /* Arith-coding tables are numbered 0..15 */ +#define MAX_COMPS_IN_SCAN 4 /* JPEG limit on # of components in one scan */ +#define MAX_SAMP_FACTOR 4 /* JPEG limit on sampling factors */ +/* Unfortunately, some bozo at Adobe saw no reason to be bound by the standard; + * the PostScript DCT filter can emit files with many more than 10 blocks/MCU. + * If you happen to run across such a file, you can up D_MAX_BLOCKS_IN_MCU + * to handle it. We even let you do this from the jconfig.h file. However, + * we strongly discourage changing C_MAX_BLOCKS_IN_MCU; just because Adobe + * sometimes emits noncompliant files doesn't mean you should too. + */ +#define C_MAX_BLOCKS_IN_MCU 10 /* compressor's limit on data units/MCU */ +#ifndef D_MAX_BLOCKS_IN_MCU +#define D_MAX_BLOCKS_IN_MCU 10 /* decompressor's limit on data units/MCU */ +#endif + + +/* Data structures for images (arrays of samples and of DCT coefficients). + */ + +typedef JSAMPLE *JSAMPROW; /* ptr to one image row of pixel samples. */ +typedef JSAMPROW *JSAMPARRAY; /* ptr to some rows (a 2-D sample array) */ +typedef JSAMPARRAY *JSAMPIMAGE; /* a 3-D sample array: top index is color */ + +typedef J12SAMPLE *J12SAMPROW; /* ptr to one image row of 12-bit pixel + samples. */ +typedef J12SAMPROW *J12SAMPARRAY; /* ptr to some 12-bit sample rows (a 2-D + 12-bit sample array) */ +typedef J12SAMPARRAY *J12SAMPIMAGE; /* a 3-D 12-bit sample array: top index is + color */ + +typedef J16SAMPLE *J16SAMPROW; /* ptr to one image row of 16-bit pixel + samples. */ +typedef J16SAMPROW *J16SAMPARRAY; /* ptr to some 16-bit sample rows (a 2-D + 16-bit sample array) */ +typedef J16SAMPARRAY *J16SAMPIMAGE; /* a 3-D 16-bit sample array: top index is + color */ + +typedef JCOEF JBLOCK[DCTSIZE2]; /* one block of coefficients */ +typedef JBLOCK *JBLOCKROW; /* pointer to one row of coefficient blocks */ +typedef JBLOCKROW *JBLOCKARRAY; /* a 2-D array of coefficient blocks */ +typedef JBLOCKARRAY *JBLOCKIMAGE; /* a 3-D array of coefficient blocks */ + +typedef JCOEF *JCOEFPTR; /* useful in a couple of places */ + + +/* Types for JPEG compression parameters and working tables. */ + + +/* DCT coefficient quantization tables. */ + +typedef struct { + /* This array gives the coefficient quantizers in natural array order + * (not the zigzag order in which they are stored in a JPEG DQT marker). + * CAUTION: IJG versions prior to v6a kept this array in zigzag order. + */ + UINT16 quantval[DCTSIZE2]; /* quantization step for each coefficient */ + /* This field is used only during compression. It's initialized FALSE when + * the table is created, and set TRUE when it's been output to the file. + * You could suppress output of a table by setting this to TRUE. + * (See jpeg_suppress_tables for an example.) + */ + boolean sent_table; /* TRUE when table has been output */ +} JQUANT_TBL; + + +/* Huffman coding tables. */ + +typedef struct { + /* These two fields directly represent the contents of a JPEG DHT marker */ + UINT8 bits[17]; /* bits[k] = # of symbols with codes of */ + /* length k bits; bits[0] is unused */ + UINT8 huffval[256]; /* The symbols, in order of incr code length */ + /* This field is used only during compression. It's initialized FALSE when + * the table is created, and set TRUE when it's been output to the file. + * You could suppress output of a table by setting this to TRUE. + * (See jpeg_suppress_tables for an example.) + */ + boolean sent_table; /* TRUE when table has been output */ +} JHUFF_TBL; + + +/* Basic info about one component (color channel). */ + +typedef struct { + /* These values are fixed over the whole image. */ + /* For compression, they must be supplied by parameter setup; */ + /* for decompression, they are read from the SOF marker. */ + int component_id; /* identifier for this component (0..255) */ + int component_index; /* its index in SOF or cinfo->comp_info[] */ + int h_samp_factor; /* horizontal sampling factor (1..4) */ + int v_samp_factor; /* vertical sampling factor (1..4) */ + int quant_tbl_no; /* quantization table selector (0..3) */ + /* These values may vary between scans. */ + /* For compression, they must be supplied by parameter setup; */ + /* for decompression, they are read from the SOS marker. */ + /* The decompressor output side may not use these variables. */ + int dc_tbl_no; /* DC entropy table selector (0..3) */ + int ac_tbl_no; /* AC entropy table selector (0..3) */ + + /* Remaining fields should be treated as private by applications. */ + + /* These values are computed during compression or decompression startup: */ + /* Component's size in data units. + * In lossy mode, any dummy blocks added to complete an MCU are not counted; + * therefore these values do not depend on whether a scan is interleaved or + * not. In lossless mode, these are always equal to the image width and + * height. + */ + JDIMENSION width_in_blocks; + JDIMENSION height_in_blocks; + /* Size of a data unit in samples. Always DCTSIZE for lossy compression. + * For lossy decompression this is the size of the output from one DCT block, + * reflecting any scaling we choose to apply during the IDCT step. + * Values from 1 to 16 are supported. Note that different components may + * receive different IDCT scalings. In lossless mode, this is always equal + * to 1. + */ +#if JPEG_LIB_VERSION >= 70 + int DCT_h_scaled_size; + int DCT_v_scaled_size; +#else + int DCT_scaled_size; +#endif + /* The downsampled dimensions are the component's actual, unpadded number + * of samples at the main buffer (preprocessing/compression interface), thus + * downsampled_width = ceil(image_width * Hi/Hmax) + * and similarly for height. For lossy decompression, IDCT scaling is + * included, so + * downsampled_width = ceil(image_width * Hi/Hmax * DCT_[h_]scaled_size/DCTSIZE) + * In lossless mode, these are always equal to the image width and height. + */ + JDIMENSION downsampled_width; /* actual width in samples */ + JDIMENSION downsampled_height; /* actual height in samples */ + /* This flag is used only for decompression. In cases where some of the + * components will be ignored (eg grayscale output from YCbCr image), + * we can skip most computations for the unused components. + */ + boolean component_needed; /* do we need the value of this component? */ + + /* These values are computed before starting a scan of the component. */ + /* The decompressor output side may not use these variables. */ + int MCU_width; /* number of data units per MCU, horizontally */ + int MCU_height; /* number of data units per MCU, vertically */ + int MCU_blocks; /* MCU_width * MCU_height */ + int MCU_sample_width; /* MCU width in samples, MCU_width*DCT_[h_]scaled_size */ + int last_col_width; /* # of non-dummy data units across in last MCU */ + int last_row_height; /* # of non-dummy data units down in last MCU */ + + /* Saved quantization table for component; NULL if none yet saved. + * See jdinput.c comments about the need for this information. + * This field is currently used only for decompression. + */ + JQUANT_TBL *quant_table; + + /* Private per-component storage for DCT or IDCT subsystem. */ + void *dct_table; +} jpeg_component_info; + + +/* The script for encoding a multiple-scan file is an array of these: */ + +typedef struct { + int comps_in_scan; /* number of components encoded in this scan */ + int component_index[MAX_COMPS_IN_SCAN]; /* their SOF/comp_info[] indexes */ + int Ss, Se; /* progressive JPEG spectral selection parms + (Ss is the predictor selection value in + lossless mode) */ + int Ah, Al; /* progressive JPEG successive approx. parms + (Al is the point transform value in lossless + mode) */ +} jpeg_scan_info; + +/* The decompressor can save APPn and COM markers in a list of these: */ + +typedef struct jpeg_marker_struct *jpeg_saved_marker_ptr; + +struct jpeg_marker_struct { + jpeg_saved_marker_ptr next; /* next in list, or NULL */ + UINT8 marker; /* marker code: JPEG_COM, or JPEG_APP0+n */ + unsigned int original_length; /* # bytes of data in the file */ + unsigned int data_length; /* # bytes of data saved at data[] */ + JOCTET *data; /* the data contained in the marker */ + /* the marker length word is not counted in data_length or original_length */ +}; + +/* Known color spaces. */ + +#define JCS_EXTENSIONS 1 +#define JCS_ALPHA_EXTENSIONS 1 + +typedef enum { + JCS_UNKNOWN, /* error/unspecified */ + JCS_GRAYSCALE, /* monochrome */ + JCS_RGB, /* red/green/blue as specified by the RGB_RED, + RGB_GREEN, RGB_BLUE, and RGB_PIXELSIZE macros */ + JCS_YCbCr, /* Y/Cb/Cr (also known as YUV) */ + JCS_CMYK, /* C/M/Y/K */ + JCS_YCCK, /* Y/Cb/Cr/K */ + JCS_EXT_RGB, /* red/green/blue */ + JCS_EXT_RGBX, /* red/green/blue/x */ + JCS_EXT_BGR, /* blue/green/red */ + JCS_EXT_BGRX, /* blue/green/red/x */ + JCS_EXT_XBGR, /* x/blue/green/red */ + JCS_EXT_XRGB, /* x/red/green/blue */ + /* When out_color_space it set to JCS_EXT_RGBX, JCS_EXT_BGRX, JCS_EXT_XBGR, + or JCS_EXT_XRGB during decompression, the X byte is undefined, and in + order to ensure the best performance, libjpeg-turbo can set that byte to + whatever value it wishes. Use the following colorspace constants to + ensure that the X byte is set to 0xFF, so that it can be interpreted as an + opaque alpha channel. */ + JCS_EXT_RGBA, /* red/green/blue/alpha */ + JCS_EXT_BGRA, /* blue/green/red/alpha */ + JCS_EXT_ABGR, /* alpha/blue/green/red */ + JCS_EXT_ARGB, /* alpha/red/green/blue */ + JCS_RGB565 /* 5-bit red/6-bit green/5-bit blue + [decompression only] */ +} J_COLOR_SPACE; + +/* DCT/IDCT algorithm options. */ + +typedef enum { + JDCT_ISLOW, /* accurate integer method */ + JDCT_IFAST, /* less accurate integer method [legacy feature] */ + JDCT_FLOAT /* floating-point method [legacy feature] */ +} J_DCT_METHOD; + +#ifndef JDCT_DEFAULT /* may be overridden in jconfig.h */ +#define JDCT_DEFAULT JDCT_ISLOW +#endif +#ifndef JDCT_FASTEST /* may be overridden in jconfig.h */ +#define JDCT_FASTEST JDCT_IFAST +#endif + +/* Dithering options for decompression. */ + +typedef enum { + JDITHER_NONE, /* no dithering */ + JDITHER_ORDERED, /* simple ordered dither */ + JDITHER_FS /* Floyd-Steinberg error diffusion dither */ +} J_DITHER_MODE; + + +/* Common fields between JPEG compression and decompression master structs. */ + +#define jpeg_common_fields \ + struct jpeg_error_mgr *err; /* Error handler module */ \ + struct jpeg_memory_mgr *mem; /* Memory manager module */ \ + struct jpeg_progress_mgr *progress; /* Progress monitor, or NULL if none */ \ + void *client_data; /* Available for use by application */ \ + boolean is_decompressor; /* So common code can tell which is which */ \ + int global_state /* For checking call sequence validity */ + +/* Routines that are to be used by both halves of the library are declared + * to receive a pointer to this structure. There are no actual instances of + * jpeg_common_struct, only of jpeg_compress_struct and jpeg_decompress_struct. + */ +struct jpeg_common_struct { + jpeg_common_fields; /* Fields common to both master struct types */ + /* Additional fields follow in an actual jpeg_compress_struct or + * jpeg_decompress_struct. All three structs must agree on these + * initial fields! (This would be a lot cleaner in C++.) + */ +}; + +typedef struct jpeg_common_struct *j_common_ptr; +typedef struct jpeg_compress_struct *j_compress_ptr; +typedef struct jpeg_decompress_struct *j_decompress_ptr; + + +/* Master record for a compression instance */ + +struct jpeg_compress_struct { + jpeg_common_fields; /* Fields shared with jpeg_decompress_struct */ + + /* Destination for compressed data */ + struct jpeg_destination_mgr *dest; + + /* Description of source image --- these fields must be filled in by + * outer application before starting compression. in_color_space must + * be correct before you can even call jpeg_set_defaults(). + */ + + JDIMENSION image_width; /* input image width */ + JDIMENSION image_height; /* input image height */ + int input_components; /* # of color components in input image */ + J_COLOR_SPACE in_color_space; /* colorspace of input image */ + + double input_gamma; /* image gamma of input image */ + + /* Compression parameters --- these fields must be set before calling + * jpeg_start_compress(). We recommend calling jpeg_set_defaults() to + * initialize everything to reasonable defaults, then changing anything + * the application specifically wants to change. That way you won't get + * burnt when new parameters are added. Also note that there are several + * helper routines to simplify changing parameters. + */ + +#if JPEG_LIB_VERSION >= 70 + unsigned int scale_num, scale_denom; /* fraction by which to scale image */ + + JDIMENSION jpeg_width; /* scaled JPEG image width */ + JDIMENSION jpeg_height; /* scaled JPEG image height */ + /* Dimensions of actual JPEG image that will be written to file, + * derived from input dimensions by scaling factors above. + * These fields are computed by jpeg_start_compress(). + * You can also use jpeg_calc_jpeg_dimensions() to determine these values + * in advance of calling jpeg_start_compress(). + */ +#endif + + int data_precision; /* bits of precision in image data */ + + int num_components; /* # of color components in JPEG image */ + J_COLOR_SPACE jpeg_color_space; /* colorspace of JPEG image */ + + jpeg_component_info *comp_info; + /* comp_info[i] describes component that appears i'th in SOF */ + + JQUANT_TBL *quant_tbl_ptrs[NUM_QUANT_TBLS]; +#if JPEG_LIB_VERSION >= 70 + int q_scale_factor[NUM_QUANT_TBLS]; +#endif + /* ptrs to coefficient quantization tables, or NULL if not defined, + * and corresponding scale factors (percentage, initialized 100). + */ + + JHUFF_TBL *dc_huff_tbl_ptrs[NUM_HUFF_TBLS]; + JHUFF_TBL *ac_huff_tbl_ptrs[NUM_HUFF_TBLS]; + /* ptrs to Huffman coding tables, or NULL if not defined */ + + UINT8 arith_dc_L[NUM_ARITH_TBLS]; /* L values for DC arith-coding tables */ + UINT8 arith_dc_U[NUM_ARITH_TBLS]; /* U values for DC arith-coding tables */ + UINT8 arith_ac_K[NUM_ARITH_TBLS]; /* Kx values for AC arith-coding tables */ + + int num_scans; /* # of entries in scan_info array */ + const jpeg_scan_info *scan_info; /* script for multi-scan file, or NULL */ + /* The default value of scan_info is NULL, which causes a single-scan + * sequential JPEG file to be emitted. To create a multi-scan file, + * set num_scans and scan_info to point to an array of scan definitions. + */ + + boolean raw_data_in; /* TRUE=caller supplies downsampled data */ + boolean arith_code; /* TRUE=arithmetic coding, FALSE=Huffman */ + boolean optimize_coding; /* TRUE=optimize entropy encoding parms */ + boolean CCIR601_sampling; /* TRUE=first samples are cosited */ +#if JPEG_LIB_VERSION >= 70 + boolean do_fancy_downsampling; /* TRUE=apply fancy downsampling */ +#endif + int smoothing_factor; /* 1..100, or 0 for no input smoothing */ + J_DCT_METHOD dct_method; /* DCT algorithm selector */ + + /* The restart interval can be specified in absolute MCUs by setting + * restart_interval, or in MCU rows by setting restart_in_rows + * (in which case the correct restart_interval will be figured + * for each scan). + */ + unsigned int restart_interval; /* MCUs per restart, or 0 for no restart */ + int restart_in_rows; /* if > 0, MCU rows per restart interval */ + + /* Parameters controlling emission of special markers. */ + + boolean write_JFIF_header; /* should a JFIF marker be written? */ + UINT8 JFIF_major_version; /* What to write for the JFIF version number */ + UINT8 JFIF_minor_version; + /* These three values are not used by the JPEG code, merely copied */ + /* into the JFIF APP0 marker. density_unit can be 0 for unknown, */ + /* 1 for dots/inch, or 2 for dots/cm. Note that the pixel aspect */ + /* ratio is defined by X_density/Y_density even when density_unit=0. */ + UINT8 density_unit; /* JFIF code for pixel size units */ + UINT16 X_density; /* Horizontal pixel density */ + UINT16 Y_density; /* Vertical pixel density */ + boolean write_Adobe_marker; /* should an Adobe marker be written? */ + + /* State variable: index of next scanline to be written to + * jpeg_write_scanlines(). Application may use this to control its + * processing loop, e.g., "while (next_scanline < image_height)". + */ + + JDIMENSION next_scanline; /* 0 .. image_height-1 */ + + /* Remaining fields are known throughout compressor, but generally + * should not be touched by a surrounding application. + */ + + /* + * These fields are computed during compression startup + */ + boolean progressive_mode; /* TRUE if scan script uses progressive mode */ + int max_h_samp_factor; /* largest h_samp_factor */ + int max_v_samp_factor; /* largest v_samp_factor */ + +#if JPEG_LIB_VERSION >= 70 + int min_DCT_h_scaled_size; /* smallest DCT_h_scaled_size of any component */ + int min_DCT_v_scaled_size; /* smallest DCT_v_scaled_size of any component */ +#endif + + JDIMENSION total_iMCU_rows; /* # of iMCU rows to be input to coefficient or + difference controller */ + /* The coefficient or difference controller receives data in units of MCU + * rows as defined for fully interleaved scans (whether the JPEG file is + * interleaved or not). In lossy mode, there are v_samp_factor * DCTSIZE + * sample rows of each component in an "iMCU" (interleaved MCU) row. In + * lossless mode, total_iMCU_rows is always equal to the image height. + */ + + /* + * These fields are valid during any one scan. + * They describe the components and MCUs actually appearing in the scan. + */ + int comps_in_scan; /* # of JPEG components in this scan */ + jpeg_component_info *cur_comp_info[MAX_COMPS_IN_SCAN]; + /* *cur_comp_info[i] describes component that appears i'th in SOS */ + + JDIMENSION MCUs_per_row; /* # of MCUs across the image */ + JDIMENSION MCU_rows_in_scan; /* # of MCU rows in the image */ + + int blocks_in_MCU; /* # of data units per MCU */ + int MCU_membership[C_MAX_BLOCKS_IN_MCU]; + /* MCU_membership[i] is index in cur_comp_info of component owning */ + /* i'th data unit in an MCU */ + + int Ss, Se, Ah, Al; /* progressive/lossless JPEG parameters for + scan */ + +#if JPEG_LIB_VERSION >= 80 + int block_size; /* the basic DCT block size: 1..16 */ + const int *natural_order; /* natural-order position array */ + int lim_Se; /* min( Se, DCTSIZE2-1 ) */ +#endif + + /* + * Links to compression subobjects (methods and private variables of modules) + */ + struct jpeg_comp_master *master; + struct jpeg_c_main_controller *main; + struct jpeg_c_prep_controller *prep; + struct jpeg_c_coef_controller *coef; + struct jpeg_marker_writer *marker; + struct jpeg_color_converter *cconvert; + struct jpeg_downsampler *downsample; + struct jpeg_forward_dct *fdct; + struct jpeg_entropy_encoder *entropy; + jpeg_scan_info *script_space; /* workspace for jpeg_simple_progression */ + int script_space_size; +}; + + +/* Master record for a decompression instance */ + +struct jpeg_decompress_struct { + jpeg_common_fields; /* Fields shared with jpeg_compress_struct */ + + /* Source of compressed data */ + struct jpeg_source_mgr *src; + + /* Basic description of image --- filled in by jpeg_read_header(). */ + /* Application may inspect these values to decide how to process image. */ + + JDIMENSION image_width; /* nominal image width (from SOF marker) */ + JDIMENSION image_height; /* nominal image height */ + int num_components; /* # of color components in JPEG image */ + J_COLOR_SPACE jpeg_color_space; /* colorspace of JPEG image */ + + /* Decompression processing parameters --- these fields must be set before + * calling jpeg_start_decompress(). Note that jpeg_read_header() initializes + * them to default values. + */ + + J_COLOR_SPACE out_color_space; /* colorspace for output */ + + unsigned int scale_num, scale_denom; /* fraction by which to scale image */ + + double output_gamma; /* image gamma wanted in output */ + + boolean buffered_image; /* TRUE=multiple output passes */ + boolean raw_data_out; /* TRUE=downsampled data wanted */ + + J_DCT_METHOD dct_method; /* IDCT algorithm selector */ + boolean do_fancy_upsampling; /* TRUE=apply fancy upsampling */ + boolean do_block_smoothing; /* TRUE=apply interblock smoothing */ + + boolean quantize_colors; /* TRUE=colormapped output wanted */ + /* the following are ignored if not quantize_colors: */ + J_DITHER_MODE dither_mode; /* type of color dithering to use */ + boolean two_pass_quantize; /* TRUE=use two-pass color quantization */ + int desired_number_of_colors; /* max # colors to use in created colormap */ + /* these are significant only in buffered-image mode: */ + boolean enable_1pass_quant; /* enable future use of 1-pass quantizer */ + boolean enable_external_quant;/* enable future use of external colormap */ + boolean enable_2pass_quant; /* enable future use of 2-pass quantizer */ + + /* Description of actual output image that will be returned to application. + * These fields are computed by jpeg_start_decompress(). + * You can also use jpeg_calc_output_dimensions() to determine these values + * in advance of calling jpeg_start_decompress(). + */ + + JDIMENSION output_width; /* scaled image width */ + JDIMENSION output_height; /* scaled image height */ + int out_color_components; /* # of color components in out_color_space */ + int output_components; /* # of color components returned */ + /* output_components is 1 (a colormap index) when quantizing colors; + * otherwise it equals out_color_components. + */ + int rec_outbuf_height; /* min recommended height of scanline buffer */ + /* If the buffer passed to jpeg_read_scanlines() is less than this many rows + * high, space and time will be wasted due to unnecessary data copying. + * Usually rec_outbuf_height will be 1 or 2, at most 4. + */ + + /* When quantizing colors, the output colormap is described by these fields. + * The application can supply a colormap by setting colormap non-NULL before + * calling jpeg_start_decompress; otherwise a colormap is created during + * jpeg_start_decompress or jpeg_start_output. + * The map has out_color_components rows and actual_number_of_colors columns. + */ + int actual_number_of_colors; /* number of entries in use */ + JSAMPARRAY colormap; /* The color map as a 2-D pixel array + If data_precision is 12, then this is + actually a J12SAMPARRAY, so callers must + type-cast it in order to read/write 12-bit + samples from/to the array. */ + + /* State variables: these variables indicate the progress of decompression. + * The application may examine these but must not modify them. + */ + + /* Row index of next scanline to be read from jpeg_read_scanlines(). + * Application may use this to control its processing loop, e.g., + * "while (output_scanline < output_height)". + */ + JDIMENSION output_scanline; /* 0 .. output_height-1 */ + + /* Current input scan number and number of iMCU rows completed in scan. + * These indicate the progress of the decompressor input side. + */ + int input_scan_number; /* Number of SOS markers seen so far */ + JDIMENSION input_iMCU_row; /* Number of iMCU rows completed */ + + /* The "output scan number" is the notional scan being displayed by the + * output side. The decompressor will not allow output scan/row number + * to get ahead of input scan/row, but it can fall arbitrarily far behind. + */ + int output_scan_number; /* Nominal scan number being displayed */ + JDIMENSION output_iMCU_row; /* Number of iMCU rows read */ + + /* Current progression status. coef_bits[c][i] indicates the precision + * with which component c's DCT coefficient i (in zigzag order) is known. + * It is -1 when no data has yet been received, otherwise it is the point + * transform (shift) value for the most recent scan of the coefficient + * (thus, 0 at completion of the progression). + * This pointer is NULL when reading a non-progressive file. + */ + int (*coef_bits)[DCTSIZE2]; /* -1 or current Al value for each coef */ + + /* Internal JPEG parameters --- the application usually need not look at + * these fields. Note that the decompressor output side may not use + * any parameters that can change between scans. + */ + + /* Quantization and Huffman tables are carried forward across input + * datastreams when processing abbreviated JPEG datastreams. + */ + + JQUANT_TBL *quant_tbl_ptrs[NUM_QUANT_TBLS]; + /* ptrs to coefficient quantization tables, or NULL if not defined */ + + JHUFF_TBL *dc_huff_tbl_ptrs[NUM_HUFF_TBLS]; + JHUFF_TBL *ac_huff_tbl_ptrs[NUM_HUFF_TBLS]; + /* ptrs to Huffman coding tables, or NULL if not defined */ + + /* These parameters are never carried across datastreams, since they + * are given in SOF/SOS markers or defined to be reset by SOI. + */ + + int data_precision; /* bits of precision in image data */ + + jpeg_component_info *comp_info; + /* comp_info[i] describes component that appears i'th in SOF */ + +#if JPEG_LIB_VERSION >= 80 + boolean is_baseline; /* TRUE if Baseline SOF0 encountered */ +#endif + boolean progressive_mode; /* TRUE if SOFn specifies progressive mode */ + boolean arith_code; /* TRUE=arithmetic coding, FALSE=Huffman */ + + UINT8 arith_dc_L[NUM_ARITH_TBLS]; /* L values for DC arith-coding tables */ + UINT8 arith_dc_U[NUM_ARITH_TBLS]; /* U values for DC arith-coding tables */ + UINT8 arith_ac_K[NUM_ARITH_TBLS]; /* Kx values for AC arith-coding tables */ + + unsigned int restart_interval; /* MCUs per restart interval, or 0 for no restart */ + + /* These fields record data obtained from optional markers recognized by + * the JPEG library. + */ + boolean saw_JFIF_marker; /* TRUE iff a JFIF APP0 marker was found */ + /* Data copied from JFIF marker; only valid if saw_JFIF_marker is TRUE: */ + UINT8 JFIF_major_version; /* JFIF version number */ + UINT8 JFIF_minor_version; + UINT8 density_unit; /* JFIF code for pixel size units */ + UINT16 X_density; /* Horizontal pixel density */ + UINT16 Y_density; /* Vertical pixel density */ + boolean saw_Adobe_marker; /* TRUE iff an Adobe APP14 marker was found */ + UINT8 Adobe_transform; /* Color transform code from Adobe marker */ + + boolean CCIR601_sampling; /* TRUE=first samples are cosited */ + + /* Aside from the specific data retained from APPn markers known to the + * library, the uninterpreted contents of any or all APPn and COM markers + * can be saved in a list for examination by the application. + */ + jpeg_saved_marker_ptr marker_list; /* Head of list of saved markers */ + + /* Remaining fields are known throughout decompressor, but generally + * should not be touched by a surrounding application. + */ + + /* + * These fields are computed during decompression startup + */ + int max_h_samp_factor; /* largest h_samp_factor */ + int max_v_samp_factor; /* largest v_samp_factor */ + +#if JPEG_LIB_VERSION >= 70 + int min_DCT_h_scaled_size; /* smallest DCT_h_scaled_size of any component */ + int min_DCT_v_scaled_size; /* smallest DCT_v_scaled_size of any component */ +#else + int min_DCT_scaled_size; /* smallest DCT_scaled_size of any component */ +#endif + + JDIMENSION total_iMCU_rows; /* # of iMCU rows in image */ + /* The coefficient or difference controller's input and output progress is + * measured in units of "iMCU" (interleaved MCU) rows. These are the same as + * MCU rows in fully interleaved JPEG scans, but are used whether the scan is + * interleaved or not. In lossy mode, we define an iMCU row as v_samp_factor + * DCT block rows of each component. Therefore, the IDCT output contains + * v_samp_factor*DCT_[v_]scaled_size sample rows of a component per iMCU row. + * In lossless mode, total_iMCU_rows is always equal to the image height. + */ + + JSAMPLE *sample_range_limit; /* table for fast range-limiting + If data_precision is 12 or 16, then this is + actually a J12SAMPLE pointer or a J16SAMPLE + pointer, so callers must type-cast it in + order to read 12-bit or 16-bit samples from + the array. */ + + /* + * These fields are valid during any one scan. + * They describe the components and MCUs actually appearing in the scan. + * Note that the decompressor output side must not use these fields. + */ + int comps_in_scan; /* # of JPEG components in this scan */ + jpeg_component_info *cur_comp_info[MAX_COMPS_IN_SCAN]; + /* *cur_comp_info[i] describes component that appears i'th in SOS */ + + JDIMENSION MCUs_per_row; /* # of MCUs across the image */ + JDIMENSION MCU_rows_in_scan; /* # of MCU rows in the image */ + + int blocks_in_MCU; /* # of data units per MCU */ + int MCU_membership[D_MAX_BLOCKS_IN_MCU]; + /* MCU_membership[i] is index in cur_comp_info of component owning */ + /* i'th data unit in an MCU */ + + int Ss, Se, Ah, Al; /* progressive/lossless JPEG parameters for + scan */ + +#if JPEG_LIB_VERSION >= 80 + /* These fields are derived from Se of first SOS marker. + */ + int block_size; /* the basic DCT block size: 1..16 */ + const int *natural_order; /* natural-order position array for entropy decode */ + int lim_Se; /* min( Se, DCTSIZE2-1 ) for entropy decode */ +#endif + + /* This field is shared between entropy decoder and marker parser. + * It is either zero or the code of a JPEG marker that has been + * read from the data source, but has not yet been processed. + */ + int unread_marker; + + /* + * Links to decompression subobjects (methods, private variables of modules) + */ + struct jpeg_decomp_master *master; + struct jpeg_d_main_controller *main; + struct jpeg_d_coef_controller *coef; + struct jpeg_d_post_controller *post; + struct jpeg_input_controller *inputctl; + struct jpeg_marker_reader *marker; + struct jpeg_entropy_decoder *entropy; + struct jpeg_inverse_dct *idct; + struct jpeg_upsampler *upsample; + struct jpeg_color_deconverter *cconvert; + struct jpeg_color_quantizer *cquantize; +}; + + +/* "Object" declarations for JPEG modules that may be supplied or called + * directly by the surrounding application. + * As with all objects in the JPEG library, these structs only define the + * publicly visible methods and state variables of a module. Additional + * private fields may exist after the public ones. + */ + + +/* Error handler object */ + +struct jpeg_error_mgr { + /* Error exit handler: does not return to caller */ + void (*error_exit) (j_common_ptr cinfo); + /* Conditionally emit a trace or warning message */ + void (*emit_message) (j_common_ptr cinfo, int msg_level); + /* Routine that actually outputs a trace or error message */ + void (*output_message) (j_common_ptr cinfo); + /* Format a message string for the most recent JPEG error or message */ + void (*format_message) (j_common_ptr cinfo, char *buffer); +#define JMSG_LENGTH_MAX 200 /* recommended size of format_message buffer */ + /* Reset error state variables at start of a new image */ + void (*reset_error_mgr) (j_common_ptr cinfo); + + /* The message ID code and any parameters are saved here. + * A message can have one string parameter or up to 8 int parameters. + */ + int msg_code; +#define JMSG_STR_PARM_MAX 80 + union { + int i[8]; + char s[JMSG_STR_PARM_MAX]; + } msg_parm; + + /* Standard state variables for error facility */ + + int trace_level; /* max msg_level that will be displayed */ + + /* For recoverable corrupt-data errors, we emit a warning message, + * but keep going unless emit_message chooses to abort. emit_message + * should count warnings in num_warnings. The surrounding application + * can check for bad data by seeing if num_warnings is nonzero at the + * end of processing. + */ + long num_warnings; /* number of corrupt-data warnings */ + + /* These fields point to the table(s) of error message strings. + * An application can change the table pointer to switch to a different + * message list (typically, to change the language in which errors are + * reported). Some applications may wish to add additional error codes + * that will be handled by the JPEG library error mechanism; the second + * table pointer is used for this purpose. + * + * First table includes all errors generated by JPEG library itself. + * Error code 0 is reserved for a "no such error string" message. + */ + const char * const *jpeg_message_table; /* Library errors */ + int last_jpeg_message; /* Table contains strings 0..last_jpeg_message */ + /* Second table can be added by application (see cjpeg/djpeg for example). + * It contains strings numbered first_addon_message..last_addon_message. + */ + const char * const *addon_message_table; /* Non-library errors */ + int first_addon_message; /* code for first string in addon table */ + int last_addon_message; /* code for last string in addon table */ +}; + + +/* Progress monitor object */ + +struct jpeg_progress_mgr { + void (*progress_monitor) (j_common_ptr cinfo); + + long pass_counter; /* work units completed in this pass */ + long pass_limit; /* total number of work units in this pass */ + int completed_passes; /* passes completed so far */ + int total_passes; /* total number of passes expected */ +}; + + +/* Data destination object for compression */ + +struct jpeg_destination_mgr { + JOCTET *next_output_byte; /* => next byte to write in buffer */ + size_t free_in_buffer; /* # of byte spaces remaining in buffer */ + + void (*init_destination) (j_compress_ptr cinfo); + boolean (*empty_output_buffer) (j_compress_ptr cinfo); + void (*term_destination) (j_compress_ptr cinfo); +}; + + +/* Data source object for decompression */ + +struct jpeg_source_mgr { + const JOCTET *next_input_byte; /* => next byte to read from buffer */ + size_t bytes_in_buffer; /* # of bytes remaining in buffer */ + + void (*init_source) (j_decompress_ptr cinfo); + boolean (*fill_input_buffer) (j_decompress_ptr cinfo); + void (*skip_input_data) (j_decompress_ptr cinfo, long num_bytes); + boolean (*resync_to_restart) (j_decompress_ptr cinfo, int desired); + void (*term_source) (j_decompress_ptr cinfo); +}; + + +/* Memory manager object. + * Allocates "small" objects (a few K total), "large" objects (tens of K), + * and "really big" objects (virtual arrays with backing store if needed). + * The memory manager does not allow individual objects to be freed; rather, + * each created object is assigned to a pool, and whole pools can be freed + * at once. This is faster and more convenient than remembering exactly what + * to free, especially where malloc()/free() are not too speedy. + * NB: alloc routines never return NULL. They exit to error_exit if not + * successful. + */ + +#define JPOOL_PERMANENT 0 /* lasts until master record is destroyed */ +#define JPOOL_IMAGE 1 /* lasts until done with image/datastream */ +#define JPOOL_NUMPOOLS 2 + +typedef struct jvirt_sarray_control *jvirt_sarray_ptr; +typedef struct jvirt_barray_control *jvirt_barray_ptr; + + +struct jpeg_memory_mgr { + /* Method pointers */ + void *(*alloc_small) (j_common_ptr cinfo, int pool_id, size_t sizeofobject); + void *(*alloc_large) (j_common_ptr cinfo, int pool_id, + size_t sizeofobject); + /* If cinfo->data_precision is 12 or 16, then this method and the + * access_virt_sarray method actually return a J12SAMPARRAY or a + * J16SAMPARRAY, so callers must type-cast the return value in order to + * read/write 12-bit or 16-bit samples from/to the array. + */ + JSAMPARRAY (*alloc_sarray) (j_common_ptr cinfo, int pool_id, + JDIMENSION samplesperrow, JDIMENSION numrows); + JBLOCKARRAY (*alloc_barray) (j_common_ptr cinfo, int pool_id, + JDIMENSION blocksperrow, JDIMENSION numrows); + jvirt_sarray_ptr (*request_virt_sarray) (j_common_ptr cinfo, int pool_id, + boolean pre_zero, + JDIMENSION samplesperrow, + JDIMENSION numrows, + JDIMENSION maxaccess); + jvirt_barray_ptr (*request_virt_barray) (j_common_ptr cinfo, int pool_id, + boolean pre_zero, + JDIMENSION blocksperrow, + JDIMENSION numrows, + JDIMENSION maxaccess); + void (*realize_virt_arrays) (j_common_ptr cinfo); + JSAMPARRAY (*access_virt_sarray) (j_common_ptr cinfo, jvirt_sarray_ptr ptr, + JDIMENSION start_row, JDIMENSION num_rows, + boolean writable); + JBLOCKARRAY (*access_virt_barray) (j_common_ptr cinfo, jvirt_barray_ptr ptr, + JDIMENSION start_row, JDIMENSION num_rows, + boolean writable); + void (*free_pool) (j_common_ptr cinfo, int pool_id); + void (*self_destruct) (j_common_ptr cinfo); + + /* Limit on memory allocation for this JPEG object. (Note that this is + * merely advisory, not a guaranteed maximum; it only affects the space + * used for virtual-array buffers.) May be changed by outer application + * after creating the JPEG object. + */ + long max_memory_to_use; + + /* Maximum allocation request accepted by alloc_large. */ + long max_alloc_chunk; +}; + + +/* Routine signature for application-supplied marker processing methods. + * Need not pass marker code since it is stored in cinfo->unread_marker. + */ +typedef boolean (*jpeg_marker_parser_method) (j_decompress_ptr cinfo); + + +/* Originally, this macro was used as a way of defining function prototypes + * for both modern compilers as well as older compilers that did not support + * prototype parameters. libjpeg-turbo has never supported these older, + * non-ANSI compilers, but the macro is still included because there is some + * software out there that uses it. + */ + +#define JPP(arglist) arglist + + +/* Default error-management setup */ +EXTERN(struct jpeg_error_mgr *) jpeg_std_error(struct jpeg_error_mgr *err); + +/* Initialization of JPEG compression objects. + * jpeg_create_compress() and jpeg_create_decompress() are the exported + * names that applications should call. These expand to calls on + * jpeg_CreateCompress and jpeg_CreateDecompress with additional information + * passed for version mismatch checking. + * NB: you must set up the error-manager BEFORE calling jpeg_create_xxx. + */ +#define jpeg_create_compress(cinfo) \ + jpeg_CreateCompress((cinfo), JPEG_LIB_VERSION, \ + (size_t)sizeof(struct jpeg_compress_struct)) +#define jpeg_create_decompress(cinfo) \ + jpeg_CreateDecompress((cinfo), JPEG_LIB_VERSION, \ + (size_t)sizeof(struct jpeg_decompress_struct)) +EXTERN(void) jpeg_CreateCompress(j_compress_ptr cinfo, int version, + size_t structsize); +EXTERN(void) jpeg_CreateDecompress(j_decompress_ptr cinfo, int version, + size_t structsize); +/* Destruction of JPEG compression objects */ +EXTERN(void) jpeg_destroy_compress(j_compress_ptr cinfo); +EXTERN(void) jpeg_destroy_decompress(j_decompress_ptr cinfo); + +/* Standard data source and destination managers: stdio streams. */ +/* Caller is responsible for opening the file before and closing after. */ +EXTERN(void) jpeg_stdio_dest(j_compress_ptr cinfo, FILE *outfile); +EXTERN(void) jpeg_stdio_src(j_decompress_ptr cinfo, FILE *infile); + +/* Data source and destination managers: memory buffers. */ +EXTERN(void) jpeg_mem_dest(j_compress_ptr cinfo, unsigned char **outbuffer, + unsigned long *outsize); +EXTERN(void) jpeg_mem_src(j_decompress_ptr cinfo, + const unsigned char *inbuffer, unsigned long insize); + +/* Default parameter setup for compression */ +EXTERN(void) jpeg_set_defaults(j_compress_ptr cinfo); +/* Compression parameter setup aids */ +EXTERN(void) jpeg_set_colorspace(j_compress_ptr cinfo, + J_COLOR_SPACE colorspace); +EXTERN(void) jpeg_default_colorspace(j_compress_ptr cinfo); +EXTERN(void) jpeg_set_quality(j_compress_ptr cinfo, int quality, + boolean force_baseline); +EXTERN(void) jpeg_set_linear_quality(j_compress_ptr cinfo, int scale_factor, + boolean force_baseline); +#if JPEG_LIB_VERSION >= 70 +EXTERN(void) jpeg_default_qtables(j_compress_ptr cinfo, + boolean force_baseline); +#endif +EXTERN(void) jpeg_add_quant_table(j_compress_ptr cinfo, int which_tbl, + const unsigned int *basic_table, + int scale_factor, boolean force_baseline); +EXTERN(int) jpeg_quality_scaling(int quality); +EXTERN(void) jpeg_enable_lossless(j_compress_ptr cinfo, + int predictor_selection_value, + int point_transform); +EXTERN(void) jpeg_simple_progression(j_compress_ptr cinfo); +EXTERN(void) jpeg_suppress_tables(j_compress_ptr cinfo, boolean suppress); +EXTERN(JQUANT_TBL *) jpeg_alloc_quant_table(j_common_ptr cinfo); +EXTERN(JHUFF_TBL *) jpeg_alloc_huff_table(j_common_ptr cinfo); + +/* Main entry points for compression */ +EXTERN(void) jpeg_start_compress(j_compress_ptr cinfo, + boolean write_all_tables); +EXTERN(JDIMENSION) jpeg_write_scanlines(j_compress_ptr cinfo, + JSAMPARRAY scanlines, + JDIMENSION num_lines); +EXTERN(JDIMENSION) jpeg12_write_scanlines(j_compress_ptr cinfo, + J12SAMPARRAY scanlines, + JDIMENSION num_lines); +EXTERN(JDIMENSION) jpeg16_write_scanlines(j_compress_ptr cinfo, + J16SAMPARRAY scanlines, + JDIMENSION num_lines); +EXTERN(void) jpeg_finish_compress(j_compress_ptr cinfo); + +#if JPEG_LIB_VERSION >= 70 +/* Precalculate JPEG dimensions for current compression parameters. */ +EXTERN(void) jpeg_calc_jpeg_dimensions(j_compress_ptr cinfo); +#endif + +/* Replaces jpeg_write_scanlines when writing raw downsampled data. */ +EXTERN(JDIMENSION) jpeg_write_raw_data(j_compress_ptr cinfo, JSAMPIMAGE data, + JDIMENSION num_lines); +EXTERN(JDIMENSION) jpeg12_write_raw_data(j_compress_ptr cinfo, + J12SAMPIMAGE data, + JDIMENSION num_lines); + +/* Write a special marker. See libjpeg.txt concerning safe usage. */ +EXTERN(void) jpeg_write_marker(j_compress_ptr cinfo, int marker, + const JOCTET *dataptr, unsigned int datalen); +/* Same, but piecemeal. */ +EXTERN(void) jpeg_write_m_header(j_compress_ptr cinfo, int marker, + unsigned int datalen); +EXTERN(void) jpeg_write_m_byte(j_compress_ptr cinfo, int val); + +/* Alternate compression function: just write an abbreviated table file */ +EXTERN(void) jpeg_write_tables(j_compress_ptr cinfo); + +/* Write ICC profile. See libjpeg.txt for usage information. */ +EXTERN(void) jpeg_write_icc_profile(j_compress_ptr cinfo, + const JOCTET *icc_data_ptr, + unsigned int icc_data_len); + + +/* Decompression startup: read start of JPEG datastream to see what's there */ +EXTERN(int) jpeg_read_header(j_decompress_ptr cinfo, boolean require_image); +/* Return value is one of: */ +#define JPEG_SUSPENDED 0 /* Suspended due to lack of input data */ +#define JPEG_HEADER_OK 1 /* Found valid image datastream */ +#define JPEG_HEADER_TABLES_ONLY 2 /* Found valid table-specs-only datastream */ +/* If you pass require_image = TRUE (normal case), you need not check for + * a TABLES_ONLY return code; an abbreviated file will cause an error exit. + * JPEG_SUSPENDED is only possible if you use a data source module that can + * give a suspension return (the stdio source module doesn't). + */ + +/* Main entry points for decompression */ +EXTERN(boolean) jpeg_start_decompress(j_decompress_ptr cinfo); +EXTERN(JDIMENSION) jpeg_read_scanlines(j_decompress_ptr cinfo, + JSAMPARRAY scanlines, + JDIMENSION max_lines); +EXTERN(JDIMENSION) jpeg12_read_scanlines(j_decompress_ptr cinfo, + J12SAMPARRAY scanlines, + JDIMENSION max_lines); +EXTERN(JDIMENSION) jpeg16_read_scanlines(j_decompress_ptr cinfo, + J16SAMPARRAY scanlines, + JDIMENSION max_lines); +EXTERN(JDIMENSION) jpeg_skip_scanlines(j_decompress_ptr cinfo, + JDIMENSION num_lines); +EXTERN(JDIMENSION) jpeg12_skip_scanlines(j_decompress_ptr cinfo, + JDIMENSION num_lines); +EXTERN(void) jpeg_crop_scanline(j_decompress_ptr cinfo, JDIMENSION *xoffset, + JDIMENSION *width); +EXTERN(void) jpeg12_crop_scanline(j_decompress_ptr cinfo, JDIMENSION *xoffset, + JDIMENSION *width); +EXTERN(boolean) jpeg_finish_decompress(j_decompress_ptr cinfo); + +/* Replaces jpeg_read_scanlines when reading raw downsampled data. */ +EXTERN(JDIMENSION) jpeg_read_raw_data(j_decompress_ptr cinfo, JSAMPIMAGE data, + JDIMENSION max_lines); +EXTERN(JDIMENSION) jpeg12_read_raw_data(j_decompress_ptr cinfo, + J12SAMPIMAGE data, + JDIMENSION max_lines); + +/* Additional entry points for buffered-image mode. */ +EXTERN(boolean) jpeg_has_multiple_scans(j_decompress_ptr cinfo); +EXTERN(boolean) jpeg_start_output(j_decompress_ptr cinfo, int scan_number); +EXTERN(boolean) jpeg_finish_output(j_decompress_ptr cinfo); +EXTERN(boolean) jpeg_input_complete(j_decompress_ptr cinfo); +EXTERN(void) jpeg_new_colormap(j_decompress_ptr cinfo); +EXTERN(int) jpeg_consume_input(j_decompress_ptr cinfo); +/* Return value is one of: */ +/* #define JPEG_SUSPENDED 0 Suspended due to lack of input data */ +#define JPEG_REACHED_SOS 1 /* Reached start of new scan */ +#define JPEG_REACHED_EOI 2 /* Reached end of image */ +#define JPEG_ROW_COMPLETED 3 /* Completed one iMCU row */ +#define JPEG_SCAN_COMPLETED 4 /* Completed last iMCU row of a scan */ + +/* Precalculate output dimensions for current decompression parameters. */ +#if JPEG_LIB_VERSION >= 80 +EXTERN(void) jpeg_core_output_dimensions(j_decompress_ptr cinfo); +#endif +EXTERN(void) jpeg_calc_output_dimensions(j_decompress_ptr cinfo); + +/* Control saving of COM and APPn markers into marker_list. */ +EXTERN(void) jpeg_save_markers(j_decompress_ptr cinfo, int marker_code, + unsigned int length_limit); + +/* Install a special processing method for COM or APPn markers. */ +EXTERN(void) jpeg_set_marker_processor(j_decompress_ptr cinfo, + int marker_code, + jpeg_marker_parser_method routine); + +/* Read or write raw DCT coefficients --- useful for lossless transcoding. */ +EXTERN(jvirt_barray_ptr *) jpeg_read_coefficients(j_decompress_ptr cinfo); +EXTERN(void) jpeg_write_coefficients(j_compress_ptr cinfo, + jvirt_barray_ptr *coef_arrays); +EXTERN(void) jpeg_copy_critical_parameters(j_decompress_ptr srcinfo, + j_compress_ptr dstinfo); + +/* If you choose to abort compression or decompression before completing + * jpeg_finish_(de)compress, then you need to clean up to release memory, + * temporary files, etc. You can just call jpeg_destroy_(de)compress + * if you're done with the JPEG object, but if you want to clean it up and + * reuse it, call this: + */ +EXTERN(void) jpeg_abort_compress(j_compress_ptr cinfo); +EXTERN(void) jpeg_abort_decompress(j_decompress_ptr cinfo); + +/* Generic versions of jpeg_abort and jpeg_destroy that work on either + * flavor of JPEG object. These may be more convenient in some places. + */ +EXTERN(void) jpeg_abort(j_common_ptr cinfo); +EXTERN(void) jpeg_destroy(j_common_ptr cinfo); + +/* Default restart-marker-resync procedure for use by data source modules */ +EXTERN(boolean) jpeg_resync_to_restart(j_decompress_ptr cinfo, int desired); + +/* Read ICC profile. See libjpeg.txt for usage information. */ +EXTERN(boolean) jpeg_read_icc_profile(j_decompress_ptr cinfo, + JOCTET **icc_data_ptr, + unsigned int *icc_data_len); + + +/* These marker codes are exported since applications and data source modules + * are likely to want to use them. + */ + +#define JPEG_RST0 0xD0 /* RST0 marker code */ +#define JPEG_EOI 0xD9 /* EOI marker code */ +#define JPEG_APP0 0xE0 /* APP0 marker code */ +#define JPEG_COM 0xFE /* COM marker code */ + + +/* If we have a brain-damaged compiler that emits warnings (or worse, errors) + * for structure definitions that are never filled in, keep it quiet by + * supplying dummy definitions for the various substructures. + */ + +#ifdef INCOMPLETE_TYPES_BROKEN +#ifndef JPEG_INTERNALS /* will be defined in jpegint.h */ +struct jvirt_sarray_control { long dummy; }; +struct jvirt_barray_control { long dummy; }; +struct jpeg_comp_master { long dummy; }; +struct jpeg_c_main_controller { long dummy; }; +struct jpeg_c_prep_controller { long dummy; }; +struct jpeg_c_coef_controller { long dummy; }; +struct jpeg_marker_writer { long dummy; }; +struct jpeg_color_converter { long dummy; }; +struct jpeg_downsampler { long dummy; }; +struct jpeg_forward_dct { long dummy; }; +struct jpeg_entropy_encoder { long dummy; }; +struct jpeg_decomp_master { long dummy; }; +struct jpeg_d_main_controller { long dummy; }; +struct jpeg_d_coef_controller { long dummy; }; +struct jpeg_d_post_controller { long dummy; }; +struct jpeg_input_controller { long dummy; }; +struct jpeg_marker_reader { long dummy; }; +struct jpeg_entropy_decoder { long dummy; }; +struct jpeg_inverse_dct { long dummy; }; +struct jpeg_upsampler { long dummy; }; +struct jpeg_color_deconverter { long dummy; }; +struct jpeg_color_quantizer { long dummy; }; +#endif /* JPEG_INTERNALS */ +#endif /* INCOMPLETE_TYPES_BROKEN */ + + +/* + * The JPEG library modules define JPEG_INTERNALS before including this file. + * The internal structure declarations are read only when that is true. + * Applications using the library should not include jpegint.h, but may wish + * to include jerror.h. + */ + +#ifdef JPEG_INTERNALS +#include "jpegint.h" /* fetch private declarations */ +#include "jerror.h" /* fetch error codes too */ +#endif + +#ifdef __cplusplus +#ifndef DONT_USE_EXTERN_C +} +#endif +#endif + +#endif /* JPEGLIB_H */ diff --git a/src/spice2x/external/libjpeg-turbo/jpegtran.c b/src/spice2x/external/libjpeg-turbo/jpegtran.c new file mode 100644 index 0000000..b7d7ebb --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jpegtran.c @@ -0,0 +1,729 @@ +/* + * jpegtran.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1995-2019, Thomas G. Lane, Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2010, 2014, 2017, 2019-2022, 2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains a command-line user interface for JPEG transcoding. + * It is very similar to cjpeg.c, and partly to djpeg.c, but provides + * lossless transcoding between different JPEG file formats. It also + * provides some lossless and sort-of-lossless transformations of JPEG data. + */ + +#ifdef _MSC_VER +#define _CRT_SECURE_NO_DEPRECATE +#endif + +#include "cdjpeg.h" /* Common decls for cjpeg/djpeg applications */ +#include "transupp.h" /* Support routines for jpegtran */ +#include "jversion.h" /* for version message */ +#include "jconfigint.h" + + +/* + * Argument-parsing code. + * The switch parser is designed to be useful with DOS-style command line + * syntax, ie, intermixed switches and file names, where only the switches + * to the left of a given file name affect processing of that file. + * The main program in this file doesn't actually use this capability... + */ + + +static const char *progname; /* program name for error messages */ +static char *icc_filename; /* for -icc switch */ +static JDIMENSION max_scans; /* for -maxscans switch */ +static char *outfilename; /* for -outfile switch */ +static char *dropfilename; /* for -drop switch */ +static boolean report; /* for -report switch */ +static boolean strict; /* for -strict switch */ +static JCOPY_OPTION copyoption; /* -copy switch */ +static jpeg_transform_info transformoption; /* image transformation options */ + + +LOCAL(void) +usage(void) +/* complain about bad command line */ +{ + fprintf(stderr, "usage: %s [switches] ", progname); +#ifdef TWO_FILE_COMMANDLINE + fprintf(stderr, "inputfile outputfile\n"); +#else + fprintf(stderr, "[inputfile]\n"); +#endif + + fprintf(stderr, "Switches (names may be abbreviated):\n"); + fprintf(stderr, " -copy none Copy no extra markers from source file\n"); + fprintf(stderr, " -copy comments Copy only comment markers (default)\n"); + fprintf(stderr, " -copy icc Copy only ICC profile markers\n"); + fprintf(stderr, " -copy all Copy all extra markers\n"); +#ifdef ENTROPY_OPT_SUPPORTED + fprintf(stderr, " -optimize Optimize Huffman table (smaller file, but slow compression)\n"); +#endif +#ifdef C_PROGRESSIVE_SUPPORTED + fprintf(stderr, " -progressive Create progressive JPEG file\n"); +#endif + fprintf(stderr, "Switches for modifying the image:\n"); +#if TRANSFORMS_SUPPORTED + fprintf(stderr, " -crop WxH+X+Y Crop to a rectangular region\n"); + fprintf(stderr, " -drop +X+Y filename Drop (insert) another image\n"); + fprintf(stderr, " -flip [horizontal|vertical] Mirror image (left-right or top-bottom)\n"); + fprintf(stderr, " -grayscale Reduce to grayscale (omit color data)\n"); + fprintf(stderr, " -perfect Fail if there is non-transformable edge blocks\n"); + fprintf(stderr, " -rotate [90|180|270] Rotate image (degrees clockwise)\n"); +#endif +#if TRANSFORMS_SUPPORTED + fprintf(stderr, " -transpose Transpose image\n"); + fprintf(stderr, " -transverse Transverse transpose image\n"); + fprintf(stderr, " -trim Drop non-transformable edge blocks\n"); + fprintf(stderr, " with -drop: Requantize drop file to match source file\n"); + fprintf(stderr, " -wipe WxH+X+Y Wipe (gray out) a rectangular region\n"); +#endif + fprintf(stderr, "Switches for advanced users:\n"); +#ifdef C_ARITH_CODING_SUPPORTED + fprintf(stderr, " -arithmetic Use arithmetic coding\n"); +#endif + fprintf(stderr, " -icc FILE Embed ICC profile contained in FILE\n"); + fprintf(stderr, " -restart N Set restart interval in rows, or in blocks with B\n"); + fprintf(stderr, " -maxmemory N Maximum memory to use (in kbytes)\n"); + fprintf(stderr, " -maxscans N Maximum number of scans to allow in input file\n"); + fprintf(stderr, " -outfile name Specify name for output file\n"); + fprintf(stderr, " -report Report transformation progress\n"); + fprintf(stderr, " -strict Treat all warnings as fatal\n"); + fprintf(stderr, " -verbose or -debug Emit debug output\n"); + fprintf(stderr, " -version Print version information and exit\n"); + fprintf(stderr, "Switches for wizards:\n"); +#ifdef C_MULTISCAN_FILES_SUPPORTED + fprintf(stderr, " -scans FILE Create multi-scan JPEG per script FILE\n"); +#endif + exit(EXIT_FAILURE); +} + + +LOCAL(void) +select_transform(JXFORM_CODE transform) +/* Silly little routine to detect multiple transform options, + * which we can't handle. + */ +{ +#if TRANSFORMS_SUPPORTED + if (transformoption.transform == JXFORM_NONE || + transformoption.transform == transform) { + transformoption.transform = transform; + } else { + fprintf(stderr, "%s: can only do one image transformation at a time\n", + progname); + usage(); + } +#else + fprintf(stderr, "%s: sorry, image transformation was not compiled\n", + progname); + exit(EXIT_FAILURE); +#endif +} + + +LOCAL(int) +parse_switches(j_compress_ptr cinfo, int argc, char **argv, + int last_file_arg_seen, boolean for_real) +/* Parse optional switches. + * Returns argv[] index of first file-name argument (== argc if none). + * Any file names with indexes <= last_file_arg_seen are ignored; + * they have presumably been processed in a previous iteration. + * (Pass 0 for last_file_arg_seen on the first or only iteration.) + * for_real is FALSE on the first (dummy) pass; we may skip any expensive + * processing. + */ +{ + int argn; + char *arg; + boolean simple_progressive; + char *scansarg = NULL; /* saves -scans parm if any */ + + /* Set up default JPEG parameters. */ + simple_progressive = FALSE; + icc_filename = NULL; + max_scans = 0; + outfilename = NULL; + report = FALSE; + strict = FALSE; + copyoption = JCOPYOPT_DEFAULT; + transformoption.transform = JXFORM_NONE; + transformoption.perfect = FALSE; + transformoption.trim = FALSE; + transformoption.force_grayscale = FALSE; + transformoption.crop = FALSE; + transformoption.slow_hflip = FALSE; + cinfo->err->trace_level = 0; + + /* Scan command line options, adjust parameters */ + + for (argn = 1; argn < argc; argn++) { + arg = argv[argn]; + if (*arg != '-') { + /* Not a switch, must be a file name argument */ + if (argn <= last_file_arg_seen) { + outfilename = NULL; /* -outfile applies to just one input file */ + continue; /* ignore this name if previously processed */ + } + break; /* else done parsing switches */ + } + arg++; /* advance past switch marker character */ + + if (keymatch(arg, "arithmetic", 1)) { + /* Use arithmetic coding. */ +#ifdef C_ARITH_CODING_SUPPORTED + cinfo->arith_code = TRUE; +#else + fprintf(stderr, "%s: sorry, arithmetic coding not supported\n", + progname); + exit(EXIT_FAILURE); +#endif + + } else if (keymatch(arg, "copy", 1)) { + /* Select which extra markers to copy. */ + if (++argn >= argc) /* advance to next argument */ + usage(); + if (keymatch(argv[argn], "none", 1)) { + copyoption = JCOPYOPT_NONE; + } else if (keymatch(argv[argn], "comments", 1)) { + copyoption = JCOPYOPT_COMMENTS; + } else if (keymatch(argv[argn], "icc", 1)) { + copyoption = JCOPYOPT_ICC; + } else if (keymatch(argv[argn], "all", 1)) { + copyoption = JCOPYOPT_ALL; + } else + usage(); + + } else if (keymatch(arg, "crop", 2)) { + /* Perform lossless cropping. */ +#if TRANSFORMS_SUPPORTED + if (++argn >= argc) /* advance to next argument */ + usage(); + if (transformoption.crop /* reject multiple crop/drop/wipe requests */ || + !jtransform_parse_crop_spec(&transformoption, argv[argn])) { + fprintf(stderr, "%s: bogus -crop argument '%s'\n", + progname, argv[argn]); + exit(EXIT_FAILURE); + } +#else + select_transform(JXFORM_NONE); /* force an error */ +#endif + + } else if (keymatch(arg, "drop", 2)) { +#if TRANSFORMS_SUPPORTED + if (++argn >= argc) /* advance to next argument */ + usage(); + if (transformoption.crop /* reject multiple crop/drop/wipe requests */ || + !jtransform_parse_crop_spec(&transformoption, argv[argn]) || + transformoption.crop_width_set != JCROP_UNSET || + transformoption.crop_height_set != JCROP_UNSET) { + fprintf(stderr, "%s: bogus -drop argument '%s'\n", + progname, argv[argn]); + exit(EXIT_FAILURE); + } + if (++argn >= argc) /* advance to next argument */ + usage(); + dropfilename = argv[argn]; + select_transform(JXFORM_DROP); +#else + select_transform(JXFORM_NONE); /* force an error */ +#endif + + } else if (keymatch(arg, "debug", 1) || keymatch(arg, "verbose", 1)) { + /* Enable debug printouts. */ + /* On first -d, print version identification */ + static boolean printed_version = FALSE; + + if (!printed_version) { + fprintf(stderr, "%s version %s (build %s)\n", + PACKAGE_NAME, VERSION, BUILD); + fprintf(stderr, JCOPYRIGHT1); + fprintf(stderr, JCOPYRIGHT2 "\n"); + fprintf(stderr, "Emulating The Independent JPEG Group's software, version %s\n\n", + JVERSION); + printed_version = TRUE; + } + cinfo->err->trace_level++; + + } else if (keymatch(arg, "version", 4)) { + fprintf(stderr, "%s version %s (build %s)\n", + PACKAGE_NAME, VERSION, BUILD); + exit(EXIT_SUCCESS); + + } else if (keymatch(arg, "flip", 1)) { + /* Mirror left-right or top-bottom. */ + if (++argn >= argc) /* advance to next argument */ + usage(); + if (keymatch(argv[argn], "horizontal", 1)) + select_transform(JXFORM_FLIP_H); + else if (keymatch(argv[argn], "vertical", 1)) + select_transform(JXFORM_FLIP_V); + else + usage(); + + } else if (keymatch(arg, "grayscale", 1) || + keymatch(arg, "greyscale", 1)) { + /* Force to grayscale. */ +#if TRANSFORMS_SUPPORTED + transformoption.force_grayscale = TRUE; +#else + select_transform(JXFORM_NONE); /* force an error */ +#endif + + } else if (keymatch(arg, "icc", 1)) { + /* Set ICC filename. */ + if (++argn >= argc) /* advance to next argument */ + usage(); + icc_filename = argv[argn]; + + } else if (keymatch(arg, "maxmemory", 3)) { + /* Maximum memory in Kb (or Mb with 'm'). */ + long lval; + char ch = 'x'; + + if (++argn >= argc) /* advance to next argument */ + usage(); + if (sscanf(argv[argn], "%ld%c", &lval, &ch) < 1) + usage(); + if (ch == 'm' || ch == 'M') + lval *= 1000L; + cinfo->mem->max_memory_to_use = lval * 1000L; + + } else if (keymatch(arg, "maxscans", 4)) { + if (++argn >= argc) /* advance to next argument */ + usage(); + if (sscanf(argv[argn], "%u", &max_scans) != 1) + usage(); + + } else if (keymatch(arg, "optimize", 1) || keymatch(arg, "optimise", 1)) { + /* Enable entropy parm optimization. */ +#ifdef ENTROPY_OPT_SUPPORTED + cinfo->optimize_coding = TRUE; +#else + fprintf(stderr, "%s: sorry, entropy optimization was not compiled\n", + progname); + exit(EXIT_FAILURE); +#endif + + } else if (keymatch(arg, "outfile", 4)) { + /* Set output file name. */ + if (++argn >= argc) /* advance to next argument */ + usage(); + outfilename = argv[argn]; /* save it away for later use */ + + } else if (keymatch(arg, "perfect", 2)) { + /* Fail if there is any partial edge MCUs that the transform can't + * handle. */ + transformoption.perfect = TRUE; + + } else if (keymatch(arg, "progressive", 1)) { + /* Select simple progressive mode. */ +#ifdef C_PROGRESSIVE_SUPPORTED + simple_progressive = TRUE; + /* We must postpone execution until num_components is known. */ +#else + fprintf(stderr, "%s: sorry, progressive output was not compiled\n", + progname); + exit(EXIT_FAILURE); +#endif + + } else if (keymatch(arg, "report", 3)) { + report = TRUE; + + } else if (keymatch(arg, "restart", 1)) { + /* Restart interval in MCU rows (or in MCUs with 'b'). */ + long lval; + char ch = 'x'; + + if (++argn >= argc) /* advance to next argument */ + usage(); + if (sscanf(argv[argn], "%ld%c", &lval, &ch) < 1) + usage(); + if (lval < 0 || lval > 65535L) + usage(); + if (ch == 'b' || ch == 'B') { + cinfo->restart_interval = (unsigned int)lval; + cinfo->restart_in_rows = 0; /* else prior '-restart n' overrides me */ + } else { + cinfo->restart_in_rows = (int)lval; + /* restart_interval will be computed during startup */ + } + + } else if (keymatch(arg, "rotate", 2)) { + /* Rotate 90, 180, or 270 degrees (measured clockwise). */ + if (++argn >= argc) /* advance to next argument */ + usage(); + if (keymatch(argv[argn], "90", 2)) + select_transform(JXFORM_ROT_90); + else if (keymatch(argv[argn], "180", 3)) + select_transform(JXFORM_ROT_180); + else if (keymatch(argv[argn], "270", 3)) + select_transform(JXFORM_ROT_270); + else + usage(); + + } else if (keymatch(arg, "scans", 1)) { + /* Set scan script. */ +#ifdef C_MULTISCAN_FILES_SUPPORTED + if (++argn >= argc) /* advance to next argument */ + usage(); + scansarg = argv[argn]; + /* We must postpone reading the file in case -progressive appears. */ +#else + fprintf(stderr, "%s: sorry, multi-scan output was not compiled\n", + progname); + exit(EXIT_FAILURE); +#endif + + } else if (keymatch(arg, "strict", 2)) { + strict = TRUE; + + } else if (keymatch(arg, "transpose", 1)) { + /* Transpose (across UL-to-LR axis). */ + select_transform(JXFORM_TRANSPOSE); + + } else if (keymatch(arg, "transverse", 6)) { + /* Transverse transpose (across UR-to-LL axis). */ + select_transform(JXFORM_TRANSVERSE); + + } else if (keymatch(arg, "trim", 3)) { + /* Trim off any partial edge MCUs that the transform can't handle. */ + transformoption.trim = TRUE; + + } else if (keymatch(arg, "wipe", 1)) { +#if TRANSFORMS_SUPPORTED + if (++argn >= argc) /* advance to next argument */ + usage(); + if (transformoption.crop /* reject multiple crop/drop/wipe requests */ || + !jtransform_parse_crop_spec(&transformoption, argv[argn])) { + fprintf(stderr, "%s: bogus -wipe argument '%s'\n", + progname, argv[argn]); + exit(EXIT_FAILURE); + } + select_transform(JXFORM_WIPE); +#else + select_transform(JXFORM_NONE); /* force an error */ +#endif + + } else { + usage(); /* bogus switch */ + } + } + + /* Post-switch-scanning cleanup */ + + if (for_real) { + +#ifdef C_PROGRESSIVE_SUPPORTED + if (simple_progressive) /* process -progressive; -scans can override */ + jpeg_simple_progression(cinfo); +#endif + +#ifdef C_MULTISCAN_FILES_SUPPORTED + if (scansarg != NULL) /* process -scans if it was present */ + if (!read_scan_script(cinfo, scansarg)) + usage(); +#endif + } + + return argn; /* return index of next arg (file name) */ +} + + +METHODDEF(void) +my_emit_message(j_common_ptr cinfo, int msg_level) +{ + if (msg_level < 0) { + /* Treat warning as fatal */ + cinfo->err->error_exit(cinfo); + } else { + if (cinfo->err->trace_level >= msg_level) + cinfo->err->output_message(cinfo); + } +} + + +/* + * The main program. + */ + +int +main(int argc, char **argv) +{ + struct jpeg_decompress_struct srcinfo; +#if TRANSFORMS_SUPPORTED + struct jpeg_decompress_struct dropinfo; + struct jpeg_error_mgr jdroperr; + FILE *drop_file; +#endif + struct jpeg_compress_struct dstinfo; + struct jpeg_error_mgr jsrcerr, jdsterr; + struct cdjpeg_progress_mgr src_progress, dst_progress; + jvirt_barray_ptr *src_coef_arrays; + jvirt_barray_ptr *dst_coef_arrays; + int file_index; + /* We assume all-in-memory processing and can therefore use only a + * single file pointer for sequential input and output operation. + */ + FILE *fp; + FILE *icc_file; + JOCTET *icc_profile = NULL; + long icc_len = 0; + + progname = argv[0]; + if (progname == NULL || progname[0] == 0) + progname = "jpegtran"; /* in case C library doesn't provide it */ + + /* Initialize the JPEG decompression object with default error handling. */ + srcinfo.err = jpeg_std_error(&jsrcerr); + jpeg_create_decompress(&srcinfo); + /* Initialize the JPEG compression object with default error handling. */ + dstinfo.err = jpeg_std_error(&jdsterr); + jpeg_create_compress(&dstinfo); + + /* Scan command line to find file names. + * It is convenient to use just one switch-parsing routine, but the switch + * values read here are mostly ignored; we will rescan the switches after + * opening the input file. Also note that most of the switches affect the + * destination JPEG object, so we parse into that and then copy over what + * needs to affect the source too. + */ + + file_index = parse_switches(&dstinfo, argc, argv, 0, FALSE); + jsrcerr.trace_level = jdsterr.trace_level; + srcinfo.mem->max_memory_to_use = dstinfo.mem->max_memory_to_use; + + if (strict) + jsrcerr.emit_message = my_emit_message; + +#ifdef TWO_FILE_COMMANDLINE + /* Must have either -outfile switch or explicit output file name */ + if (outfilename == NULL) { + if (file_index != argc - 2) { + fprintf(stderr, "%s: must name one input and one output file\n", + progname); + usage(); + } + outfilename = argv[file_index + 1]; + } else { + if (file_index != argc - 1) { + fprintf(stderr, "%s: must name one input and one output file\n", + progname); + usage(); + } + } +#else + /* Unix style: expect zero or one file name */ + if (file_index < argc - 1) { + fprintf(stderr, "%s: only one input file\n", progname); + usage(); + } +#endif /* TWO_FILE_COMMANDLINE */ + + /* Open the input file. */ + if (file_index < argc) { + if ((fp = fopen(argv[file_index], READ_BINARY)) == NULL) { + fprintf(stderr, "%s: can't open %s for reading\n", progname, + argv[file_index]); + exit(EXIT_FAILURE); + } + } else { + /* default input file is stdin */ + fp = read_stdin(); + } + + if (icc_filename != NULL) { + if ((icc_file = fopen(icc_filename, READ_BINARY)) == NULL) { + fprintf(stderr, "%s: can't open %s\n", progname, icc_filename); + exit(EXIT_FAILURE); + } + if (fseek(icc_file, 0, SEEK_END) < 0 || + (icc_len = ftell(icc_file)) < 1 || + fseek(icc_file, 0, SEEK_SET) < 0) { + fprintf(stderr, "%s: can't determine size of %s\n", progname, + icc_filename); + exit(EXIT_FAILURE); + } + if ((icc_profile = (JOCTET *)malloc(icc_len)) == NULL) { + fprintf(stderr, "%s: can't allocate memory for ICC profile\n", progname); + fclose(icc_file); + exit(EXIT_FAILURE); + } + if (fread(icc_profile, icc_len, 1, icc_file) < 1) { + fprintf(stderr, "%s: can't read ICC profile from %s\n", progname, + icc_filename); + free(icc_profile); + fclose(icc_file); + exit(EXIT_FAILURE); + } + fclose(icc_file); + if (copyoption == JCOPYOPT_ALL) + copyoption = JCOPYOPT_ALL_EXCEPT_ICC; + if (copyoption == JCOPYOPT_ICC) + copyoption = JCOPYOPT_NONE; + } + + if (report) { + start_progress_monitor((j_common_ptr)&dstinfo, &dst_progress); + dst_progress.report = report; + } + if (report || max_scans != 0) { + start_progress_monitor((j_common_ptr)&srcinfo, &src_progress); + src_progress.report = report; + src_progress.max_scans = max_scans; + } +#if TRANSFORMS_SUPPORTED + /* Open the drop file. */ + if (dropfilename != NULL) { + if ((drop_file = fopen(dropfilename, READ_BINARY)) == NULL) { + fprintf(stderr, "%s: can't open %s for reading\n", progname, + dropfilename); + exit(EXIT_FAILURE); + } + dropinfo.err = jpeg_std_error(&jdroperr); + jpeg_create_decompress(&dropinfo); + jpeg_stdio_src(&dropinfo, drop_file); + } else { + drop_file = NULL; + } +#endif + + /* Specify data source for decompression */ + jpeg_stdio_src(&srcinfo, fp); + + /* Enable saving of extra markers that we want to copy */ + jcopy_markers_setup(&srcinfo, copyoption); + + /* Read file header */ + (void)jpeg_read_header(&srcinfo, TRUE); + +#if TRANSFORMS_SUPPORTED + if (dropfilename != NULL) { + (void)jpeg_read_header(&dropinfo, TRUE); + transformoption.crop_width = dropinfo.image_width; + transformoption.crop_width_set = JCROP_POS; + transformoption.crop_height = dropinfo.image_height; + transformoption.crop_height_set = JCROP_POS; + transformoption.drop_ptr = &dropinfo; + } +#endif + + /* Any space needed by a transform option must be requested before + * jpeg_read_coefficients so that memory allocation will be done right. + */ +#if TRANSFORMS_SUPPORTED + /* Fail right away if -perfect is given and transformation is not perfect. + */ + if (!jtransform_request_workspace(&srcinfo, &transformoption)) { + fprintf(stderr, "%s: transformation is not perfect\n", progname); + exit(EXIT_FAILURE); + } +#endif + + /* Read source file as DCT coefficients */ + src_coef_arrays = jpeg_read_coefficients(&srcinfo); + +#if TRANSFORMS_SUPPORTED + if (dropfilename != NULL) { + transformoption.drop_coef_arrays = jpeg_read_coefficients(&dropinfo); + } +#endif + + /* Initialize destination compression parameters from source values */ + jpeg_copy_critical_parameters(&srcinfo, &dstinfo); + + /* Adjust destination parameters if required by transform options; + * also find out which set of coefficient arrays will hold the output. + */ +#if TRANSFORMS_SUPPORTED + dst_coef_arrays = jtransform_adjust_parameters(&srcinfo, &dstinfo, + src_coef_arrays, + &transformoption); +#else + dst_coef_arrays = src_coef_arrays; +#endif + + /* Close input file, if we opened it. + * Note: we assume that jpeg_read_coefficients consumed all input + * until JPEG_REACHED_EOI, and that jpeg_finish_decompress will + * only consume more while (!cinfo->inputctl->eoi_reached). + * We cannot call jpeg_finish_decompress here since we still need the + * virtual arrays allocated from the source object for processing. + */ + if (fp != stdin) + fclose(fp); + + /* Open the output file. */ + if (outfilename != NULL) { + if ((fp = fopen(outfilename, WRITE_BINARY)) == NULL) { + fprintf(stderr, "%s: can't open %s for writing\n", progname, + outfilename); + exit(EXIT_FAILURE); + } + } else { + /* default output file is stdout */ + fp = write_stdout(); + } + + /* Adjust default compression parameters by re-parsing the options */ + file_index = parse_switches(&dstinfo, argc, argv, 0, TRUE); + + /* Specify data destination for compression */ + jpeg_stdio_dest(&dstinfo, fp); + + /* Start compressor (note no image data is actually written here) */ + jpeg_write_coefficients(&dstinfo, dst_coef_arrays); + + /* Copy to the output file any extra markers that we want to preserve */ + jcopy_markers_execute(&srcinfo, &dstinfo, copyoption); + + if (icc_profile != NULL) + jpeg_write_icc_profile(&dstinfo, icc_profile, (unsigned int)icc_len); + + /* Execute image transformation, if any */ +#if TRANSFORMS_SUPPORTED + jtransform_execute_transformation(&srcinfo, &dstinfo, src_coef_arrays, + &transformoption); +#endif + + /* Finish compression and release memory */ + jpeg_finish_compress(&dstinfo); + jpeg_destroy_compress(&dstinfo); +#if TRANSFORMS_SUPPORTED + if (dropfilename != NULL) { + (void)jpeg_finish_decompress(&dropinfo); + jpeg_destroy_decompress(&dropinfo); + } +#endif + (void)jpeg_finish_decompress(&srcinfo); + jpeg_destroy_decompress(&srcinfo); + + /* Close output file, if we opened it */ + if (fp != stdout) + fclose(fp); +#if TRANSFORMS_SUPPORTED + if (drop_file != NULL) + fclose(drop_file); +#endif + + if (report) + end_progress_monitor((j_common_ptr)&dstinfo); + if (report || max_scans != 0) + end_progress_monitor((j_common_ptr)&srcinfo); + + free(icc_profile); + + /* All done. */ +#if TRANSFORMS_SUPPORTED + if (dropfilename != NULL) + exit(jsrcerr.num_warnings + jdroperr.num_warnings + + jdsterr.num_warnings ? EXIT_WARNING : EXIT_SUCCESS); +#endif + exit(jsrcerr.num_warnings + jdsterr.num_warnings ? + EXIT_WARNING : EXIT_SUCCESS); + return 0; /* suppress no-return-value warnings */ +} diff --git a/src/spice2x/external/libjpeg-turbo/jquant1.c b/src/spice2x/external/libjpeg-turbo/jquant1.c new file mode 100644 index 0000000..bd68dcb --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jquant1.c @@ -0,0 +1,864 @@ +/* + * jquant1.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2009, 2015, 2022-2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains 1-pass color quantization (color mapping) routines. + * These routines provide mapping to a fixed color map using equally spaced + * color values. Optional Floyd-Steinberg or ordered dithering is available. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + +#if defined(QUANT_1PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 + + +/* + * The main purpose of 1-pass quantization is to provide a fast, if not very + * high quality, colormapped output capability. A 2-pass quantizer usually + * gives better visual quality; however, for quantized grayscale output this + * quantizer is perfectly adequate. Dithering is highly recommended with this + * quantizer, though you can turn it off if you really want to. + * + * In 1-pass quantization the colormap must be chosen in advance of seeing the + * image. We use a map consisting of all combinations of Ncolors[i] color + * values for the i'th component. The Ncolors[] values are chosen so that + * their product, the total number of colors, is no more than that requested. + * (In most cases, the product will be somewhat less.) + * + * Since the colormap is orthogonal, the representative value for each color + * component can be determined without considering the other components; + * then these indexes can be combined into a colormap index by a standard + * N-dimensional-array-subscript calculation. Most of the arithmetic involved + * can be precalculated and stored in the lookup table colorindex[]. + * colorindex[i][j] maps pixel value j in component i to the nearest + * representative value (grid plane) for that component; this index is + * multiplied by the array stride for component i, so that the + * index of the colormap entry closest to a given pixel value is just + * sum( colorindex[component-number][pixel-component-value] ) + * Aside from being fast, this scheme allows for variable spacing between + * representative values with no additional lookup cost. + * + * If gamma correction has been applied in color conversion, it might be wise + * to adjust the color grid spacing so that the representative colors are + * equidistant in linear space. At this writing, gamma correction is not + * implemented by jdcolor, so nothing is done here. + */ + + +/* Declarations for ordered dithering. + * + * We use a standard 16x16 ordered dither array. The basic concept of ordered + * dithering is described in many references, for instance Dale Schumacher's + * chapter II.2 of Graphics Gems II (James Arvo, ed. Academic Press, 1991). + * In place of Schumacher's comparisons against a "threshold" value, we add a + * "dither" value to the input pixel and then round the result to the nearest + * output value. The dither value is equivalent to (0.5 - threshold) times + * the distance between output values. For ordered dithering, we assume that + * the output colors are equally spaced; if not, results will probably be + * worse, since the dither may be too much or too little at a given point. + * + * The normal calculation would be to form pixel value + dither, range-limit + * this to 0.._MAXJSAMPLE, and then index into the colorindex table as usual. + * We can skip the separate range-limiting step by extending the colorindex + * table in both directions. + */ + +#define ODITHER_SIZE 16 /* dimension of dither matrix */ +/* NB: if ODITHER_SIZE is not a power of 2, ODITHER_MASK uses will break */ +#define ODITHER_CELLS (ODITHER_SIZE * ODITHER_SIZE) /* # cells in matrix */ +#define ODITHER_MASK (ODITHER_SIZE - 1) /* mask for wrapping around + counters */ + +typedef int ODITHER_MATRIX[ODITHER_SIZE][ODITHER_SIZE]; +typedef int (*ODITHER_MATRIX_PTR)[ODITHER_SIZE]; + +static const UINT8 base_dither_matrix[ODITHER_SIZE][ODITHER_SIZE] = { + /* Bayer's order-4 dither array. Generated by the code given in + * Stephen Hawley's article "Ordered Dithering" in Graphics Gems I. + * The values in this array must range from 0 to ODITHER_CELLS-1. + */ + { 0, 192, 48, 240, 12, 204, 60, 252, 3, 195, 51, 243, 15, 207, 63, 255 }, + { 128, 64, 176, 112, 140, 76, 188, 124, 131, 67, 179, 115, 143, 79, 191, 127 }, + { 32, 224, 16, 208, 44, 236, 28, 220, 35, 227, 19, 211, 47, 239, 31, 223 }, + { 160, 96, 144, 80, 172, 108, 156, 92, 163, 99, 147, 83, 175, 111, 159, 95 }, + { 8, 200, 56, 248, 4, 196, 52, 244, 11, 203, 59, 251, 7, 199, 55, 247 }, + { 136, 72, 184, 120, 132, 68, 180, 116, 139, 75, 187, 123, 135, 71, 183, 119 }, + { 40, 232, 24, 216, 36, 228, 20, 212, 43, 235, 27, 219, 39, 231, 23, 215 }, + { 168, 104, 152, 88, 164, 100, 148, 84, 171, 107, 155, 91, 167, 103, 151, 87 }, + { 2, 194, 50, 242, 14, 206, 62, 254, 1, 193, 49, 241, 13, 205, 61, 253 }, + { 130, 66, 178, 114, 142, 78, 190, 126, 129, 65, 177, 113, 141, 77, 189, 125 }, + { 34, 226, 18, 210, 46, 238, 30, 222, 33, 225, 17, 209, 45, 237, 29, 221 }, + { 162, 98, 146, 82, 174, 110, 158, 94, 161, 97, 145, 81, 173, 109, 157, 93 }, + { 10, 202, 58, 250, 6, 198, 54, 246, 9, 201, 57, 249, 5, 197, 53, 245 }, + { 138, 74, 186, 122, 134, 70, 182, 118, 137, 73, 185, 121, 133, 69, 181, 117 }, + { 42, 234, 26, 218, 38, 230, 22, 214, 41, 233, 25, 217, 37, 229, 21, 213 }, + { 170, 106, 154, 90, 166, 102, 150, 86, 169, 105, 153, 89, 165, 101, 149, 85 } +}; + + +/* Declarations for Floyd-Steinberg dithering. + * + * Errors are accumulated into the array fserrors[], at a resolution of + * 1/16th of a pixel count. The error at a given pixel is propagated + * to its not-yet-processed neighbors using the standard F-S fractions, + * ... (here) 7/16 + * 3/16 5/16 1/16 + * We work left-to-right on even rows, right-to-left on odd rows. + * + * We can get away with a single array (holding one row's worth of errors) + * by using it to store the current row's errors at pixel columns not yet + * processed, but the next row's errors at columns already processed. We + * need only a few extra variables to hold the errors immediately around the + * current column. (If we are lucky, those variables are in registers, but + * even if not, they're probably cheaper to access than array elements are.) + * + * The fserrors[] array is indexed [component#][position]. + * We provide (#columns + 2) entries per component; the extra entry at each + * end saves us from special-casing the first and last pixels. + */ + +#if BITS_IN_JSAMPLE == 8 +typedef INT16 FSERROR; /* 16 bits should be enough */ +typedef int LOCFSERROR; /* use 'int' for calculation temps */ +#else +typedef JLONG FSERROR; /* may need more than 16 bits */ +typedef JLONG LOCFSERROR; /* be sure calculation temps are big enough */ +#endif + +typedef FSERROR *FSERRPTR; /* pointer to error array */ + + +/* Private subobject */ + +#define MAX_Q_COMPS 4 /* max components I can handle */ + +typedef struct { + struct jpeg_color_quantizer pub; /* public fields */ + + /* Initially allocated colormap is saved here */ + _JSAMPARRAY sv_colormap; /* The color map as a 2-D pixel array */ + int sv_actual; /* number of entries in use */ + + _JSAMPARRAY colorindex; /* Precomputed mapping for speed */ + /* colorindex[i][j] = index of color closest to pixel value j in component i, + * premultiplied as described above. Since colormap indexes must fit into + * _JSAMPLEs, the entries of this array will too. + */ + boolean is_padded; /* is the colorindex padded for odither? */ + + int Ncolors[MAX_Q_COMPS]; /* # of values allocated to each component */ + + /* Variables for ordered dithering */ + int row_index; /* cur row's vertical index in dither matrix */ + ODITHER_MATRIX_PTR odither[MAX_Q_COMPS]; /* one dither array per component */ + + /* Variables for Floyd-Steinberg dithering */ + FSERRPTR fserrors[MAX_Q_COMPS]; /* accumulated errors */ + boolean on_odd_row; /* flag to remember which row we are on */ +} my_cquantizer; + +typedef my_cquantizer *my_cquantize_ptr; + + +/* + * Policy-making subroutines for create_colormap and create_colorindex. + * These routines determine the colormap to be used. The rest of the module + * only assumes that the colormap is orthogonal. + * + * * select_ncolors decides how to divvy up the available colors + * among the components. + * * output_value defines the set of representative values for a component. + * * largest_input_value defines the mapping from input values to + * representative values for a component. + * Note that the latter two routines may impose different policies for + * different components, though this is not currently done. + */ + + +LOCAL(int) +select_ncolors(j_decompress_ptr cinfo, int Ncolors[]) +/* Determine allocation of desired colors to components, */ +/* and fill in Ncolors[] array to indicate choice. */ +/* Return value is total number of colors (product of Ncolors[] values). */ +{ + int nc = cinfo->out_color_components; /* number of color components */ + int max_colors = cinfo->desired_number_of_colors; + int total_colors, iroot, i, j; + boolean changed; + long temp; + int RGB_order[3] = { RGB_GREEN, RGB_RED, RGB_BLUE }; + RGB_order[0] = rgb_green[cinfo->out_color_space]; + RGB_order[1] = rgb_red[cinfo->out_color_space]; + RGB_order[2] = rgb_blue[cinfo->out_color_space]; + + /* We can allocate at least the nc'th root of max_colors per component. */ + /* Compute floor(nc'th root of max_colors). */ + iroot = 1; + do { + iroot++; + temp = iroot; /* set temp = iroot ** nc */ + for (i = 1; i < nc; i++) + temp *= iroot; + } while (temp <= (long)max_colors); /* repeat till iroot exceeds root */ + iroot--; /* now iroot = floor(root) */ + + /* Must have at least 2 color values per component */ + if (iroot < 2) + ERREXIT1(cinfo, JERR_QUANT_FEW_COLORS, (int)temp); + + /* Initialize to iroot color values for each component */ + total_colors = 1; + for (i = 0; i < nc; i++) { + Ncolors[i] = iroot; + total_colors *= iroot; + } + /* We may be able to increment the count for one or more components without + * exceeding max_colors, though we know not all can be incremented. + * Sometimes, the first component can be incremented more than once! + * (Example: for 16 colors, we start at 2*2*2, go to 3*2*2, then 4*2*2.) + * In RGB colorspace, try to increment G first, then R, then B. + */ + do { + changed = FALSE; + for (i = 0; i < nc; i++) { + j = (cinfo->out_color_space == JCS_RGB ? RGB_order[i] : i); + /* calculate new total_colors if Ncolors[j] is incremented */ + temp = total_colors / Ncolors[j]; + temp *= Ncolors[j] + 1; /* done in long arith to avoid oflo */ + if (temp > (long)max_colors) + break; /* won't fit, done with this pass */ + Ncolors[j]++; /* OK, apply the increment */ + total_colors = (int)temp; + changed = TRUE; + } + } while (changed); + + return total_colors; +} + + +LOCAL(int) +output_value(j_decompress_ptr cinfo, int ci, int j, int maxj) +/* Return j'th output value, where j will range from 0 to maxj */ +/* The output values must fall in 0.._MAXJSAMPLE in increasing order */ +{ + /* We always provide values 0 and _MAXJSAMPLE for each component; + * any additional values are equally spaced between these limits. + * (Forcing the upper and lower values to the limits ensures that + * dithering can't produce a color outside the selected gamut.) + */ + return (int)(((JLONG)j * _MAXJSAMPLE + maxj / 2) / maxj); +} + + +LOCAL(int) +largest_input_value(j_decompress_ptr cinfo, int ci, int j, int maxj) +/* Return largest input value that should map to j'th output value */ +/* Must have largest(j=0) >= 0, and largest(j=maxj) >= _MAXJSAMPLE */ +{ + /* Breakpoints are halfway between values returned by output_value */ + return (int)(((JLONG)(2 * j + 1) * _MAXJSAMPLE + maxj) / (2 * maxj)); +} + + +/* + * Create the colormap. + */ + +LOCAL(void) +create_colormap(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + _JSAMPARRAY colormap; /* Created colormap */ + int total_colors; /* Number of distinct output colors */ + int i, j, k, nci, blksize, blkdist, ptr, val; + + /* Select number of colors for each component */ + total_colors = select_ncolors(cinfo, cquantize->Ncolors); + + /* Report selected color counts */ + if (cinfo->out_color_components == 3) + TRACEMS4(cinfo, 1, JTRC_QUANT_3_NCOLORS, total_colors, + cquantize->Ncolors[0], cquantize->Ncolors[1], + cquantize->Ncolors[2]); + else + TRACEMS1(cinfo, 1, JTRC_QUANT_NCOLORS, total_colors); + + /* Allocate and fill in the colormap. */ + /* The colors are ordered in the map in standard row-major order, */ + /* i.e. rightmost (highest-indexed) color changes most rapidly. */ + + colormap = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (JDIMENSION)total_colors, (JDIMENSION)cinfo->out_color_components); + + /* blksize is number of adjacent repeated entries for a component */ + /* blkdist is distance between groups of identical entries for a component */ + blkdist = total_colors; + + for (i = 0; i < cinfo->out_color_components; i++) { + /* fill in colormap entries for i'th color component */ + nci = cquantize->Ncolors[i]; /* # of distinct values for this color */ + blksize = blkdist / nci; + for (j = 0; j < nci; j++) { + /* Compute j'th output value (out of nci) for component */ + val = output_value(cinfo, i, j, nci - 1); + /* Fill in all colormap entries that have this value of this component */ + for (ptr = j * blksize; ptr < total_colors; ptr += blkdist) { + /* fill in blksize entries beginning at ptr */ + for (k = 0; k < blksize; k++) + colormap[i][ptr + k] = (_JSAMPLE)val; + } + } + blkdist = blksize; /* blksize of this color is blkdist of next */ + } + + /* Save the colormap in private storage, + * where it will survive color quantization mode changes. + */ + cquantize->sv_colormap = colormap; + cquantize->sv_actual = total_colors; +} + + +/* + * Create the color index table. + */ + +LOCAL(void) +create_colorindex(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + _JSAMPROW indexptr; + int i, j, k, nci, blksize, val, pad; + + /* For ordered dither, we pad the color index tables by _MAXJSAMPLE in + * each direction (input index values can be -_MAXJSAMPLE .. 2*_MAXJSAMPLE). + * This is not necessary in the other dithering modes. However, we + * flag whether it was done in case user changes dithering mode. + */ + if (cinfo->dither_mode == JDITHER_ORDERED) { + pad = _MAXJSAMPLE * 2; + cquantize->is_padded = TRUE; + } else { + pad = 0; + cquantize->is_padded = FALSE; + } + + cquantize->colorindex = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (JDIMENSION)(_MAXJSAMPLE + 1 + pad), + (JDIMENSION)cinfo->out_color_components); + + /* blksize is number of adjacent repeated entries for a component */ + blksize = cquantize->sv_actual; + + for (i = 0; i < cinfo->out_color_components; i++) { + /* fill in colorindex entries for i'th color component */ + nci = cquantize->Ncolors[i]; /* # of distinct values for this color */ + blksize = blksize / nci; + + /* adjust colorindex pointers to provide padding at negative indexes. */ + if (pad) + cquantize->colorindex[i] += _MAXJSAMPLE; + + /* in loop, val = index of current output value, */ + /* and k = largest j that maps to current val */ + indexptr = cquantize->colorindex[i]; + val = 0; + k = largest_input_value(cinfo, i, 0, nci - 1); + for (j = 0; j <= _MAXJSAMPLE; j++) { + while (j > k) /* advance val if past boundary */ + k = largest_input_value(cinfo, i, ++val, nci - 1); + /* premultiply so that no multiplication needed in main processing */ + indexptr[j] = (_JSAMPLE)(val * blksize); + } + /* Pad at both ends if necessary */ + if (pad) + for (j = 1; j <= _MAXJSAMPLE; j++) { + indexptr[-j] = indexptr[0]; + indexptr[_MAXJSAMPLE + j] = indexptr[_MAXJSAMPLE]; + } + } +} + + +/* + * Create an ordered-dither array for a component having ncolors + * distinct output values. + */ + +LOCAL(ODITHER_MATRIX_PTR) +make_odither_array(j_decompress_ptr cinfo, int ncolors) +{ + ODITHER_MATRIX_PTR odither; + int j, k; + JLONG num, den; + + odither = (ODITHER_MATRIX_PTR) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(ODITHER_MATRIX)); + /* The inter-value distance for this color is _MAXJSAMPLE/(ncolors-1). + * Hence the dither value for the matrix cell with fill order f + * (f=0..N-1) should be (N-1-2*f)/(2*N) * _MAXJSAMPLE/(ncolors-1). + * On 16-bit-int machine, be careful to avoid overflow. + */ + den = 2 * ODITHER_CELLS * ((JLONG)(ncolors - 1)); + for (j = 0; j < ODITHER_SIZE; j++) { + for (k = 0; k < ODITHER_SIZE; k++) { + num = ((JLONG)(ODITHER_CELLS - 1 - + 2 * ((int)base_dither_matrix[j][k]))) * _MAXJSAMPLE; + /* Ensure round towards zero despite C's lack of consistency + * about rounding negative values in integer division... + */ + odither[j][k] = (int)(num < 0 ? -((-num) / den) : num / den); + } + } + return odither; +} + + +/* + * Create the ordered-dither tables. + * Components having the same number of representative colors may + * share a dither table. + */ + +LOCAL(void) +create_odither_tables(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + ODITHER_MATRIX_PTR odither; + int i, j, nci; + + for (i = 0; i < cinfo->out_color_components; i++) { + nci = cquantize->Ncolors[i]; /* # of distinct values for this color */ + odither = NULL; /* search for matching prior component */ + for (j = 0; j < i; j++) { + if (nci == cquantize->Ncolors[j]) { + odither = cquantize->odither[j]; + break; + } + } + if (odither == NULL) /* need a new table? */ + odither = make_odither_array(cinfo, nci); + cquantize->odither[i] = odither; + } +} + + +/* + * Map some rows of pixels to the output colormapped representation. + */ + +METHODDEF(void) +color_quantize(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +/* General case, no dithering */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + _JSAMPARRAY colorindex = cquantize->colorindex; + register int pixcode, ci; + register _JSAMPROW ptrin, ptrout; + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + register int nc = cinfo->out_color_components; + + for (row = 0; row < num_rows; row++) { + ptrin = input_buf[row]; + ptrout = output_buf[row]; + for (col = width; col > 0; col--) { + pixcode = 0; + for (ci = 0; ci < nc; ci++) { + pixcode += colorindex[ci][*ptrin++]; + } + *ptrout++ = (_JSAMPLE)pixcode; + } + } +} + + +METHODDEF(void) +color_quantize3(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +/* Fast path for out_color_components==3, no dithering */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + register int pixcode; + register _JSAMPROW ptrin, ptrout; + _JSAMPROW colorindex0 = cquantize->colorindex[0]; + _JSAMPROW colorindex1 = cquantize->colorindex[1]; + _JSAMPROW colorindex2 = cquantize->colorindex[2]; + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + + for (row = 0; row < num_rows; row++) { + ptrin = input_buf[row]; + ptrout = output_buf[row]; + for (col = width; col > 0; col--) { + pixcode = colorindex0[*ptrin++]; + pixcode += colorindex1[*ptrin++]; + pixcode += colorindex2[*ptrin++]; + *ptrout++ = (_JSAMPLE)pixcode; + } + } +} + + +METHODDEF(void) +quantize_ord_dither(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +/* General case, with ordered dithering */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + register _JSAMPROW input_ptr; + register _JSAMPROW output_ptr; + _JSAMPROW colorindex_ci; + int *dither; /* points to active row of dither matrix */ + int row_index, col_index; /* current indexes into dither matrix */ + int nc = cinfo->out_color_components; + int ci; + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + + for (row = 0; row < num_rows; row++) { + /* Initialize output values to 0 so can process components separately */ + jzero_far((void *)output_buf[row], (size_t)(width * sizeof(_JSAMPLE))); + row_index = cquantize->row_index; + for (ci = 0; ci < nc; ci++) { + input_ptr = input_buf[row] + ci; + output_ptr = output_buf[row]; + colorindex_ci = cquantize->colorindex[ci]; + dither = cquantize->odither[ci][row_index]; + col_index = 0; + + for (col = width; col > 0; col--) { + /* Form pixel value + dither, range-limit to 0.._MAXJSAMPLE, + * select output value, accumulate into output code for this pixel. + * Range-limiting need not be done explicitly, as we have extended + * the colorindex table to produce the right answers for out-of-range + * inputs. The maximum dither is +- _MAXJSAMPLE; this sets the + * required amount of padding. + */ + *output_ptr += + colorindex_ci[*input_ptr + dither[col_index]]; + input_ptr += nc; + output_ptr++; + col_index = (col_index + 1) & ODITHER_MASK; + } + } + /* Advance row index for next row */ + row_index = (row_index + 1) & ODITHER_MASK; + cquantize->row_index = row_index; + } +} + + +METHODDEF(void) +quantize3_ord_dither(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +/* Fast path for out_color_components==3, with ordered dithering */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + register int pixcode; + register _JSAMPROW input_ptr; + register _JSAMPROW output_ptr; + _JSAMPROW colorindex0 = cquantize->colorindex[0]; + _JSAMPROW colorindex1 = cquantize->colorindex[1]; + _JSAMPROW colorindex2 = cquantize->colorindex[2]; + int *dither0; /* points to active row of dither matrix */ + int *dither1; + int *dither2; + int row_index, col_index; /* current indexes into dither matrix */ + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + + for (row = 0; row < num_rows; row++) { + row_index = cquantize->row_index; + input_ptr = input_buf[row]; + output_ptr = output_buf[row]; + dither0 = cquantize->odither[0][row_index]; + dither1 = cquantize->odither[1][row_index]; + dither2 = cquantize->odither[2][row_index]; + col_index = 0; + + for (col = width; col > 0; col--) { + pixcode = colorindex0[(*input_ptr++) + dither0[col_index]]; + pixcode += colorindex1[(*input_ptr++) + dither1[col_index]]; + pixcode += colorindex2[(*input_ptr++) + dither2[col_index]]; + *output_ptr++ = (_JSAMPLE)pixcode; + col_index = (col_index + 1) & ODITHER_MASK; + } + row_index = (row_index + 1) & ODITHER_MASK; + cquantize->row_index = row_index; + } +} + + +METHODDEF(void) +quantize_fs_dither(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +/* General case, with Floyd-Steinberg dithering */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + register LOCFSERROR cur; /* current error or pixel value */ + LOCFSERROR belowerr; /* error for pixel below cur */ + LOCFSERROR bpreverr; /* error for below/prev col */ + LOCFSERROR bnexterr; /* error for below/next col */ + LOCFSERROR delta; + register FSERRPTR errorptr; /* => fserrors[] at column before current */ + register _JSAMPROW input_ptr; + register _JSAMPROW output_ptr; + _JSAMPROW colorindex_ci; + _JSAMPROW colormap_ci; + int pixcode; + int nc = cinfo->out_color_components; + int dir; /* 1 for left-to-right, -1 for right-to-left */ + int dirnc; /* dir * nc */ + int ci; + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + SHIFT_TEMPS + + for (row = 0; row < num_rows; row++) { + /* Initialize output values to 0 so can process components separately */ + jzero_far((void *)output_buf[row], (size_t)(width * sizeof(_JSAMPLE))); + for (ci = 0; ci < nc; ci++) { + input_ptr = input_buf[row] + ci; + output_ptr = output_buf[row]; + if (cquantize->on_odd_row) { + /* work right to left in this row */ + input_ptr += (width - 1) * nc; /* so point to rightmost pixel */ + output_ptr += width - 1; + dir = -1; + dirnc = -nc; + errorptr = cquantize->fserrors[ci] + (width + 1); /* => entry after last column */ + } else { + /* work left to right in this row */ + dir = 1; + dirnc = nc; + errorptr = cquantize->fserrors[ci]; /* => entry before first column */ + } + colorindex_ci = cquantize->colorindex[ci]; + colormap_ci = cquantize->sv_colormap[ci]; + /* Preset error values: no error propagated to first pixel from left */ + cur = 0; + /* and no error propagated to row below yet */ + belowerr = bpreverr = 0; + + for (col = width; col > 0; col--) { + /* cur holds the error propagated from the previous pixel on the + * current line. Add the error propagated from the previous line + * to form the complete error correction term for this pixel, and + * round the error term (which is expressed * 16) to an integer. + * RIGHT_SHIFT rounds towards minus infinity, so adding 8 is correct + * for either sign of the error value. + * Note: errorptr points to *previous* column's array entry. + */ + cur = RIGHT_SHIFT(cur + errorptr[dir] + 8, 4); + /* Form pixel value + error, and range-limit to 0.._MAXJSAMPLE. + * The maximum error is +- _MAXJSAMPLE; this sets the required size + * of the range_limit array. + */ + cur += *input_ptr; + cur = range_limit[cur]; + /* Select output value, accumulate into output code for this pixel */ + pixcode = colorindex_ci[cur]; + *output_ptr += (_JSAMPLE)pixcode; + /* Compute actual representation error at this pixel */ + /* Note: we can do this even though we don't have the final */ + /* pixel code, because the colormap is orthogonal. */ + cur -= colormap_ci[pixcode]; + /* Compute error fractions to be propagated to adjacent pixels. + * Add these into the running sums, and simultaneously shift the + * next-line error sums left by 1 column. + */ + bnexterr = cur; + delta = cur * 2; + cur += delta; /* form error * 3 */ + errorptr[0] = (FSERROR)(bpreverr + cur); + cur += delta; /* form error * 5 */ + bpreverr = belowerr + cur; + belowerr = bnexterr; + cur += delta; /* form error * 7 */ + /* At this point cur contains the 7/16 error value to be propagated + * to the next pixel on the current line, and all the errors for the + * next line have been shifted over. We are therefore ready to move on. + */ + input_ptr += dirnc; /* advance input ptr to next column */ + output_ptr += dir; /* advance output ptr to next column */ + errorptr += dir; /* advance errorptr to current column */ + } + /* Post-loop cleanup: we must unload the final error value into the + * final fserrors[] entry. Note we need not unload belowerr because + * it is for the dummy column before or after the actual array. + */ + errorptr[0] = (FSERROR)bpreverr; /* unload prev err into array */ + } + cquantize->on_odd_row = (cquantize->on_odd_row ? FALSE : TRUE); + } +} + + +/* + * Allocate workspace for Floyd-Steinberg errors. + */ + +LOCAL(void) +alloc_fs_workspace(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + size_t arraysize; + int i; + + arraysize = (size_t)((cinfo->output_width + 2) * sizeof(FSERROR)); + for (i = 0; i < cinfo->out_color_components; i++) { + cquantize->fserrors[i] = (FSERRPTR) + (*cinfo->mem->alloc_large) ((j_common_ptr)cinfo, JPOOL_IMAGE, arraysize); + } +} + + +/* + * Initialize for one-pass color quantization. + */ + +METHODDEF(void) +start_pass_1_quant(j_decompress_ptr cinfo, boolean is_pre_scan) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + size_t arraysize; + int i; + + /* Install my colormap. */ + cinfo->colormap = (JSAMPARRAY)cquantize->sv_colormap; + cinfo->actual_number_of_colors = cquantize->sv_actual; + + /* Initialize for desired dithering mode. */ + switch (cinfo->dither_mode) { + case JDITHER_NONE: + if (cinfo->out_color_components == 3) + cquantize->pub._color_quantize = color_quantize3; + else + cquantize->pub._color_quantize = color_quantize; + break; + case JDITHER_ORDERED: + if (cinfo->out_color_components == 3) + cquantize->pub._color_quantize = quantize3_ord_dither; + else + cquantize->pub._color_quantize = quantize_ord_dither; + cquantize->row_index = 0; /* initialize state for ordered dither */ + /* If user changed to ordered dither from another mode, + * we must recreate the color index table with padding. + * This will cost extra space, but probably isn't very likely. + */ + if (!cquantize->is_padded) + create_colorindex(cinfo); + /* Create ordered-dither tables if we didn't already. */ + if (cquantize->odither[0] == NULL) + create_odither_tables(cinfo); + break; + case JDITHER_FS: + cquantize->pub._color_quantize = quantize_fs_dither; + cquantize->on_odd_row = FALSE; /* initialize state for F-S dither */ + /* Allocate Floyd-Steinberg workspace if didn't already. */ + if (cquantize->fserrors[0] == NULL) + alloc_fs_workspace(cinfo); + /* Initialize the propagated errors to zero. */ + arraysize = (size_t)((cinfo->output_width + 2) * sizeof(FSERROR)); + for (i = 0; i < cinfo->out_color_components; i++) + jzero_far((void *)cquantize->fserrors[i], arraysize); + break; + default: + ERREXIT(cinfo, JERR_NOT_COMPILED); + break; + } +} + + +/* + * Finish up at the end of the pass. + */ + +METHODDEF(void) +finish_pass_1_quant(j_decompress_ptr cinfo) +{ + /* no work in 1-pass case */ +} + + +/* + * Switch to a new external colormap between output passes. + * Shouldn't get to this module! + */ + +METHODDEF(void) +new_color_map_1_quant(j_decompress_ptr cinfo) +{ + ERREXIT(cinfo, JERR_MODE_CHANGE); +} + + +/* + * Module initialization routine for 1-pass color quantization. + */ + +GLOBAL(void) +_jinit_1pass_quantizer(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + /* Color quantization is not supported with lossless JPEG images */ + if (cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + cquantize = (my_cquantize_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_cquantizer)); + cinfo->cquantize = (struct jpeg_color_quantizer *)cquantize; + cquantize->pub.start_pass = start_pass_1_quant; + cquantize->pub.finish_pass = finish_pass_1_quant; + cquantize->pub.new_color_map = new_color_map_1_quant; + cquantize->fserrors[0] = NULL; /* Flag FS workspace not allocated */ + cquantize->odither[0] = NULL; /* Also flag odither arrays not allocated */ + + /* Make sure my internal arrays won't overflow */ + if (cinfo->out_color_components > MAX_Q_COMPS) + ERREXIT1(cinfo, JERR_QUANT_COMPONENTS, MAX_Q_COMPS); + /* Make sure colormap indexes can be represented by _JSAMPLEs */ + if (cinfo->desired_number_of_colors > (_MAXJSAMPLE + 1)) + ERREXIT1(cinfo, JERR_QUANT_MANY_COLORS, _MAXJSAMPLE + 1); + + /* Create the colormap and color index table. */ + create_colormap(cinfo); + create_colorindex(cinfo); + + /* Allocate Floyd-Steinberg workspace now if requested. + * We do this now since it may affect the memory manager's space + * calculations. If the user changes to FS dither mode in a later pass, we + * will allocate the space then, and will possibly overrun the + * max_memory_to_use setting. + */ + if (cinfo->dither_mode == JDITHER_FS) + alloc_fs_workspace(cinfo); +} + +#endif /* defined(QUANT_1PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 */ diff --git a/src/spice2x/external/libjpeg-turbo/jquant2.c b/src/spice2x/external/libjpeg-turbo/jquant2.c new file mode 100644 index 0000000..9ba51fa --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jquant2.c @@ -0,0 +1,1293 @@ +/* + * jquant2.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2009, 2014-2015, 2020, 2022-2023, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains 2-pass color quantization (color mapping) routines. + * These routines provide selection of a custom color map for an image, + * followed by mapping of the image to that color map, with optional + * Floyd-Steinberg dithering. + * It is also possible to use just the second pass to map to an arbitrary + * externally-given color map. + * + * Note: ordered dithering is not supported, since there isn't any fast + * way to compute intercolor distances; it's unclear that ordered dither's + * fundamental assumptions even hold with an irregularly spaced color map. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + +#if defined(QUANT_2PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 + + +/* + * This module implements the well-known Heckbert paradigm for color + * quantization. Most of the ideas used here can be traced back to + * Heckbert's seminal paper + * Heckbert, Paul. "Color Image Quantization for Frame Buffer Display", + * Proc. SIGGRAPH '82, Computer Graphics v.16 #3 (July 1982), pp 297-304. + * + * In the first pass over the image, we accumulate a histogram showing the + * usage count of each possible color. To keep the histogram to a reasonable + * size, we reduce the precision of the input; typical practice is to retain + * 5 or 6 bits per color, so that 8 or 4 different input values are counted + * in the same histogram cell. + * + * Next, the color-selection step begins with a box representing the whole + * color space, and repeatedly splits the "largest" remaining box until we + * have as many boxes as desired colors. Then the mean color in each + * remaining box becomes one of the possible output colors. + * + * The second pass over the image maps each input pixel to the closest output + * color (optionally after applying a Floyd-Steinberg dithering correction). + * This mapping is logically trivial, but making it go fast enough requires + * considerable care. + * + * Heckbert-style quantizers vary a good deal in their policies for choosing + * the "largest" box and deciding where to cut it. The particular policies + * used here have proved out well in experimental comparisons, but better ones + * may yet be found. + * + * In earlier versions of the IJG code, this module quantized in YCbCr color + * space, processing the raw upsampled data without a color conversion step. + * This allowed the color conversion math to be done only once per colormap + * entry, not once per pixel. However, that optimization precluded other + * useful optimizations (such as merging color conversion with upsampling) + * and it also interfered with desired capabilities such as quantizing to an + * externally-supplied colormap. We have therefore abandoned that approach. + * The present code works in the post-conversion color space, typically RGB. + * + * To improve the visual quality of the results, we actually work in scaled + * RGB space, giving G distances more weight than R, and R in turn more than + * B. To do everything in integer math, we must use integer scale factors. + * The 2/3/1 scale factors used here correspond loosely to the relative + * weights of the colors in the NTSC grayscale equation. + * If you want to use this code to quantize a non-RGB color space, you'll + * probably need to change these scale factors. + */ + +#define R_SCALE 2 /* scale R distances by this much */ +#define G_SCALE 3 /* scale G distances by this much */ +#define B_SCALE 1 /* and B by this much */ + +static const int c_scales[3] = { R_SCALE, G_SCALE, B_SCALE }; +#define C0_SCALE c_scales[rgb_red[cinfo->out_color_space]] +#define C1_SCALE c_scales[rgb_green[cinfo->out_color_space]] +#define C2_SCALE c_scales[rgb_blue[cinfo->out_color_space]] + +/* + * First we have the histogram data structure and routines for creating it. + * + * The number of bits of precision can be adjusted by changing these symbols. + * We recommend keeping 6 bits for G and 5 each for R and B. + * If you have plenty of memory and cycles, 6 bits all around gives marginally + * better results; if you are short of memory, 5 bits all around will save + * some space but degrade the results. + * To maintain a fully accurate histogram, we'd need to allocate a "long" + * (preferably unsigned long) for each cell. In practice this is overkill; + * we can get by with 16 bits per cell. Few of the cell counts will overflow, + * and clamping those that do overflow to the maximum value will give close- + * enough results. This reduces the recommended histogram size from 256Kb + * to 128Kb, which is a useful savings on PC-class machines. + * (In the second pass the histogram space is re-used for pixel mapping data; + * in that capacity, each cell must be able to store zero to the number of + * desired colors. 16 bits/cell is plenty for that too.) + * Since the JPEG code is intended to run in small memory model on 80x86 + * machines, we can't just allocate the histogram in one chunk. Instead + * of a true 3-D array, we use a row of pointers to 2-D arrays. Each + * pointer corresponds to a C0 value (typically 2^5 = 32 pointers) and + * each 2-D array has 2^6*2^5 = 2048 or 2^6*2^6 = 4096 entries. + */ + +#define MAXNUMCOLORS (_MAXJSAMPLE + 1) /* maximum size of colormap */ + +/* These will do the right thing for either R,G,B or B,G,R color order, + * but you may not like the results for other color orders. + */ +#define HIST_C0_BITS 5 /* bits of precision in R/B histogram */ +#define HIST_C1_BITS 6 /* bits of precision in G histogram */ +#define HIST_C2_BITS 5 /* bits of precision in B/R histogram */ + +/* Number of elements along histogram axes. */ +#define HIST_C0_ELEMS (1 << HIST_C0_BITS) +#define HIST_C1_ELEMS (1 << HIST_C1_BITS) +#define HIST_C2_ELEMS (1 << HIST_C2_BITS) + +/* These are the amounts to shift an input value to get a histogram index. */ +#define C0_SHIFT (BITS_IN_JSAMPLE - HIST_C0_BITS) +#define C1_SHIFT (BITS_IN_JSAMPLE - HIST_C1_BITS) +#define C2_SHIFT (BITS_IN_JSAMPLE - HIST_C2_BITS) + + +typedef UINT16 histcell; /* histogram cell; prefer an unsigned type */ + +typedef histcell *histptr; /* for pointers to histogram cells */ + +typedef histcell hist1d[HIST_C2_ELEMS]; /* typedefs for the array */ +typedef hist1d *hist2d; /* type for the 2nd-level pointers */ +typedef hist2d *hist3d; /* type for top-level pointer */ + + +/* Declarations for Floyd-Steinberg dithering. + * + * Errors are accumulated into the array fserrors[], at a resolution of + * 1/16th of a pixel count. The error at a given pixel is propagated + * to its not-yet-processed neighbors using the standard F-S fractions, + * ... (here) 7/16 + * 3/16 5/16 1/16 + * We work left-to-right on even rows, right-to-left on odd rows. + * + * We can get away with a single array (holding one row's worth of errors) + * by using it to store the current row's errors at pixel columns not yet + * processed, but the next row's errors at columns already processed. We + * need only a few extra variables to hold the errors immediately around the + * current column. (If we are lucky, those variables are in registers, but + * even if not, they're probably cheaper to access than array elements are.) + * + * The fserrors[] array has (#columns + 2) entries; the extra entry at + * each end saves us from special-casing the first and last pixels. + * Each entry is three values long, one value for each color component. + */ + +#if BITS_IN_JSAMPLE == 8 +typedef INT16 FSERROR; /* 16 bits should be enough */ +typedef int LOCFSERROR; /* use 'int' for calculation temps */ +#else +typedef JLONG FSERROR; /* may need more than 16 bits */ +typedef JLONG LOCFSERROR; /* be sure calculation temps are big enough */ +#endif + +typedef FSERROR *FSERRPTR; /* pointer to error array */ + + +/* Private subobject */ + +typedef struct { + struct jpeg_color_quantizer pub; /* public fields */ + + /* Space for the eventually created colormap is stashed here */ + _JSAMPARRAY sv_colormap; /* colormap allocated at init time */ + int desired; /* desired # of colors = size of colormap */ + + /* Variables for accumulating image statistics */ + hist3d histogram; /* pointer to the histogram */ + + boolean needs_zeroed; /* TRUE if next pass must zero histogram */ + + /* Variables for Floyd-Steinberg dithering */ + FSERRPTR fserrors; /* accumulated errors */ + boolean on_odd_row; /* flag to remember which row we are on */ + int *error_limiter; /* table for clamping the applied error */ +} my_cquantizer; + +typedef my_cquantizer *my_cquantize_ptr; + + +/* + * Prescan some rows of pixels. + * In this module the prescan simply updates the histogram, which has been + * initialized to zeroes by start_pass. + * An output_buf parameter is required by the method signature, but no data + * is actually output (in fact the buffer controller is probably passing a + * NULL pointer). + */ + +METHODDEF(void) +prescan_quantize(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + register _JSAMPROW ptr; + register histptr histp; + register hist3d histogram = cquantize->histogram; + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + + for (row = 0; row < num_rows; row++) { + ptr = input_buf[row]; + for (col = width; col > 0; col--) { + /* get pixel value and index into the histogram */ + histp = &histogram[ptr[0] >> C0_SHIFT] + [ptr[1] >> C1_SHIFT] + [ptr[2] >> C2_SHIFT]; + /* increment, check for overflow and undo increment if so. */ + if (++(*histp) <= 0) + (*histp)--; + ptr += 3; + } + } +} + + +/* + * Next we have the really interesting routines: selection of a colormap + * given the completed histogram. + * These routines work with a list of "boxes", each representing a rectangular + * subset of the input color space (to histogram precision). + */ + +typedef struct { + /* The bounds of the box (inclusive); expressed as histogram indexes */ + int c0min, c0max; + int c1min, c1max; + int c2min, c2max; + /* The volume (actually 2-norm) of the box */ + JLONG volume; + /* The number of nonzero histogram cells within this box */ + long colorcount; +} box; + +typedef box *boxptr; + + +LOCAL(boxptr) +find_biggest_color_pop(boxptr boxlist, int numboxes) +/* Find the splittable box with the largest color population */ +/* Returns NULL if no splittable boxes remain */ +{ + register boxptr boxp; + register int i; + register long maxc = 0; + boxptr which = NULL; + + for (i = 0, boxp = boxlist; i < numboxes; i++, boxp++) { + if (boxp->colorcount > maxc && boxp->volume > 0) { + which = boxp; + maxc = boxp->colorcount; + } + } + return which; +} + + +LOCAL(boxptr) +find_biggest_volume(boxptr boxlist, int numboxes) +/* Find the splittable box with the largest (scaled) volume */ +/* Returns NULL if no splittable boxes remain */ +{ + register boxptr boxp; + register int i; + register JLONG maxv = 0; + boxptr which = NULL; + + for (i = 0, boxp = boxlist; i < numboxes; i++, boxp++) { + if (boxp->volume > maxv) { + which = boxp; + maxv = boxp->volume; + } + } + return which; +} + + +LOCAL(void) +update_box(j_decompress_ptr cinfo, boxptr boxp) +/* Shrink the min/max bounds of a box to enclose only nonzero elements, */ +/* and recompute its volume and population */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + hist3d histogram = cquantize->histogram; + histptr histp; + int c0, c1, c2; + int c0min, c0max, c1min, c1max, c2min, c2max; + JLONG dist0, dist1, dist2; + long ccount; + + c0min = boxp->c0min; c0max = boxp->c0max; + c1min = boxp->c1min; c1max = boxp->c1max; + c2min = boxp->c2min; c2max = boxp->c2max; + + if (c0max > c0min) + for (c0 = c0min; c0 <= c0max; c0++) + for (c1 = c1min; c1 <= c1max; c1++) { + histp = &histogram[c0][c1][c2min]; + for (c2 = c2min; c2 <= c2max; c2++) + if (*histp++ != 0) { + boxp->c0min = c0min = c0; + goto have_c0min; + } + } +have_c0min: + if (c0max > c0min) + for (c0 = c0max; c0 >= c0min; c0--) + for (c1 = c1min; c1 <= c1max; c1++) { + histp = &histogram[c0][c1][c2min]; + for (c2 = c2min; c2 <= c2max; c2++) + if (*histp++ != 0) { + boxp->c0max = c0max = c0; + goto have_c0max; + } + } +have_c0max: + if (c1max > c1min) + for (c1 = c1min; c1 <= c1max; c1++) + for (c0 = c0min; c0 <= c0max; c0++) { + histp = &histogram[c0][c1][c2min]; + for (c2 = c2min; c2 <= c2max; c2++) + if (*histp++ != 0) { + boxp->c1min = c1min = c1; + goto have_c1min; + } + } +have_c1min: + if (c1max > c1min) + for (c1 = c1max; c1 >= c1min; c1--) + for (c0 = c0min; c0 <= c0max; c0++) { + histp = &histogram[c0][c1][c2min]; + for (c2 = c2min; c2 <= c2max; c2++) + if (*histp++ != 0) { + boxp->c1max = c1max = c1; + goto have_c1max; + } + } +have_c1max: + if (c2max > c2min) + for (c2 = c2min; c2 <= c2max; c2++) + for (c0 = c0min; c0 <= c0max; c0++) { + histp = &histogram[c0][c1min][c2]; + for (c1 = c1min; c1 <= c1max; c1++, histp += HIST_C2_ELEMS) + if (*histp != 0) { + boxp->c2min = c2min = c2; + goto have_c2min; + } + } +have_c2min: + if (c2max > c2min) + for (c2 = c2max; c2 >= c2min; c2--) + for (c0 = c0min; c0 <= c0max; c0++) { + histp = &histogram[c0][c1min][c2]; + for (c1 = c1min; c1 <= c1max; c1++, histp += HIST_C2_ELEMS) + if (*histp != 0) { + boxp->c2max = c2max = c2; + goto have_c2max; + } + } +have_c2max: + + /* Update box volume. + * We use 2-norm rather than real volume here; this biases the method + * against making long narrow boxes, and it has the side benefit that + * a box is splittable iff norm > 0. + * Since the differences are expressed in histogram-cell units, + * we have to shift back to _JSAMPLE units to get consistent distances; + * after which, we scale according to the selected distance scale factors. + */ + dist0 = ((c0max - c0min) << C0_SHIFT) * C0_SCALE; + dist1 = ((c1max - c1min) << C1_SHIFT) * C1_SCALE; + dist2 = ((c2max - c2min) << C2_SHIFT) * C2_SCALE; + boxp->volume = dist0 * dist0 + dist1 * dist1 + dist2 * dist2; + + /* Now scan remaining volume of box and compute population */ + ccount = 0; + for (c0 = c0min; c0 <= c0max; c0++) + for (c1 = c1min; c1 <= c1max; c1++) { + histp = &histogram[c0][c1][c2min]; + for (c2 = c2min; c2 <= c2max; c2++, histp++) + if (*histp != 0) { + ccount++; + } + } + boxp->colorcount = ccount; +} + + +LOCAL(int) +median_cut(j_decompress_ptr cinfo, boxptr boxlist, int numboxes, + int desired_colors) +/* Repeatedly select and split the largest box until we have enough boxes */ +{ + int n, lb; + int c0, c1, c2, cmax; + register boxptr b1, b2; + + while (numboxes < desired_colors) { + /* Select box to split. + * Current algorithm: by population for first half, then by volume. + */ + if (numboxes * 2 <= desired_colors) { + b1 = find_biggest_color_pop(boxlist, numboxes); + } else { + b1 = find_biggest_volume(boxlist, numboxes); + } + if (b1 == NULL) /* no splittable boxes left! */ + break; + b2 = &boxlist[numboxes]; /* where new box will go */ + /* Copy the color bounds to the new box. */ + b2->c0max = b1->c0max; b2->c1max = b1->c1max; b2->c2max = b1->c2max; + b2->c0min = b1->c0min; b2->c1min = b1->c1min; b2->c2min = b1->c2min; + /* Choose which axis to split the box on. + * Current algorithm: longest scaled axis. + * See notes in update_box about scaling distances. + */ + c0 = ((b1->c0max - b1->c0min) << C0_SHIFT) * C0_SCALE; + c1 = ((b1->c1max - b1->c1min) << C1_SHIFT) * C1_SCALE; + c2 = ((b1->c2max - b1->c2min) << C2_SHIFT) * C2_SCALE; + /* We want to break any ties in favor of green, then red, blue last. + * This code does the right thing for R,G,B or B,G,R color orders only. + */ + if (rgb_red[cinfo->out_color_space] == 0) { + cmax = c1; n = 1; + if (c0 > cmax) { cmax = c0; n = 0; } + if (c2 > cmax) { n = 2; } + } else { + cmax = c1; n = 1; + if (c2 > cmax) { cmax = c2; n = 2; } + if (c0 > cmax) { n = 0; } + } + /* Choose split point along selected axis, and update box bounds. + * Current algorithm: split at halfway point. + * (Since the box has been shrunk to minimum volume, + * any split will produce two nonempty subboxes.) + * Note that lb value is max for lower box, so must be < old max. + */ + switch (n) { + case 0: + lb = (b1->c0max + b1->c0min) / 2; + b1->c0max = lb; + b2->c0min = lb + 1; + break; + case 1: + lb = (b1->c1max + b1->c1min) / 2; + b1->c1max = lb; + b2->c1min = lb + 1; + break; + case 2: + lb = (b1->c2max + b1->c2min) / 2; + b1->c2max = lb; + b2->c2min = lb + 1; + break; + } + /* Update stats for boxes */ + update_box(cinfo, b1); + update_box(cinfo, b2); + numboxes++; + } + return numboxes; +} + + +LOCAL(void) +compute_color(j_decompress_ptr cinfo, boxptr boxp, int icolor) +/* Compute representative color for a box, put it in colormap[icolor] */ +{ + /* Current algorithm: mean weighted by pixels (not colors) */ + /* Note it is important to get the rounding correct! */ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + hist3d histogram = cquantize->histogram; + histptr histp; + int c0, c1, c2; + int c0min, c0max, c1min, c1max, c2min, c2max; + long count; + long total = 0; + long c0total = 0; + long c1total = 0; + long c2total = 0; + + c0min = boxp->c0min; c0max = boxp->c0max; + c1min = boxp->c1min; c1max = boxp->c1max; + c2min = boxp->c2min; c2max = boxp->c2max; + + for (c0 = c0min; c0 <= c0max; c0++) + for (c1 = c1min; c1 <= c1max; c1++) { + histp = &histogram[c0][c1][c2min]; + for (c2 = c2min; c2 <= c2max; c2++) { + if ((count = *histp++) != 0) { + total += count; + c0total += ((c0 << C0_SHIFT) + ((1 << C0_SHIFT) >> 1)) * count; + c1total += ((c1 << C1_SHIFT) + ((1 << C1_SHIFT) >> 1)) * count; + c2total += ((c2 << C2_SHIFT) + ((1 << C2_SHIFT) >> 1)) * count; + } + } + } + + ((_JSAMPARRAY)cinfo->colormap)[0][icolor] = + (_JSAMPLE)((c0total + (total >> 1)) / total); + ((_JSAMPARRAY)cinfo->colormap)[1][icolor] = + (_JSAMPLE)((c1total + (total >> 1)) / total); + ((_JSAMPARRAY)cinfo->colormap)[2][icolor] = + (_JSAMPLE)((c2total + (total >> 1)) / total); +} + + +LOCAL(void) +select_colors(j_decompress_ptr cinfo, int desired_colors) +/* Master routine for color selection */ +{ + boxptr boxlist; + int numboxes; + int i; + + /* Allocate workspace for box list */ + boxlist = (boxptr)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, desired_colors * sizeof(box)); + /* Initialize one box containing whole space */ + numboxes = 1; + boxlist[0].c0min = 0; + boxlist[0].c0max = _MAXJSAMPLE >> C0_SHIFT; + boxlist[0].c1min = 0; + boxlist[0].c1max = _MAXJSAMPLE >> C1_SHIFT; + boxlist[0].c2min = 0; + boxlist[0].c2max = _MAXJSAMPLE >> C2_SHIFT; + /* Shrink it to actually-used volume and set its statistics */ + update_box(cinfo, &boxlist[0]); + /* Perform median-cut to produce final box list */ + numboxes = median_cut(cinfo, boxlist, numboxes, desired_colors); + /* Compute the representative color for each box, fill colormap */ + for (i = 0; i < numboxes; i++) + compute_color(cinfo, &boxlist[i], i); + cinfo->actual_number_of_colors = numboxes; + TRACEMS1(cinfo, 1, JTRC_QUANT_SELECTED, numboxes); +} + + +/* + * These routines are concerned with the time-critical task of mapping input + * colors to the nearest color in the selected colormap. + * + * We re-use the histogram space as an "inverse color map", essentially a + * cache for the results of nearest-color searches. All colors within a + * histogram cell will be mapped to the same colormap entry, namely the one + * closest to the cell's center. This may not be quite the closest entry to + * the actual input color, but it's almost as good. A zero in the cache + * indicates we haven't found the nearest color for that cell yet; the array + * is cleared to zeroes before starting the mapping pass. When we find the + * nearest color for a cell, its colormap index plus one is recorded in the + * cache for future use. The pass2 scanning routines call fill_inverse_cmap + * when they need to use an unfilled entry in the cache. + * + * Our method of efficiently finding nearest colors is based on the "locally + * sorted search" idea described by Heckbert and on the incremental distance + * calculation described by Spencer W. Thomas in chapter III.1 of Graphics + * Gems II (James Arvo, ed. Academic Press, 1991). Thomas points out that + * the distances from a given colormap entry to each cell of the histogram can + * be computed quickly using an incremental method: the differences between + * distances to adjacent cells themselves differ by a constant. This allows a + * fairly fast implementation of the "brute force" approach of computing the + * distance from every colormap entry to every histogram cell. Unfortunately, + * it needs a work array to hold the best-distance-so-far for each histogram + * cell (because the inner loop has to be over cells, not colormap entries). + * The work array elements have to be JLONGs, so the work array would need + * 256Kb at our recommended precision. This is not feasible in DOS machines. + * + * To get around these problems, we apply Thomas' method to compute the + * nearest colors for only the cells within a small subbox of the histogram. + * The work array need be only as big as the subbox, so the memory usage + * problem is solved. Furthermore, we need not fill subboxes that are never + * referenced in pass2; many images use only part of the color gamut, so a + * fair amount of work is saved. An additional advantage of this + * approach is that we can apply Heckbert's locality criterion to quickly + * eliminate colormap entries that are far away from the subbox; typically + * three-fourths of the colormap entries are rejected by Heckbert's criterion, + * and we need not compute their distances to individual cells in the subbox. + * The speed of this approach is heavily influenced by the subbox size: too + * small means too much overhead, too big loses because Heckbert's criterion + * can't eliminate as many colormap entries. Empirically the best subbox + * size seems to be about 1/512th of the histogram (1/8th in each direction). + * + * Thomas' article also describes a refined method which is asymptotically + * faster than the brute-force method, but it is also far more complex and + * cannot efficiently be applied to small subboxes. It is therefore not + * useful for programs intended to be portable to DOS machines. On machines + * with plenty of memory, filling the whole histogram in one shot with Thomas' + * refined method might be faster than the present code --- but then again, + * it might not be any faster, and it's certainly more complicated. + */ + + +/* log2(histogram cells in update box) for each axis; this can be adjusted */ +#define BOX_C0_LOG (HIST_C0_BITS - 3) +#define BOX_C1_LOG (HIST_C1_BITS - 3) +#define BOX_C2_LOG (HIST_C2_BITS - 3) + +#define BOX_C0_ELEMS (1 << BOX_C0_LOG) /* # of hist cells in update box */ +#define BOX_C1_ELEMS (1 << BOX_C1_LOG) +#define BOX_C2_ELEMS (1 << BOX_C2_LOG) + +#define BOX_C0_SHIFT (C0_SHIFT + BOX_C0_LOG) +#define BOX_C1_SHIFT (C1_SHIFT + BOX_C1_LOG) +#define BOX_C2_SHIFT (C2_SHIFT + BOX_C2_LOG) + + +/* + * The next three routines implement inverse colormap filling. They could + * all be folded into one big routine, but splitting them up this way saves + * some stack space (the mindist[] and bestdist[] arrays need not coexist) + * and may allow some compilers to produce better code by registerizing more + * inner-loop variables. + */ + +LOCAL(int) +find_nearby_colors(j_decompress_ptr cinfo, int minc0, int minc1, int minc2, + _JSAMPLE colorlist[]) +/* Locate the colormap entries close enough to an update box to be candidates + * for the nearest entry to some cell(s) in the update box. The update box + * is specified by the center coordinates of its first cell. The number of + * candidate colormap entries is returned, and their colormap indexes are + * placed in colorlist[]. + * This routine uses Heckbert's "locally sorted search" criterion to select + * the colors that need further consideration. + */ +{ + int numcolors = cinfo->actual_number_of_colors; + int maxc0, maxc1, maxc2; + int centerc0, centerc1, centerc2; + int i, x, ncolors; + JLONG minmaxdist, min_dist, max_dist, tdist; + JLONG mindist[MAXNUMCOLORS]; /* min distance to colormap entry i */ + + /* Compute true coordinates of update box's upper corner and center. + * Actually we compute the coordinates of the center of the upper-corner + * histogram cell, which are the upper bounds of the volume we care about. + * Note that since ">>" rounds down, the "center" values may be closer to + * min than to max; hence comparisons to them must be "<=", not "<". + */ + maxc0 = minc0 + ((1 << BOX_C0_SHIFT) - (1 << C0_SHIFT)); + centerc0 = (minc0 + maxc0) >> 1; + maxc1 = minc1 + ((1 << BOX_C1_SHIFT) - (1 << C1_SHIFT)); + centerc1 = (minc1 + maxc1) >> 1; + maxc2 = minc2 + ((1 << BOX_C2_SHIFT) - (1 << C2_SHIFT)); + centerc2 = (minc2 + maxc2) >> 1; + + /* For each color in colormap, find: + * 1. its minimum squared-distance to any point in the update box + * (zero if color is within update box); + * 2. its maximum squared-distance to any point in the update box. + * Both of these can be found by considering only the corners of the box. + * We save the minimum distance for each color in mindist[]; + * only the smallest maximum distance is of interest. + */ + minmaxdist = 0x7FFFFFFFL; + + for (i = 0; i < numcolors; i++) { + /* We compute the squared-c0-distance term, then add in the other two. */ + x = ((_JSAMPARRAY)cinfo->colormap)[0][i]; + if (x < minc0) { + tdist = (x - minc0) * C0_SCALE; + min_dist = tdist * tdist; + tdist = (x - maxc0) * C0_SCALE; + max_dist = tdist * tdist; + } else if (x > maxc0) { + tdist = (x - maxc0) * C0_SCALE; + min_dist = tdist * tdist; + tdist = (x - minc0) * C0_SCALE; + max_dist = tdist * tdist; + } else { + /* within cell range so no contribution to min_dist */ + min_dist = 0; + if (x <= centerc0) { + tdist = (x - maxc0) * C0_SCALE; + max_dist = tdist * tdist; + } else { + tdist = (x - minc0) * C0_SCALE; + max_dist = tdist * tdist; + } + } + + x = ((_JSAMPARRAY)cinfo->colormap)[1][i]; + if (x < minc1) { + tdist = (x - minc1) * C1_SCALE; + min_dist += tdist * tdist; + tdist = (x - maxc1) * C1_SCALE; + max_dist += tdist * tdist; + } else if (x > maxc1) { + tdist = (x - maxc1) * C1_SCALE; + min_dist += tdist * tdist; + tdist = (x - minc1) * C1_SCALE; + max_dist += tdist * tdist; + } else { + /* within cell range so no contribution to min_dist */ + if (x <= centerc1) { + tdist = (x - maxc1) * C1_SCALE; + max_dist += tdist * tdist; + } else { + tdist = (x - minc1) * C1_SCALE; + max_dist += tdist * tdist; + } + } + + x = ((_JSAMPARRAY)cinfo->colormap)[2][i]; + if (x < minc2) { + tdist = (x - minc2) * C2_SCALE; + min_dist += tdist * tdist; + tdist = (x - maxc2) * C2_SCALE; + max_dist += tdist * tdist; + } else if (x > maxc2) { + tdist = (x - maxc2) * C2_SCALE; + min_dist += tdist * tdist; + tdist = (x - minc2) * C2_SCALE; + max_dist += tdist * tdist; + } else { + /* within cell range so no contribution to min_dist */ + if (x <= centerc2) { + tdist = (x - maxc2) * C2_SCALE; + max_dist += tdist * tdist; + } else { + tdist = (x - minc2) * C2_SCALE; + max_dist += tdist * tdist; + } + } + + mindist[i] = min_dist; /* save away the results */ + if (max_dist < minmaxdist) + minmaxdist = max_dist; + } + + /* Now we know that no cell in the update box is more than minmaxdist + * away from some colormap entry. Therefore, only colors that are + * within minmaxdist of some part of the box need be considered. + */ + ncolors = 0; + for (i = 0; i < numcolors; i++) { + if (mindist[i] <= minmaxdist) + colorlist[ncolors++] = (_JSAMPLE)i; + } + return ncolors; +} + + +LOCAL(void) +find_best_colors(j_decompress_ptr cinfo, int minc0, int minc1, int minc2, + int numcolors, _JSAMPLE colorlist[], _JSAMPLE bestcolor[]) +/* Find the closest colormap entry for each cell in the update box, + * given the list of candidate colors prepared by find_nearby_colors. + * Return the indexes of the closest entries in the bestcolor[] array. + * This routine uses Thomas' incremental distance calculation method to + * find the distance from a colormap entry to successive cells in the box. + */ +{ + int ic0, ic1, ic2; + int i, icolor; + register JLONG *bptr; /* pointer into bestdist[] array */ + _JSAMPLE *cptr; /* pointer into bestcolor[] array */ + JLONG dist0, dist1; /* initial distance values */ + register JLONG dist2; /* current distance in inner loop */ + JLONG xx0, xx1; /* distance increments */ + register JLONG xx2; + JLONG inc0, inc1, inc2; /* initial values for increments */ + /* This array holds the distance to the nearest-so-far color for each cell */ + JLONG bestdist[BOX_C0_ELEMS * BOX_C1_ELEMS * BOX_C2_ELEMS]; + + /* Initialize best-distance for each cell of the update box */ + bptr = bestdist; + for (i = BOX_C0_ELEMS * BOX_C1_ELEMS * BOX_C2_ELEMS - 1; i >= 0; i--) + *bptr++ = 0x7FFFFFFFL; + + /* For each color selected by find_nearby_colors, + * compute its distance to the center of each cell in the box. + * If that's less than best-so-far, update best distance and color number. + */ + + /* Nominal steps between cell centers ("x" in Thomas article) */ +#define STEP_C0 ((1 << C0_SHIFT) * C0_SCALE) +#define STEP_C1 ((1 << C1_SHIFT) * C1_SCALE) +#define STEP_C2 ((1 << C2_SHIFT) * C2_SCALE) + + for (i = 0; i < numcolors; i++) { + icolor = colorlist[i]; + /* Compute (square of) distance from minc0/c1/c2 to this color */ + inc0 = (minc0 - ((_JSAMPARRAY)cinfo->colormap)[0][icolor]) * C0_SCALE; + dist0 = inc0 * inc0; + inc1 = (minc1 - ((_JSAMPARRAY)cinfo->colormap)[1][icolor]) * C1_SCALE; + dist0 += inc1 * inc1; + inc2 = (minc2 - ((_JSAMPARRAY)cinfo->colormap)[2][icolor]) * C2_SCALE; + dist0 += inc2 * inc2; + /* Form the initial difference increments */ + inc0 = inc0 * (2 * STEP_C0) + STEP_C0 * STEP_C0; + inc1 = inc1 * (2 * STEP_C1) + STEP_C1 * STEP_C1; + inc2 = inc2 * (2 * STEP_C2) + STEP_C2 * STEP_C2; + /* Now loop over all cells in box, updating distance per Thomas method */ + bptr = bestdist; + cptr = bestcolor; + xx0 = inc0; + for (ic0 = BOX_C0_ELEMS - 1; ic0 >= 0; ic0--) { + dist1 = dist0; + xx1 = inc1; + for (ic1 = BOX_C1_ELEMS - 1; ic1 >= 0; ic1--) { + dist2 = dist1; + xx2 = inc2; + for (ic2 = BOX_C2_ELEMS - 1; ic2 >= 0; ic2--) { + if (dist2 < *bptr) { + *bptr = dist2; + *cptr = (_JSAMPLE)icolor; + } + dist2 += xx2; + xx2 += 2 * STEP_C2 * STEP_C2; + bptr++; + cptr++; + } + dist1 += xx1; + xx1 += 2 * STEP_C1 * STEP_C1; + } + dist0 += xx0; + xx0 += 2 * STEP_C0 * STEP_C0; + } + } +} + + +LOCAL(void) +fill_inverse_cmap(j_decompress_ptr cinfo, int c0, int c1, int c2) +/* Fill the inverse-colormap entries in the update box that contains */ +/* histogram cell c0/c1/c2. (Only that one cell MUST be filled, but */ +/* we can fill as many others as we wish.) */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + hist3d histogram = cquantize->histogram; + int minc0, minc1, minc2; /* lower left corner of update box */ + int ic0, ic1, ic2; + register _JSAMPLE *cptr; /* pointer into bestcolor[] array */ + register histptr cachep; /* pointer into main cache array */ + /* This array lists the candidate colormap indexes. */ + _JSAMPLE colorlist[MAXNUMCOLORS]; + int numcolors; /* number of candidate colors */ + /* This array holds the actually closest colormap index for each cell. */ + _JSAMPLE bestcolor[BOX_C0_ELEMS * BOX_C1_ELEMS * BOX_C2_ELEMS]; + + /* Convert cell coordinates to update box ID */ + c0 >>= BOX_C0_LOG; + c1 >>= BOX_C1_LOG; + c2 >>= BOX_C2_LOG; + + /* Compute true coordinates of update box's origin corner. + * Actually we compute the coordinates of the center of the corner + * histogram cell, which are the lower bounds of the volume we care about. + */ + minc0 = (c0 << BOX_C0_SHIFT) + ((1 << C0_SHIFT) >> 1); + minc1 = (c1 << BOX_C1_SHIFT) + ((1 << C1_SHIFT) >> 1); + minc2 = (c2 << BOX_C2_SHIFT) + ((1 << C2_SHIFT) >> 1); + + /* Determine which colormap entries are close enough to be candidates + * for the nearest entry to some cell in the update box. + */ + numcolors = find_nearby_colors(cinfo, minc0, minc1, minc2, colorlist); + + /* Determine the actually nearest colors. */ + find_best_colors(cinfo, minc0, minc1, minc2, numcolors, colorlist, + bestcolor); + + /* Save the best color numbers (plus 1) in the main cache array */ + c0 <<= BOX_C0_LOG; /* convert ID back to base cell indexes */ + c1 <<= BOX_C1_LOG; + c2 <<= BOX_C2_LOG; + cptr = bestcolor; + for (ic0 = 0; ic0 < BOX_C0_ELEMS; ic0++) { + for (ic1 = 0; ic1 < BOX_C1_ELEMS; ic1++) { + cachep = &histogram[c0 + ic0][c1 + ic1][c2]; + for (ic2 = 0; ic2 < BOX_C2_ELEMS; ic2++) { + *cachep++ = (histcell)((*cptr++) + 1); + } + } + } +} + + +/* + * Map some rows of pixels to the output colormapped representation. + */ + +METHODDEF(void) +pass2_no_dither(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +/* This version performs no dithering */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + hist3d histogram = cquantize->histogram; + register _JSAMPROW inptr, outptr; + register histptr cachep; + register int c0, c1, c2; + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + + for (row = 0; row < num_rows; row++) { + inptr = input_buf[row]; + outptr = output_buf[row]; + for (col = width; col > 0; col--) { + /* get pixel value and index into the cache */ + c0 = (*inptr++) >> C0_SHIFT; + c1 = (*inptr++) >> C1_SHIFT; + c2 = (*inptr++) >> C2_SHIFT; + cachep = &histogram[c0][c1][c2]; + /* If we have not seen this color before, find nearest colormap entry */ + /* and update the cache */ + if (*cachep == 0) + fill_inverse_cmap(cinfo, c0, c1, c2); + /* Now emit the colormap index for this cell */ + *outptr++ = (_JSAMPLE)(*cachep - 1); + } + } +} + + +METHODDEF(void) +pass2_fs_dither(j_decompress_ptr cinfo, _JSAMPARRAY input_buf, + _JSAMPARRAY output_buf, int num_rows) +/* This version performs Floyd-Steinberg dithering */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + hist3d histogram = cquantize->histogram; + register LOCFSERROR cur0, cur1, cur2; /* current error or pixel value */ + LOCFSERROR belowerr0, belowerr1, belowerr2; /* error for pixel below cur */ + LOCFSERROR bpreverr0, bpreverr1, bpreverr2; /* error for below/prev col */ + register FSERRPTR errorptr; /* => fserrors[] at column before current */ + _JSAMPROW inptr; /* => current input pixel */ + _JSAMPROW outptr; /* => current output pixel */ + histptr cachep; + int dir; /* +1 or -1 depending on direction */ + int dir3; /* 3*dir, for advancing inptr & errorptr */ + int row; + JDIMENSION col; + JDIMENSION width = cinfo->output_width; + _JSAMPLE *range_limit = (_JSAMPLE *)cinfo->sample_range_limit; + int *error_limit = cquantize->error_limiter; + _JSAMPROW colormap0 = ((_JSAMPARRAY)cinfo->colormap)[0]; + _JSAMPROW colormap1 = ((_JSAMPARRAY)cinfo->colormap)[1]; + _JSAMPROW colormap2 = ((_JSAMPARRAY)cinfo->colormap)[2]; + SHIFT_TEMPS + + for (row = 0; row < num_rows; row++) { + inptr = input_buf[row]; + outptr = output_buf[row]; + if (cquantize->on_odd_row) { + /* work right to left in this row */ + inptr += (width - 1) * 3; /* so point to rightmost pixel */ + outptr += width - 1; + dir = -1; + dir3 = -3; + errorptr = cquantize->fserrors + (width + 1) * 3; /* => entry after last column */ + cquantize->on_odd_row = FALSE; /* flip for next time */ + } else { + /* work left to right in this row */ + dir = 1; + dir3 = 3; + errorptr = cquantize->fserrors; /* => entry before first real column */ + cquantize->on_odd_row = TRUE; /* flip for next time */ + } + /* Preset error values: no error propagated to first pixel from left */ + cur0 = cur1 = cur2 = 0; + /* and no error propagated to row below yet */ + belowerr0 = belowerr1 = belowerr2 = 0; + bpreverr0 = bpreverr1 = bpreverr2 = 0; + + for (col = width; col > 0; col--) { + /* curN holds the error propagated from the previous pixel on the + * current line. Add the error propagated from the previous line + * to form the complete error correction term for this pixel, and + * round the error term (which is expressed * 16) to an integer. + * RIGHT_SHIFT rounds towards minus infinity, so adding 8 is correct + * for either sign of the error value. + * Note: errorptr points to *previous* column's array entry. + */ + cur0 = RIGHT_SHIFT(cur0 + errorptr[dir3 + 0] + 8, 4); + cur1 = RIGHT_SHIFT(cur1 + errorptr[dir3 + 1] + 8, 4); + cur2 = RIGHT_SHIFT(cur2 + errorptr[dir3 + 2] + 8, 4); + /* Limit the error using transfer function set by init_error_limit. + * See comments with init_error_limit for rationale. + */ + cur0 = error_limit[cur0]; + cur1 = error_limit[cur1]; + cur2 = error_limit[cur2]; + /* Form pixel value + error, and range-limit to 0.._MAXJSAMPLE. + * The maximum error is +- _MAXJSAMPLE (or less with error limiting); + * this sets the required size of the range_limit array. + */ + cur0 += inptr[0]; + cur1 += inptr[1]; + cur2 += inptr[2]; + cur0 = range_limit[cur0]; + cur1 = range_limit[cur1]; + cur2 = range_limit[cur2]; + /* Index into the cache with adjusted pixel value */ + cachep = + &histogram[cur0 >> C0_SHIFT][cur1 >> C1_SHIFT][cur2 >> C2_SHIFT]; + /* If we have not seen this color before, find nearest colormap */ + /* entry and update the cache */ + if (*cachep == 0) + fill_inverse_cmap(cinfo, cur0 >> C0_SHIFT, cur1 >> C1_SHIFT, + cur2 >> C2_SHIFT); + /* Now emit the colormap index for this cell */ + { + register int pixcode = *cachep - 1; + *outptr = (_JSAMPLE)pixcode; + /* Compute representation error for this pixel */ + cur0 -= colormap0[pixcode]; + cur1 -= colormap1[pixcode]; + cur2 -= colormap2[pixcode]; + } + /* Compute error fractions to be propagated to adjacent pixels. + * Add these into the running sums, and simultaneously shift the + * next-line error sums left by 1 column. + */ + { + register LOCFSERROR bnexterr; + + bnexterr = cur0; /* Process component 0 */ + errorptr[0] = (FSERROR)(bpreverr0 + cur0 * 3); + bpreverr0 = belowerr0 + cur0 * 5; + belowerr0 = bnexterr; + cur0 *= 7; + bnexterr = cur1; /* Process component 1 */ + errorptr[1] = (FSERROR)(bpreverr1 + cur1 * 3); + bpreverr1 = belowerr1 + cur1 * 5; + belowerr1 = bnexterr; + cur1 *= 7; + bnexterr = cur2; /* Process component 2 */ + errorptr[2] = (FSERROR)(bpreverr2 + cur2 * 3); + bpreverr2 = belowerr2 + cur2 * 5; + belowerr2 = bnexterr; + cur2 *= 7; + } + /* At this point curN contains the 7/16 error value to be propagated + * to the next pixel on the current line, and all the errors for the + * next line have been shifted over. We are therefore ready to move on. + */ + inptr += dir3; /* Advance pixel pointers to next column */ + outptr += dir; + errorptr += dir3; /* advance errorptr to current column */ + } + /* Post-loop cleanup: we must unload the final error values into the + * final fserrors[] entry. Note we need not unload belowerrN because + * it is for the dummy column before or after the actual array. + */ + errorptr[0] = (FSERROR)bpreverr0; /* unload prev errs into array */ + errorptr[1] = (FSERROR)bpreverr1; + errorptr[2] = (FSERROR)bpreverr2; + } +} + + +/* + * Initialize the error-limiting transfer function (lookup table). + * The raw F-S error computation can potentially compute error values of up to + * +- _MAXJSAMPLE. But we want the maximum correction applied to a pixel to be + * much less, otherwise obviously wrong pixels will be created. (Typical + * effects include weird fringes at color-area boundaries, isolated bright + * pixels in a dark area, etc.) The standard advice for avoiding this problem + * is to ensure that the "corners" of the color cube are allocated as output + * colors; then repeated errors in the same direction cannot cause cascading + * error buildup. However, that only prevents the error from getting + * completely out of hand; Aaron Giles reports that error limiting improves + * the results even with corner colors allocated. + * A simple clamping of the error values to about +- _MAXJSAMPLE/8 works pretty + * well, but the smoother transfer function used below is even better. Thanks + * to Aaron Giles for this idea. + */ + +LOCAL(void) +init_error_limit(j_decompress_ptr cinfo) +/* Allocate and fill in the error_limiter table */ +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + int *table; + int in, out; + + table = (int *)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, (_MAXJSAMPLE * 2 + 1) * sizeof(int)); + table += _MAXJSAMPLE; /* so can index -_MAXJSAMPLE .. +_MAXJSAMPLE */ + cquantize->error_limiter = table; + +#define STEPSIZE ((_MAXJSAMPLE + 1) / 16) + /* Map errors 1:1 up to +- _MAXJSAMPLE/16 */ + out = 0; + for (in = 0; in < STEPSIZE; in++, out++) { + table[in] = out; table[-in] = -out; + } + /* Map errors 1:2 up to +- 3*_MAXJSAMPLE/16 */ + for (; in < STEPSIZE * 3; in++, out += (in & 1) ? 0 : 1) { + table[in] = out; table[-in] = -out; + } + /* Clamp the rest to final out value (which is (_MAXJSAMPLE+1)/8) */ + for (; in <= _MAXJSAMPLE; in++) { + table[in] = out; table[-in] = -out; + } +#undef STEPSIZE +} + + +/* + * Finish up at the end of each pass. + */ + +METHODDEF(void) +finish_pass1(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + + /* Select the representative colors and fill in cinfo->colormap */ + cinfo->colormap = (JSAMPARRAY)cquantize->sv_colormap; + select_colors(cinfo, cquantize->desired); + /* Force next pass to zero the color index table */ + cquantize->needs_zeroed = TRUE; +} + + +METHODDEF(void) +finish_pass2(j_decompress_ptr cinfo) +{ + /* no work */ +} + + +/* + * Initialize for each processing pass. + */ + +METHODDEF(void) +start_pass_2_quant(j_decompress_ptr cinfo, boolean is_pre_scan) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + hist3d histogram = cquantize->histogram; + int i; + + /* Only F-S dithering or no dithering is supported. */ + /* If user asks for ordered dither, give them F-S. */ + if (cinfo->dither_mode != JDITHER_NONE) + cinfo->dither_mode = JDITHER_FS; + + if (is_pre_scan) { + /* Set up method pointers */ + cquantize->pub._color_quantize = prescan_quantize; + cquantize->pub.finish_pass = finish_pass1; + cquantize->needs_zeroed = TRUE; /* Always zero histogram */ + } else { + /* Set up method pointers */ + if (cinfo->dither_mode == JDITHER_FS) + cquantize->pub._color_quantize = pass2_fs_dither; + else + cquantize->pub._color_quantize = pass2_no_dither; + cquantize->pub.finish_pass = finish_pass2; + + /* Make sure color count is acceptable */ + i = cinfo->actual_number_of_colors; + if (i < 1) + ERREXIT1(cinfo, JERR_QUANT_FEW_COLORS, 1); + if (i > MAXNUMCOLORS) + ERREXIT1(cinfo, JERR_QUANT_MANY_COLORS, MAXNUMCOLORS); + + if (cinfo->dither_mode == JDITHER_FS) { + size_t arraysize = + (size_t)((cinfo->output_width + 2) * (3 * sizeof(FSERROR))); + /* Allocate Floyd-Steinberg workspace if we didn't already. */ + if (cquantize->fserrors == NULL) + cquantize->fserrors = (FSERRPTR)(*cinfo->mem->alloc_large) + ((j_common_ptr)cinfo, JPOOL_IMAGE, arraysize); + /* Initialize the propagated errors to zero. */ + jzero_far((void *)cquantize->fserrors, arraysize); + /* Make the error-limit table if we didn't already. */ + if (cquantize->error_limiter == NULL) + init_error_limit(cinfo); + cquantize->on_odd_row = FALSE; + } + + } + /* Zero the histogram or inverse color map, if necessary */ + if (cquantize->needs_zeroed) { + for (i = 0; i < HIST_C0_ELEMS; i++) { + jzero_far((void *)histogram[i], + HIST_C1_ELEMS * HIST_C2_ELEMS * sizeof(histcell)); + } + cquantize->needs_zeroed = FALSE; + } +} + + +/* + * Switch to a new external colormap between output passes. + */ + +METHODDEF(void) +new_color_map_2_quant(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize = (my_cquantize_ptr)cinfo->cquantize; + + /* Reset the inverse color map */ + cquantize->needs_zeroed = TRUE; +} + + +/* + * Module initialization routine for 2-pass color quantization. + */ + +GLOBAL(void) +_jinit_2pass_quantizer(j_decompress_ptr cinfo) +{ + my_cquantize_ptr cquantize; + int i; + + if (cinfo->data_precision != BITS_IN_JSAMPLE) + ERREXIT1(cinfo, JERR_BAD_PRECISION, cinfo->data_precision); + + cquantize = (my_cquantize_ptr) + (*cinfo->mem->alloc_small) ((j_common_ptr)cinfo, JPOOL_IMAGE, + sizeof(my_cquantizer)); + cinfo->cquantize = (struct jpeg_color_quantizer *)cquantize; + cquantize->pub.start_pass = start_pass_2_quant; + cquantize->pub.new_color_map = new_color_map_2_quant; + cquantize->fserrors = NULL; /* flag optional arrays not allocated */ + cquantize->error_limiter = NULL; + + /* Make sure jdmaster didn't give me a case I can't handle */ + if (cinfo->out_color_components != 3 || + cinfo->out_color_space == JCS_RGB565 || cinfo->master->lossless) + ERREXIT(cinfo, JERR_NOTIMPL); + + /* Allocate the histogram/inverse colormap storage */ + cquantize->histogram = (hist3d)(*cinfo->mem->alloc_small) + ((j_common_ptr)cinfo, JPOOL_IMAGE, HIST_C0_ELEMS * sizeof(hist2d)); + for (i = 0; i < HIST_C0_ELEMS; i++) { + cquantize->histogram[i] = (hist2d)(*cinfo->mem->alloc_large) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + HIST_C1_ELEMS * HIST_C2_ELEMS * sizeof(histcell)); + } + cquantize->needs_zeroed = TRUE; /* histogram is garbage now */ + + /* Allocate storage for the completed colormap, if required. + * We do this now since it may affect the memory manager's space + * calculations. + */ + if (cinfo->enable_2pass_quant) { + /* Make sure color count is acceptable */ + int desired = cinfo->desired_number_of_colors; + /* Lower bound on # of colors ... somewhat arbitrary as long as > 0 */ + if (desired < 8) + ERREXIT1(cinfo, JERR_QUANT_FEW_COLORS, 8); + /* Make sure colormap indexes can be represented by _JSAMPLEs */ + if (desired > MAXNUMCOLORS) + ERREXIT1(cinfo, JERR_QUANT_MANY_COLORS, MAXNUMCOLORS); + cquantize->sv_colormap = (_JSAMPARRAY)(*cinfo->mem->alloc_sarray) + ((j_common_ptr)cinfo, JPOOL_IMAGE, (JDIMENSION)desired, (JDIMENSION)3); + cquantize->desired = desired; + } else + cquantize->sv_colormap = NULL; + + /* Only F-S dithering or no dithering is supported. */ + /* If user asks for ordered dither, give them F-S. */ + if (cinfo->dither_mode != JDITHER_NONE) + cinfo->dither_mode = JDITHER_FS; + + /* Allocate Floyd-Steinberg workspace if necessary. + * This isn't really needed until pass 2, but again it may affect the memory + * manager's space calculations. Although we will cope with a later change + * in dither_mode, we do not promise to honor max_memory_to_use if + * dither_mode changes. + */ + if (cinfo->dither_mode == JDITHER_FS) { + cquantize->fserrors = (FSERRPTR)(*cinfo->mem->alloc_large) + ((j_common_ptr)cinfo, JPOOL_IMAGE, + (size_t)((cinfo->output_width + 2) * (3 * sizeof(FSERROR)))); + /* Might as well create the error-limiting table too. */ + init_error_limit(cinfo); + } +} + +#endif /* defined(QUANT_2PASS_SUPPORTED) && BITS_IN_JSAMPLE != 16 */ diff --git a/src/spice2x/external/libjpeg-turbo/jsamplecomp.h b/src/spice2x/external/libjpeg-turbo/jsamplecomp.h new file mode 100644 index 0000000..f3f275e --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jsamplecomp.h @@ -0,0 +1,336 @@ +/* + * jsamplecomp.h + * + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + */ + +/* In source files that must be compiled for multiple data precisions, we + * prefix all precision-dependent data types, macros, methods, fields, and + * function names with an underscore. Including this file replaces those + * precision-independent tokens with their precision-dependent equivalents, + * based on the value of BITS_IN_JSAMPLE. + */ + +#ifndef JSAMPLECOMP_H +#define JSAMPLECOMP_H + +#if BITS_IN_JSAMPLE == 16 + +/* Sample data types and macros (jmorecfg.h) */ +#define _JSAMPLE J16SAMPLE + +#define _MAXJSAMPLE MAXJ16SAMPLE +#define _CENTERJSAMPLE CENTERJ16SAMPLE + +#define _JSAMPROW J16SAMPROW +#define _JSAMPARRAY J16SAMPARRAY +#define _JSAMPIMAGE J16SAMPIMAGE + +/* External functions (jpeglib.h) */ +#define _jpeg_write_scanlines jpeg16_write_scanlines +#define _jpeg_read_scanlines jpeg16_read_scanlines + +/* Internal methods (jpegint.h) */ + +#ifdef C_LOSSLESS_SUPPORTED +/* Use the 16-bit method in the jpeg_c_main_controller structure. */ +#define _process_data process_data_16 +/* Use the 16-bit method in the jpeg_c_prep_controller structure. */ +#define _pre_process_data pre_process_data_16 +/* Use the 16-bit method in the jpeg_c_coef_controller structure. */ +#define _compress_data compress_data_16 +/* Use the 16-bit method in the jpeg_color_converter structure. */ +#define _color_convert color_convert_16 +/* Use the 16-bit method in the jpeg_downsampler structure. */ +#define _downsample downsample_16 +#endif +#ifdef D_LOSSLESS_SUPPORTED +/* Use the 16-bit method in the jpeg_d_main_controller structure. */ +#define _process_data process_data_16 +/* Use the 16-bit method in the jpeg_d_coef_controller structure. */ +#define _decompress_data decompress_data_16 +/* Use the 16-bit method in the jpeg_d_post_controller structure. */ +#define _post_process_data post_process_data_16 +/* Use the 16-bit method in the jpeg_upsampler structure. */ +#define _upsample upsample_16 +/* Use the 16-bit method in the jpeg_color_converter structure. */ +#define _color_convert color_convert_16 +#endif + +/* Global internal functions (jpegint.h) */ +#ifdef C_LOSSLESS_SUPPORTED +#define _jinit_c_main_controller j16init_c_main_controller +#define _jinit_c_prep_controller j16init_c_prep_controller +#define _jinit_color_converter j16init_color_converter +#define _jinit_downsampler j16init_downsampler +#define _jinit_c_diff_controller j16init_c_diff_controller +#define _jinit_lossless_compressor j16init_lossless_compressor +#endif + +#ifdef D_LOSSLESS_SUPPORTED +#define _jinit_d_main_controller j16init_d_main_controller +#define _jinit_d_post_controller j16init_d_post_controller +#define _jinit_upsampler j16init_upsampler +#define _jinit_color_deconverter j16init_color_deconverter +#define _jinit_merged_upsampler j16init_merged_upsampler +#define _jinit_d_diff_controller j16init_d_diff_controller +#define _jinit_lossless_decompressor j16init_lossless_decompressor +#endif + +#if defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) +#define _jcopy_sample_rows j16copy_sample_rows +#endif + +/* Internal fields (cdjpeg.h) */ + +#if defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) +/* Use the 16-bit buffer in the cjpeg_source_struct and djpeg_dest_struct + structures. */ +#define _buffer buffer16 +#endif + +/* Image I/O functions (cdjpeg.h) */ +#ifdef C_LOSSLESS_SUPPORTED +#define _jinit_read_gif j16init_read_gif +#define _jinit_read_ppm j16init_read_ppm +#endif + +#ifdef D_LOSSLESS_SUPPORTED +#define _jinit_write_ppm j16init_write_ppm +#endif + +#elif BITS_IN_JSAMPLE == 12 + +/* Sample data types and macros (jmorecfg.h) */ +#define _JSAMPLE J12SAMPLE + +#define _MAXJSAMPLE MAXJ12SAMPLE +#define _CENTERJSAMPLE CENTERJ12SAMPLE + +#define _JSAMPROW J12SAMPROW +#define _JSAMPARRAY J12SAMPARRAY +#define _JSAMPIMAGE J12SAMPIMAGE + +/* External functions (jpeglib.h) */ +#define _jpeg_write_scanlines jpeg12_write_scanlines +#define _jpeg_write_raw_data jpeg12_write_raw_data +#define _jpeg_read_scanlines jpeg12_read_scanlines +#define _jpeg_skip_scanlines jpeg12_skip_scanlines +#define _jpeg_crop_scanline jpeg12_crop_scanline +#define _jpeg_read_raw_data jpeg12_read_raw_data + +/* Internal methods (jpegint.h) */ + +/* Use the 12-bit method in the jpeg_c_main_controller structure. */ +#define _process_data process_data_12 +/* Use the 12-bit method in the jpeg_c_prep_controller structure. */ +#define _pre_process_data pre_process_data_12 +/* Use the 12-bit method in the jpeg_c_coef_controller structure. */ +#define _compress_data compress_data_12 +/* Use the 12-bit method in the jpeg_color_converter structure. */ +#define _color_convert color_convert_12 +/* Use the 12-bit method in the jpeg_downsampler structure. */ +#define _downsample downsample_12 +/* Use the 12-bit method in the jpeg_forward_dct structure. */ +#define _forward_DCT forward_DCT_12 +/* Use the 12-bit method in the jpeg_d_main_controller structure. */ +#define _process_data process_data_12 +/* Use the 12-bit method in the jpeg_d_coef_controller structure. */ +#define _decompress_data decompress_data_12 +/* Use the 12-bit method in the jpeg_d_post_controller structure. */ +#define _post_process_data post_process_data_12 +/* Use the 12-bit method in the jpeg_inverse_dct structure. */ +#define _inverse_DCT_method_ptr inverse_DCT_12_method_ptr +#define _inverse_DCT inverse_DCT_12 +/* Use the 12-bit method in the jpeg_upsampler structure. */ +#define _upsample upsample_12 +/* Use the 12-bit method in the jpeg_color_converter structure. */ +#define _color_convert color_convert_12 +/* Use the 12-bit method in the jpeg_color_quantizer structure. */ +#define _color_quantize color_quantize_12 + +/* Global internal functions (jpegint.h) */ +#define _jinit_c_main_controller j12init_c_main_controller +#define _jinit_c_prep_controller j12init_c_prep_controller +#define _jinit_c_coef_controller j12init_c_coef_controller +#define _jinit_color_converter j12init_color_converter +#define _jinit_downsampler j12init_downsampler +#define _jinit_forward_dct j12init_forward_dct +#ifdef C_LOSSLESS_SUPPORTED +#define _jinit_c_diff_controller j12init_c_diff_controller +#define _jinit_lossless_compressor j12init_lossless_compressor +#endif + +#define _jinit_d_main_controller j12init_d_main_controller +#define _jinit_d_coef_controller j12init_d_coef_controller +#define _jinit_d_post_controller j12init_d_post_controller +#define _jinit_inverse_dct j12init_inverse_dct +#define _jinit_upsampler j12init_upsampler +#define _jinit_color_deconverter j12init_color_deconverter +#define _jinit_1pass_quantizer j12init_1pass_quantizer +#define _jinit_2pass_quantizer j12init_2pass_quantizer +#define _jinit_merged_upsampler j12init_merged_upsampler +#ifdef D_LOSSLESS_SUPPORTED +#define _jinit_d_diff_controller j12init_d_diff_controller +#define _jinit_lossless_decompressor j12init_lossless_decompressor +#endif + +#define _jcopy_sample_rows j12copy_sample_rows + +/* Global internal functions (jdct.h) */ +#define _jpeg_fdct_islow jpeg12_fdct_islow +#define _jpeg_fdct_ifast jpeg12_fdct_ifast + +#define _jpeg_idct_islow jpeg12_idct_islow +#define _jpeg_idct_ifast jpeg12_idct_ifast +#define _jpeg_idct_float jpeg12_idct_float +#define _jpeg_idct_7x7 jpeg12_idct_7x7 +#define _jpeg_idct_6x6 jpeg12_idct_6x6 +#define _jpeg_idct_5x5 jpeg12_idct_5x5 +#define _jpeg_idct_4x4 jpeg12_idct_4x4 +#define _jpeg_idct_3x3 jpeg12_idct_3x3 +#define _jpeg_idct_2x2 jpeg12_idct_2x2 +#define _jpeg_idct_1x1 jpeg12_idct_1x1 +#define _jpeg_idct_9x9 jpeg12_idct_9x9 +#define _jpeg_idct_10x10 jpeg12_idct_10x10 +#define _jpeg_idct_11x11 jpeg12_idct_11x11 +#define _jpeg_idct_12x12 jpeg12_idct_12x12 +#define _jpeg_idct_13x13 jpeg12_idct_13x13 +#define _jpeg_idct_14x14 jpeg12_idct_14x14 +#define _jpeg_idct_15x15 jpeg12_idct_15x15 +#define _jpeg_idct_16x16 jpeg12_idct_16x16 + +/* Internal fields (cdjpeg.h) */ + +/* Use the 12-bit buffer in the cjpeg_source_struct and djpeg_dest_struct + structures. */ +#define _buffer buffer12 + +/* Image I/O functions (cdjpeg.h) */ +#define _jinit_read_gif j12init_read_gif +#define _jinit_write_gif j12init_write_gif +#define _jinit_read_ppm j12init_read_ppm +#define _jinit_write_ppm j12init_write_ppm + +#define _read_color_map read_color_map_12 + +#else /* BITS_IN_JSAMPLE */ + +/* Sample data types and macros (jmorecfg.h) */ +#define _JSAMPLE JSAMPLE + +#define _MAXJSAMPLE MAXJSAMPLE +#define _CENTERJSAMPLE CENTERJSAMPLE + +#define _JSAMPROW JSAMPROW +#define _JSAMPARRAY JSAMPARRAY +#define _JSAMPIMAGE JSAMPIMAGE + +/* External functions (jpeglib.h) */ +#define _jpeg_write_scanlines jpeg_write_scanlines +#define _jpeg_write_raw_data jpeg_write_raw_data +#define _jpeg_read_scanlines jpeg_read_scanlines +#define _jpeg_skip_scanlines jpeg_skip_scanlines +#define _jpeg_crop_scanline jpeg_crop_scanline +#define _jpeg_read_raw_data jpeg_read_raw_data + +/* Internal methods (jpegint.h) */ + +/* Use the 8-bit method in the jpeg_c_main_controller structure. */ +#define _process_data process_data +/* Use the 8-bit method in the jpeg_c_prep_controller structure. */ +#define _pre_process_data pre_process_data +/* Use the 8-bit method in the jpeg_c_coef_controller structure. */ +#define _compress_data compress_data +/* Use the 8-bit method in the jpeg_color_converter structure. */ +#define _color_convert color_convert +/* Use the 8-bit method in the jpeg_downsampler structure. */ +#define _downsample downsample +/* Use the 8-bit method in the jpeg_forward_dct structure. */ +#define _forward_DCT forward_DCT +/* Use the 8-bit method in the jpeg_d_main_controller structure. */ +#define _process_data process_data +/* Use the 8-bit method in the jpeg_d_coef_controller structure. */ +#define _decompress_data decompress_data +/* Use the 8-bit method in the jpeg_d_post_controller structure. */ +#define _post_process_data post_process_data +/* Use the 8-bit method in the jpeg_inverse_dct structure. */ +#define _inverse_DCT_method_ptr inverse_DCT_method_ptr +#define _inverse_DCT inverse_DCT +/* Use the 8-bit method in the jpeg_upsampler structure. */ +#define _upsample upsample +/* Use the 8-bit method in the jpeg_color_converter structure. */ +#define _color_convert color_convert +/* Use the 8-bit method in the jpeg_color_quantizer structure. */ +#define _color_quantize color_quantize + +/* Global internal functions (jpegint.h) */ +#define _jinit_c_main_controller jinit_c_main_controller +#define _jinit_c_prep_controller jinit_c_prep_controller +#define _jinit_c_coef_controller jinit_c_coef_controller +#define _jinit_color_converter jinit_color_converter +#define _jinit_downsampler jinit_downsampler +#define _jinit_forward_dct jinit_forward_dct +#ifdef C_LOSSLESS_SUPPORTED +#define _jinit_c_diff_controller jinit_c_diff_controller +#define _jinit_lossless_compressor jinit_lossless_compressor +#endif + +#define _jinit_d_main_controller jinit_d_main_controller +#define _jinit_d_coef_controller jinit_d_coef_controller +#define _jinit_d_post_controller jinit_d_post_controller +#define _jinit_inverse_dct jinit_inverse_dct +#define _jinit_upsampler jinit_upsampler +#define _jinit_color_deconverter jinit_color_deconverter +#define _jinit_1pass_quantizer jinit_1pass_quantizer +#define _jinit_2pass_quantizer jinit_2pass_quantizer +#define _jinit_merged_upsampler jinit_merged_upsampler +#ifdef D_LOSSLESS_SUPPORTED +#define _jinit_d_diff_controller jinit_d_diff_controller +#define _jinit_lossless_decompressor jinit_lossless_decompressor +#endif + +#define _jcopy_sample_rows jcopy_sample_rows + +/* Global internal functions (jdct.h) */ +#define _jpeg_fdct_islow jpeg_fdct_islow +#define _jpeg_fdct_ifast jpeg_fdct_ifast + +#define _jpeg_idct_islow jpeg_idct_islow +#define _jpeg_idct_ifast jpeg_idct_ifast +#define _jpeg_idct_float jpeg_idct_float +#define _jpeg_idct_7x7 jpeg_idct_7x7 +#define _jpeg_idct_6x6 jpeg_idct_6x6 +#define _jpeg_idct_5x5 jpeg_idct_5x5 +#define _jpeg_idct_4x4 jpeg_idct_4x4 +#define _jpeg_idct_3x3 jpeg_idct_3x3 +#define _jpeg_idct_2x2 jpeg_idct_2x2 +#define _jpeg_idct_1x1 jpeg_idct_1x1 +#define _jpeg_idct_9x9 jpeg_idct_9x9 +#define _jpeg_idct_10x10 jpeg_idct_10x10 +#define _jpeg_idct_11x11 jpeg_idct_11x11 +#define _jpeg_idct_12x12 jpeg_idct_12x12 +#define _jpeg_idct_13x13 jpeg_idct_13x13 +#define _jpeg_idct_14x14 jpeg_idct_14x14 +#define _jpeg_idct_15x15 jpeg_idct_15x15 +#define _jpeg_idct_16x16 jpeg_idct_16x16 + +/* Internal fields (cdjpeg.h) */ + +/* Use the 8-bit buffer in the cjpeg_source_struct and djpeg_dest_struct + structures. */ +#define _buffer buffer + +/* Image I/O functions (cdjpeg.h) */ +#define _jinit_read_gif jinit_read_gif +#define _jinit_write_gif jinit_write_gif +#define _jinit_read_ppm jinit_read_ppm +#define _jinit_write_ppm jinit_write_ppm + +#define _read_color_map read_color_map + +#endif /* BITS_IN_JSAMPLE */ + +#endif /* JSAMPLECOMP_H */ diff --git a/src/spice2x/external/libjpeg-turbo/jsimd.h b/src/spice2x/external/libjpeg-turbo/jsimd.h new file mode 100644 index 0000000..6ae021a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jsimd.h @@ -0,0 +1,127 @@ +/* + * jsimd.h + * + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2011, 2014, 2022, D. R. Commander. + * Copyright (C) 2015-2016, 2018, 2022, Matthieu Darbois. + * Copyright (C) 2020, Arm Limited. + * + * Based on the x86 SIMD extension for IJG JPEG library, + * Copyright (C) 1999-2006, MIYASAKA Masaru. + * For conditions of distribution and use, see copyright notice in jsimdext.inc + * + */ + +#ifdef WITH_SIMD + +#include "jchuff.h" /* Declarations shared with jcphuff.c */ + +EXTERN(int) jsimd_can_rgb_ycc(void); +EXTERN(int) jsimd_can_rgb_gray(void); +EXTERN(int) jsimd_can_ycc_rgb(void); +EXTERN(int) jsimd_can_ycc_rgb565(void); +EXTERN(int) jsimd_c_can_null_convert(void); + +EXTERN(void) jsimd_rgb_ycc_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_rgb_gray_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_ycc_rgb_convert(j_decompress_ptr cinfo, + JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_rgb565_convert(j_decompress_ptr cinfo, + JSAMPIMAGE input_buf, + JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_c_null_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows); + +EXTERN(int) jsimd_can_h2v2_downsample(void); +EXTERN(int) jsimd_can_h2v1_downsample(void); + +EXTERN(void) jsimd_h2v2_downsample(j_compress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY output_data); + +EXTERN(int) jsimd_can_h2v2_smooth_downsample(void); + +EXTERN(void) jsimd_h2v2_smooth_downsample(j_compress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v1_downsample(j_compress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY output_data); + +EXTERN(int) jsimd_can_h2v2_upsample(void); +EXTERN(int) jsimd_can_h2v1_upsample(void); +EXTERN(int) jsimd_can_int_upsample(void); + +EXTERN(void) jsimd_h2v2_upsample(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v1_upsample(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_int_upsample(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(int) jsimd_can_h2v2_fancy_upsample(void); +EXTERN(int) jsimd_can_h2v1_fancy_upsample(void); +EXTERN(int) jsimd_can_h1v2_fancy_upsample(void); + +EXTERN(void) jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h1v2_fancy_upsample(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(int) jsimd_can_h2v2_merged_upsample(void); +EXTERN(int) jsimd_can_h2v1_merged_upsample(void); + +EXTERN(void) jsimd_h2v2_merged_upsample(j_decompress_ptr cinfo, + JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_merged_upsample(j_decompress_ptr cinfo, + JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(int) jsimd_can_huff_encode_one_block(void); + +EXTERN(JOCTET *) jsimd_huff_encode_one_block(void *state, JOCTET *buffer, + JCOEFPTR block, int last_dc_val, + c_derived_tbl *dctbl, + c_derived_tbl *actbl); + +EXTERN(int) jsimd_can_encode_mcu_AC_first_prepare(void); + +EXTERN(void) jsimd_encode_mcu_AC_first_prepare + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *values, size_t *zerobits); + +EXTERN(int) jsimd_can_encode_mcu_AC_refine_prepare(void); + +EXTERN(int) jsimd_encode_mcu_AC_refine_prepare + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *absvalues, size_t *bits); + +#endif /* WITH_SIMD */ diff --git a/src/spice2x/external/libjpeg-turbo/jsimddct.h b/src/spice2x/external/libjpeg-turbo/jsimddct.h new file mode 100644 index 0000000..55ee8cf --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jsimddct.h @@ -0,0 +1,70 @@ +/* + * jsimddct.h + * + * Copyright 2009 Pierre Ossman for Cendio AB + * + * Based on the x86 SIMD extension for IJG JPEG library, + * Copyright (C) 1999-2006, MIYASAKA Masaru. + * For conditions of distribution and use, see copyright notice in jsimdext.inc + * + */ + +EXTERN(int) jsimd_can_convsamp(void); +EXTERN(int) jsimd_can_convsamp_float(void); + +EXTERN(void) jsimd_convsamp(JSAMPARRAY sample_data, JDIMENSION start_col, + DCTELEM *workspace); +EXTERN(void) jsimd_convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col, + FAST_FLOAT *workspace); + +EXTERN(int) jsimd_can_fdct_islow(void); +EXTERN(int) jsimd_can_fdct_ifast(void); +EXTERN(int) jsimd_can_fdct_float(void); + +EXTERN(void) jsimd_fdct_islow(DCTELEM *data); +EXTERN(void) jsimd_fdct_ifast(DCTELEM *data); +EXTERN(void) jsimd_fdct_float(FAST_FLOAT *data); + +EXTERN(int) jsimd_can_quantize(void); +EXTERN(int) jsimd_can_quantize_float(void); + +EXTERN(void) jsimd_quantize(JCOEFPTR coef_block, DCTELEM *divisors, + DCTELEM *workspace); +EXTERN(void) jsimd_quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors, + FAST_FLOAT *workspace); + +EXTERN(int) jsimd_can_idct_2x2(void); +EXTERN(int) jsimd_can_idct_4x4(void); +EXTERN(int) jsimd_can_idct_6x6(void); +EXTERN(int) jsimd_can_idct_12x12(void); + +EXTERN(void) jsimd_idct_2x2(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) jsimd_idct_4x4(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) jsimd_idct_6x6(j_decompress_ptr cinfo, + jpeg_component_info *compptr, JCOEFPTR coef_block, + JSAMPARRAY output_buf, JDIMENSION output_col); +EXTERN(void) jsimd_idct_12x12(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(int) jsimd_can_idct_islow(void); +EXTERN(int) jsimd_can_idct_ifast(void); +EXTERN(int) jsimd_can_idct_float(void); + +EXTERN(void) jsimd_idct_islow(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) jsimd_idct_ifast(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) jsimd_idct_float(j_decompress_ptr cinfo, + jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); diff --git a/src/spice2x/external/libjpeg-turbo/jstdhuff.c b/src/spice2x/external/libjpeg-turbo/jstdhuff.c new file mode 100644 index 0000000..345b513 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jstdhuff.c @@ -0,0 +1,144 @@ +/* + * jstdhuff.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1998, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2013, 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains routines to set the default Huffman tables, if they are + * not already set. + */ + +/* + * Huffman table setup routines + */ + +LOCAL(void) +add_huff_table(j_common_ptr cinfo, JHUFF_TBL **htblptr, const UINT8 *bits, + const UINT8 *val) +/* Define a Huffman table */ +{ + int nsymbols, len; + + if (*htblptr == NULL) + *htblptr = jpeg_alloc_huff_table(cinfo); + else + return; + + /* Copy the number-of-symbols-of-each-code-length counts */ + memcpy((*htblptr)->bits, bits, sizeof((*htblptr)->bits)); + + /* Validate the counts. We do this here mainly so we can copy the right + * number of symbols from the val[] array, without risking marching off + * the end of memory. jchuff.c will do a more thorough test later. + */ + nsymbols = 0; + for (len = 1; len <= 16; len++) + nsymbols += bits[len]; + if (nsymbols < 1 || nsymbols > 256) + ERREXIT(cinfo, JERR_BAD_HUFF_TABLE); + + memcpy((*htblptr)->huffval, val, nsymbols * sizeof(UINT8)); + memset(&((*htblptr)->huffval[nsymbols]), 0, + (256 - nsymbols) * sizeof(UINT8)); + + /* Initialize sent_table FALSE so table will be written to JPEG file. */ + (*htblptr)->sent_table = FALSE; +} + + +LOCAL(void) +std_huff_tables(j_common_ptr cinfo) +/* Set up the standard Huffman tables (cf. JPEG standard section K.3) */ +/* IMPORTANT: these are only valid for 8-bit data precision! */ +{ + JHUFF_TBL **dc_huff_tbl_ptrs, **ac_huff_tbl_ptrs; + + static const UINT8 bits_dc_luminance[17] = { + /* 0-base */ 0, 0, 1, 5, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0 + }; + static const UINT8 val_dc_luminance[] = { + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 + }; + + static const UINT8 bits_dc_chrominance[17] = { + /* 0-base */ 0, 0, 3, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0 + }; + static const UINT8 val_dc_chrominance[] = { + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 + }; + + static const UINT8 bits_ac_luminance[17] = { + /* 0-base */ 0, 0, 2, 1, 3, 3, 2, 4, 3, 5, 5, 4, 4, 0, 0, 1, 0x7d + }; + static const UINT8 val_ac_luminance[] = { + 0x01, 0x02, 0x03, 0x00, 0x04, 0x11, 0x05, 0x12, + 0x21, 0x31, 0x41, 0x06, 0x13, 0x51, 0x61, 0x07, + 0x22, 0x71, 0x14, 0x32, 0x81, 0x91, 0xa1, 0x08, + 0x23, 0x42, 0xb1, 0xc1, 0x15, 0x52, 0xd1, 0xf0, + 0x24, 0x33, 0x62, 0x72, 0x82, 0x09, 0x0a, 0x16, + 0x17, 0x18, 0x19, 0x1a, 0x25, 0x26, 0x27, 0x28, + 0x29, 0x2a, 0x34, 0x35, 0x36, 0x37, 0x38, 0x39, + 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, + 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, + 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, 0x69, + 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, 0x79, + 0x7a, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, + 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, + 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, 0xa6, 0xa7, + 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, 0xb5, 0xb6, + 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, 0xc4, 0xc5, + 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, 0xd3, 0xd4, + 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, 0xe1, 0xe2, + 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, 0xea, + 0xf1, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, + 0xf9, 0xfa + }; + + static const UINT8 bits_ac_chrominance[17] = { + /* 0-base */ 0, 0, 2, 1, 2, 4, 4, 3, 4, 7, 5, 4, 4, 0, 1, 2, 0x77 + }; + static const UINT8 val_ac_chrominance[] = { + 0x00, 0x01, 0x02, 0x03, 0x11, 0x04, 0x05, 0x21, + 0x31, 0x06, 0x12, 0x41, 0x51, 0x07, 0x61, 0x71, + 0x13, 0x22, 0x32, 0x81, 0x08, 0x14, 0x42, 0x91, + 0xa1, 0xb1, 0xc1, 0x09, 0x23, 0x33, 0x52, 0xf0, + 0x15, 0x62, 0x72, 0xd1, 0x0a, 0x16, 0x24, 0x34, + 0xe1, 0x25, 0xf1, 0x17, 0x18, 0x19, 0x1a, 0x26, + 0x27, 0x28, 0x29, 0x2a, 0x35, 0x36, 0x37, 0x38, + 0x39, 0x3a, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, + 0x49, 0x4a, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, + 0x59, 0x5a, 0x63, 0x64, 0x65, 0x66, 0x67, 0x68, + 0x69, 0x6a, 0x73, 0x74, 0x75, 0x76, 0x77, 0x78, + 0x79, 0x7a, 0x82, 0x83, 0x84, 0x85, 0x86, 0x87, + 0x88, 0x89, 0x8a, 0x92, 0x93, 0x94, 0x95, 0x96, + 0x97, 0x98, 0x99, 0x9a, 0xa2, 0xa3, 0xa4, 0xa5, + 0xa6, 0xa7, 0xa8, 0xa9, 0xaa, 0xb2, 0xb3, 0xb4, + 0xb5, 0xb6, 0xb7, 0xb8, 0xb9, 0xba, 0xc2, 0xc3, + 0xc4, 0xc5, 0xc6, 0xc7, 0xc8, 0xc9, 0xca, 0xd2, + 0xd3, 0xd4, 0xd5, 0xd6, 0xd7, 0xd8, 0xd9, 0xda, + 0xe2, 0xe3, 0xe4, 0xe5, 0xe6, 0xe7, 0xe8, 0xe9, + 0xea, 0xf2, 0xf3, 0xf4, 0xf5, 0xf6, 0xf7, 0xf8, + 0xf9, 0xfa + }; + + if (cinfo->is_decompressor) { + dc_huff_tbl_ptrs = ((j_decompress_ptr)cinfo)->dc_huff_tbl_ptrs; + ac_huff_tbl_ptrs = ((j_decompress_ptr)cinfo)->ac_huff_tbl_ptrs; + } else { + dc_huff_tbl_ptrs = ((j_compress_ptr)cinfo)->dc_huff_tbl_ptrs; + ac_huff_tbl_ptrs = ((j_compress_ptr)cinfo)->ac_huff_tbl_ptrs; + } + + add_huff_table(cinfo, &dc_huff_tbl_ptrs[0], bits_dc_luminance, + val_dc_luminance); + add_huff_table(cinfo, &ac_huff_tbl_ptrs[0], bits_ac_luminance, + val_ac_luminance); + add_huff_table(cinfo, &dc_huff_tbl_ptrs[1], bits_dc_chrominance, + val_dc_chrominance); + add_huff_table(cinfo, &ac_huff_tbl_ptrs[1], bits_ac_chrominance, + val_ac_chrominance); +} diff --git a/src/spice2x/external/libjpeg-turbo/jutils.c b/src/spice2x/external/libjpeg-turbo/jutils.c new file mode 100644 index 0000000..24caac1 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jutils.c @@ -0,0 +1,148 @@ +/* + * jutils.c + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-1996, Thomas G. Lane. + * libjpeg-turbo Modifications: + * Copyright (C) 2022, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains tables and miscellaneous utility routines needed + * for both compression and decompression. + * Note we prefix all global names with "j" to minimize conflicts with + * a surrounding application. + */ + +#define JPEG_INTERNALS +#include "jinclude.h" +#include "jpeglib.h" +#include "jsamplecomp.h" + + +#if BITS_IN_JSAMPLE == 8 + +/* + * jpeg_zigzag_order[i] is the zigzag-order position of the i'th element + * of a DCT block read in natural order (left to right, top to bottom). + */ + +#if 0 /* This table is not actually needed in v6a */ + +const int jpeg_zigzag_order[DCTSIZE2] = { + 0, 1, 5, 6, 14, 15, 27, 28, + 2, 4, 7, 13, 16, 26, 29, 42, + 3, 8, 12, 17, 25, 30, 41, 43, + 9, 11, 18, 24, 31, 40, 44, 53, + 10, 19, 23, 32, 39, 45, 52, 54, + 20, 22, 33, 38, 46, 51, 55, 60, + 21, 34, 37, 47, 50, 56, 59, 61, + 35, 36, 48, 49, 57, 58, 62, 63 +}; + +#endif + +/* + * jpeg_natural_order[i] is the natural-order position of the i'th element + * of zigzag order. + * + * When reading corrupted data, the Huffman decoders could attempt + * to reference an entry beyond the end of this array (if the decoded + * zero run length reaches past the end of the block). To prevent + * wild stores without adding an inner-loop test, we put some extra + * "63"s after the real entries. This will cause the extra coefficient + * to be stored in location 63 of the block, not somewhere random. + * The worst case would be a run-length of 15, which means we need 16 + * fake entries. + */ + +const int jpeg_natural_order[DCTSIZE2 + 16] = { + 0, 1, 8, 16, 9, 2, 3, 10, + 17, 24, 32, 25, 18, 11, 4, 5, + 12, 19, 26, 33, 40, 48, 41, 34, + 27, 20, 13, 6, 7, 14, 21, 28, + 35, 42, 49, 56, 57, 50, 43, 36, + 29, 22, 15, 23, 30, 37, 44, 51, + 58, 59, 52, 45, 38, 31, 39, 46, + 53, 60, 61, 54, 47, 55, 62, 63, + 63, 63, 63, 63, 63, 63, 63, 63, /* extra entries for safety in decoder */ + 63, 63, 63, 63, 63, 63, 63, 63 +}; + + +/* + * Arithmetic utilities + */ + +GLOBAL(long) +jdiv_round_up(long a, long b) +/* Compute a/b rounded up to next integer, ie, ceil(a/b) */ +/* Assumes a >= 0, b > 0 */ +{ + return (a + b - 1L) / b; +} + + +GLOBAL(long) +jround_up(long a, long b) +/* Compute a rounded up to next multiple of b, ie, ceil(a/b)*b */ +/* Assumes a >= 0, b > 0 */ +{ + a += b - 1L; + return a - (a % b); +} + +#endif /* BITS_IN_JSAMPLE == 8 */ + + +#if BITS_IN_JSAMPLE != 16 || \ + defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) + +GLOBAL(void) +_jcopy_sample_rows(_JSAMPARRAY input_array, int source_row, + _JSAMPARRAY output_array, int dest_row, int num_rows, + JDIMENSION num_cols) +/* Copy some rows of samples from one place to another. + * num_rows rows are copied from input_array[source_row++] + * to output_array[dest_row++]; these areas may overlap for duplication. + * The source and destination arrays must be at least as wide as num_cols. + */ +{ + register _JSAMPROW inptr, outptr; + register size_t count = (size_t)(num_cols * sizeof(_JSAMPLE)); + register int row; + + input_array += source_row; + output_array += dest_row; + + for (row = num_rows; row > 0; row--) { + inptr = *input_array++; + outptr = *output_array++; + memcpy(outptr, inptr, count); + } +} + +#endif /* BITS_IN_JSAMPLE != 16 || + defined(C_LOSSLESS_SUPPORTED) || defined(D_LOSSLESS_SUPPORTED) */ + + +#if BITS_IN_JSAMPLE == 8 + +GLOBAL(void) +jcopy_block_row(JBLOCKROW input_row, JBLOCKROW output_row, + JDIMENSION num_blocks) +/* Copy a row of coefficient blocks from one place to another. */ +{ + memcpy(output_row, input_row, num_blocks * (DCTSIZE2 * sizeof(JCOEF))); +} + + +GLOBAL(void) +jzero_far(void *target, size_t bytestozero) +/* Zero out a chunk of memory. */ +/* This might be sample-array data, block-array data, or alloc_large data. */ +{ + memset(target, 0, bytestozero); +} + +#endif /* BITS_IN_JSAMPLE == 8 */ diff --git a/src/spice2x/external/libjpeg-turbo/jversion.h.in b/src/spice2x/external/libjpeg-turbo/jversion.h.in new file mode 100644 index 0000000..fc0ce3e --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/jversion.h.in @@ -0,0 +1,56 @@ +/* + * jversion.h + * + * This file was part of the Independent JPEG Group's software: + * Copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding. + * libjpeg-turbo Modifications: + * Copyright (C) 2010, 2012-2024, D. R. Commander. + * For conditions of distribution and use, see the accompanying README.ijg + * file. + * + * This file contains software version identification. + */ + + +#if JPEG_LIB_VERSION >= 80 + +#define JVERSION "8d 15-Jan-2012" + +#elif JPEG_LIB_VERSION >= 70 + +#define JVERSION "7 27-Jun-2009" + +#else + +#define JVERSION "6b 27-Mar-1998" + +#endif + +/* + * NOTE: It is our convention to place the authors in the following order: + * - libjpeg-turbo authors (2009-) in descending order of the date of their + * most recent contribution to the project, then in ascending order of the + * date of their first contribution to the project, then in alphabetical + * order + * - Upstream authors in descending order of the date of the first inclusion of + * their code + */ + +#define JCOPYRIGHT1 \ + "Copyright (C) 2009-2024 D. R. Commander\n" \ + "Copyright (C) 2015, 2020 Google, Inc.\n" \ + "Copyright (C) 2019-2020 Arm Limited\n" \ + "Copyright (C) 2015-2016, 2018 Matthieu Darbois\n" \ + "Copyright (C) 2011-2016 Siarhei Siamashka\n" \ + "Copyright (C) 2015 Intel Corporation\n" +#define JCOPYRIGHT2 \ + "Copyright (C) 2013-2014 Linaro Limited\n" \ + "Copyright (C) 2013-2014 MIPS Technologies, Inc.\n" \ + "Copyright (C) 2009, 2012 Pierre Ossman for Cendio AB\n" \ + "Copyright (C) 2009-2011 Nokia Corporation and/or its subsidiary(-ies)\n" \ + "Copyright (C) 1999-2006 MIYASAKA Masaru\n" \ + "Copyright (C) 1999 Ken Murchison\n" \ + "Copyright (C) 1991-2020 Thomas G. Lane, Guido Vollbeding\n" + +#define JCOPYRIGHT_SHORT \ + "Copyright (C) @COPYRIGHT_YEAR@ The libjpeg-turbo Project and many others" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-avx2.asm new file mode 100644 index 0000000..28ac952 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-avx2.asm @@ -0,0 +1,574 @@ +; +; jccolext.asm - colorspace conversion (AVX2) +; +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_ycc_convert_avx2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +%define img_width(b) (b) + 8 ; JDIMENSION img_width +%define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf +%define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf +%define output_row(b) (b) + 20 ; JDIMENSION output_row +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_YMMWORD + ; ymmword wk[WK_NUM] +%define WK_NUM 8 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_ycc_convert_avx2) + +EXTN(jsimd_rgb_ycc_convert_avx2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [img_width(eax)] + test ecx, ecx + jz near .return + + push ecx + + mov esi, JSAMPIMAGE [output_buf(eax)] + mov ecx, JDIMENSION [output_row(eax)] + mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY] + lea edi, [edi+ecx*SIZEOF_JSAMPROW] + lea ebx, [ebx+ecx*SIZEOF_JSAMPROW] + lea edx, [edx+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov esi, JSAMPARRAY [input_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + PUSHPIC eax + push edx + push ebx + push edi + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr0 + mov ebx, JSAMPROW [ebx] ; outptr1 + mov edx, JSAMPROW [edx] ; outptr2 + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + cmp ecx, byte SIZEOF_YMMWORD + jae near .columnloop + ALIGNX 16, 7 + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push eax + push edx + lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub ecx, byte SIZEOF_BYTE + movzx eax, byte [esi+ecx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub ecx, byte SIZEOF_WORD + movzx edx, word [esi+ecx] + shl eax, WORD_BIT + or eax, edx +.column_ld4: + vmovd xmmA, eax + pop edx + pop eax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub ecx, byte SIZEOF_DWORD + vmovd xmmF, XMM_DWORD [esi+ecx] + vpslldq xmmA, xmmA, SIZEOF_DWORD + vpor xmmA, xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub ecx, byte SIZEOF_MMWORD + vmovq xmmB, XMM_MMWORD [esi+ecx] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + sub ecx, byte SIZEOF_XMMWORD + vmovdqu xmmB, XMM_MMWORD [esi+ecx] + vperm2i128 ymmA, ymmA, ymmA, 1 + vpor ymmA, ymmB +.column_ld32: + test cl, SIZEOF_YMMWORD + jz short .column_ld64 + sub ecx, byte SIZEOF_YMMWORD + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] +.column_ld64: + test cl, 2*SIZEOF_YMMWORD + mov ecx, SIZEOF_YMMWORD + jz short .rgb_ycc_cnv + vmovdqa ymmB, ymmA + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + jmp short .rgb_ycc_cnv + ALIGNX 16, 7 + +.columnloop: + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + vmovdqu ymmB, YMMWORD [esi+2*SIZEOF_YMMWORD] + +.rgb_ycc_cnv: + ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vmovdqu ymmC, ymmA + vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + + vmovdqa ymmG, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12 + ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I) + vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I + ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A + ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q) + vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27 + ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N) + + vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D + ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T) + vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F + ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V) + + vmovdqa ymmD, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09 + ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P) + vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P + ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T) + vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B + ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R) + + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E + ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F + ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V) + + vmovdqa ymmE, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C + ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S) + vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S + ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D + ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T) + + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F + ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F + ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V) + + vpxor ymmH, ymmH, ymmH + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmB, ymmE + vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + + vmovdqa ymmF, ymmD + vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub ecx, byte SIZEOF_XMMWORD/16 + vmovd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub ecx, byte SIZEOF_XMMWORD/8 + vmovq xmmF, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmF +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub ecx, byte SIZEOF_XMMWORD/4 + vmovdqa xmmF, xmmA + vperm2i128 ymmF, ymmF, ymmF, 1 + vmovdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE] + vpor ymmA, ymmA, ymmF +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + jz short .column_ld16 + sub ecx, byte SIZEOF_XMMWORD/2 + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld16: + test cl, SIZEOF_XMMWORD + mov ecx, SIZEOF_YMMWORD + jz short .rgb_ycc_cnv + vmovdqa ymmE, ymmA + vmovdqa ymmH, ymmF + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + jmp short .rgb_ycc_cnv + ALIGNX 16, 7 + +.columnloop: + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + vmovdqu ymmE, YMMWORD [esi+2*SIZEOF_YMMWORD] + vmovdqu ymmH, YMMWORD [esi+3*SIZEOF_YMMWORD] + +.rgb_ycc_cnv: + ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmB, ymmA + vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + + vmovdqa ymmB, ymmF + vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmD, ymmA + vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35 + ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L) + vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37 + ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N) + + vmovdqa ymmC, ymmF + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D + ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F + ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V) + + vmovdqa ymmB, ymmA + vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S) + vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D + ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T) + + vmovdqa ymmG, ymmD + vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E + ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U) + vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F + ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V) + + vmovdqa ymmE, ymmA + vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E + ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U) + + vmovdqa ymmH, ymmB + vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F + ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F + ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V) + + vpxor ymmF, ymmF, ymmF + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmD, ymmB + vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + + vmovdqa ymmG, ymmE + vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U) + + vpunpcklbw ymmF, ymmF, ymmH + vpunpckhbw ymmH, ymmH, ymmH + vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE + ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + + vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=RE + vmovdqa YMMWORD [wk(1)], ymm1 ; wk(1)=RO + vmovdqa YMMWORD [wk(2)], ymm4 ; wk(2)=BE + vmovdqa YMMWORD [wk(3)], ymm5 ; wk(3)=BO + + vmovdqa ymm6, ymm1 + vpunpcklwd ymm1, ymm1, ymm3 + vpunpckhwd ymm6, ymm6, ymm3 + vmovdqa ymm7, ymm1 + vmovdqa ymm4, ymm6 + vpmaddwd ymm1, ymm1, [GOTOFF(eax,PW_F0299_F0337)] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337) + vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337) + vpmaddwd ymm7, ymm7, [GOTOFF(eax,PW_MF016_MF033)] ; ymm7=ROL*-FIX(0.168)+GOL*-FIX(0.331) + vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_MF016_MF033)] ; ymm4=ROH*-FIX(0.168)+GOH*-FIX(0.331) + + vmovdqa YMMWORD [wk(4)], ymm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337) + vmovdqa YMMWORD [wk(5)], ymm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337) + + vpxor ymm1, ymm1, ymm1 + vpxor ymm6, ymm6, ymm6 + vpunpcklwd ymm1, ymm1, ymm5 ; ymm1=BOL + vpunpckhwd ymm6, ymm6, ymm5 ; ymm6=BOH + vpsrld ymm1, ymm1, 1 ; ymm1=BOL*FIX(0.500) + vpsrld ymm6, ymm6, 1 ; ymm6=BOH*FIX(0.500) + + vmovdqa ymm5, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm5=[PD_ONEHALFM1_CJ] + + vpaddd ymm7, ymm7, ymm1 + vpaddd ymm4, ymm4, ymm6 + vpaddd ymm7, ymm7, ymm5 + vpaddd ymm4, ymm4, ymm5 + vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CbOL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbOH + vpackssdw ymm7, ymm7, ymm4 ; ymm7=CbO + + vmovdqa ymm1, YMMWORD [wk(2)] ; ymm1=BE + + vmovdqa ymm6, ymm0 + vpunpcklwd ymm0, ymm0, ymm2 + vpunpckhwd ymm6, ymm6, ymm2 + vmovdqa ymm5, ymm0 + vmovdqa ymm4, ymm6 + vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0299_F0337)] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337) + vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337) + vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF016_MF033)] ; ymm5=REL*-FIX(0.168)+GEL*-FIX(0.331) + vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_MF016_MF033)] ; ymm4=REH*-FIX(0.168)+GEH*-FIX(0.331) + + vmovdqa YMMWORD [wk(6)], ymm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337) + vmovdqa YMMWORD [wk(7)], ymm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337) + + vpxor ymm0, ymm0, ymm0 + vpxor ymm6, ymm6, ymm6 + vpunpcklwd ymm0, ymm0, ymm1 ; ymm0=BEL + vpunpckhwd ymm6, ymm6, ymm1 ; ymm6=BEH + vpsrld ymm0, ymm0, 1 ; ymm0=BEL*FIX(0.500) + vpsrld ymm6, ymm6, 1 ; ymm6=BEH*FIX(0.500) + + vmovdqa ymm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm1=[PD_ONEHALFM1_CJ] + + vpaddd ymm5, ymm5, ymm0 + vpaddd ymm4, ymm4, ymm6 + vpaddd ymm5, ymm5, ymm1 + vpaddd ymm4, ymm4, ymm1 + vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CbEL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbEH + vpackssdw ymm5, ymm5, ymm4 ; ymm5=CbE + + vpsllw ymm7, ymm7, BYTE_BIT + vpor ymm5, ymm5, ymm7 ; ymm5=Cb + vmovdqu YMMWORD [ebx], ymm5 ; Save Cb + + vmovdqa ymm0, YMMWORD [wk(3)] ; ymm0=BO + vmovdqa ymm6, YMMWORD [wk(2)] ; ymm6=BE + vmovdqa ymm1, YMMWORD [wk(1)] ; ymm1=RO + + vmovdqa ymm4, ymm0 + vpunpcklwd ymm0, ymm0, ymm3 + vpunpckhwd ymm4, ymm4, ymm3 + vmovdqa ymm7, ymm0 + vmovdqa ymm5, ymm4 + vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0114_F0250)] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250) + vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250) + vpmaddwd ymm7, ymm7, [GOTOFF(eax,PW_MF008_MF041)] ; ymm7=BOL*-FIX(0.081)+GOL*-FIX(0.418) + vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF008_MF041)] ; ymm5=BOH*-FIX(0.081)+GOH*-FIX(0.418) + + vmovdqa ymm3, [GOTOFF(eax,PD_ONEHALF)] ; ymm3=[PD_ONEHALF] + + vpaddd ymm0, ymm0, YMMWORD [wk(4)] + vpaddd ymm4, ymm4, YMMWORD [wk(5)] + vpaddd ymm0, ymm0, ymm3 + vpaddd ymm4, ymm4, ymm3 + vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH + vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO + + vpxor ymm3, ymm3, ymm3 + vpxor ymm4, ymm4, ymm4 + vpunpcklwd ymm3, ymm3, ymm1 ; ymm3=ROL + vpunpckhwd ymm4, ymm4, ymm1 ; ymm4=ROH + vpsrld ymm3, ymm3, 1 ; ymm3=ROL*FIX(0.500) + vpsrld ymm4, ymm4, 1 ; ymm4=ROH*FIX(0.500) + + vmovdqa ymm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm1=[PD_ONEHALFM1_CJ] + + vpaddd ymm7, ymm7, ymm3 + vpaddd ymm5, ymm5, ymm4 + vpaddd ymm7, ymm7, ymm1 + vpaddd ymm5, ymm5, ymm1 + vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CrOL + vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrOH + vpackssdw ymm7, ymm7, ymm5 ; ymm7=CrO + + vmovdqa ymm3, YMMWORD [wk(0)] ; ymm3=RE + + vmovdqa ymm4, ymm6 + vpunpcklwd ymm6, ymm6, ymm2 + vpunpckhwd ymm4, ymm4, ymm2 + vmovdqa ymm1, ymm6 + vmovdqa ymm5, ymm4 + vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0114_F0250)] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250) + vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250) + vpmaddwd ymm1, ymm1, [GOTOFF(eax,PW_MF008_MF041)] ; ymm1=BEL*-FIX(0.081)+GEL*-FIX(0.418) + vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF008_MF041)] ; ymm5=BEH*-FIX(0.081)+GEH*-FIX(0.418) + + vmovdqa ymm2, [GOTOFF(eax,PD_ONEHALF)] ; ymm2=[PD_ONEHALF] + + vpaddd ymm6, ymm6, YMMWORD [wk(6)] + vpaddd ymm4, ymm4, YMMWORD [wk(7)] + vpaddd ymm6, ymm6, ymm2 + vpaddd ymm4, ymm4, ymm2 + vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH + vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE + + vpsllw ymm0, ymm0, BYTE_BIT + vpor ymm6, ymm6, ymm0 ; ymm6=Y + vmovdqu YMMWORD [edi], ymm6 ; Save Y + + vpxor ymm2, ymm2, ymm2 + vpxor ymm4, ymm4, ymm4 + vpunpcklwd ymm2, ymm2, ymm3 ; ymm2=REL + vpunpckhwd ymm4, ymm4, ymm3 ; ymm4=REH + vpsrld ymm2, ymm2, 1 ; ymm2=REL*FIX(0.500) + vpsrld ymm4, ymm4, 1 ; ymm4=REH*FIX(0.500) + + vmovdqa ymm0, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; ymm0=[PD_ONEHALFM1_CJ] + + vpaddd ymm1, ymm1, ymm2 + vpaddd ymm5, ymm5, ymm4 + vpaddd ymm1, ymm1, ymm0 + vpaddd ymm5, ymm5, ymm0 + vpsrld ymm1, ymm1, SCALEBITS ; ymm1=CrEL + vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrEH + vpackssdw ymm1, ymm1, ymm5 ; ymm1=CrE + + vpsllw ymm7, ymm7, BYTE_BIT + vpor ymm1, ymm1, ymm7 ; ymm1=Cr + vmovdqu YMMWORD [edx], ymm1 ; Save Cr + + sub ecx, byte SIZEOF_YMMWORD + add esi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr + add edi, byte SIZEOF_YMMWORD ; outptr0 + add ebx, byte SIZEOF_YMMWORD ; outptr1 + add edx, byte SIZEOF_YMMWORD ; outptr2 + cmp ecx, byte SIZEOF_YMMWORD + jae near .columnloop + test ecx, ecx + jnz near .column_ld1 + + pop ecx ; col + pop esi + pop edi + pop ebx + pop edx + POPPIC eax + + add esi, byte SIZEOF_JSAMPROW ; input_buf + add edi, byte SIZEOF_JSAMPROW + add ebx, byte SIZEOF_JSAMPROW + add edx, byte SIZEOF_JSAMPROW + dec eax ; num_rows + jg near .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-mmx.asm new file mode 100644 index 0000000..44b6251 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-mmx.asm @@ -0,0 +1,472 @@ +; +; jccolext.asm - colorspace conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_ycc_convert_mmx(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +%define img_width(b) (b) + 8 ; JDIMENSION img_width +%define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf +%define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf +%define output_row(b) (b) + 20 ; JDIMENSION output_row +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD + ; mmword wk[WK_NUM] +%define WK_NUM 8 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_ycc_convert_mmx) + +EXTN(jsimd_rgb_ycc_convert_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [img_width(eax)] ; num_cols + test ecx, ecx + jz near .return + + push ecx + + mov esi, JSAMPIMAGE [output_buf(eax)] + mov ecx, JDIMENSION [output_row(eax)] + mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY] + lea edi, [edi+ecx*SIZEOF_JSAMPROW] + lea ebx, [ebx+ecx*SIZEOF_JSAMPROW] + lea edx, [edx+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov esi, JSAMPARRAY [input_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + PUSHPIC eax + push edx + push ebx + push edi + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr0 + mov ebx, JSAMPROW [ebx] ; outptr1 + mov edx, JSAMPROW [edx] ; outptr2 + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + cmp ecx, byte SIZEOF_MMWORD + jae short .columnloop + ALIGNX 16, 7 + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push eax + push edx + lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub ecx, byte SIZEOF_BYTE + xor eax, eax + mov al, byte [esi+ecx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub ecx, byte SIZEOF_WORD + xor edx, edx + mov dx, word [esi+ecx] + shl eax, WORD_BIT + or eax, edx +.column_ld4: + movd mmA, eax + pop edx + pop eax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub ecx, byte SIZEOF_DWORD + movd mmG, dword [esi+ecx] + psllq mmA, DWORD_BIT + por mmA, mmG +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + movq mmG, mmA + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + mov ecx, SIZEOF_MMWORD + jmp short .rgb_ycc_cnv +.column_ld16: + test cl, 2*SIZEOF_MMWORD + mov ecx, SIZEOF_MMWORD + jz short .rgb_ycc_cnv + movq mmF, mmA + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmG, MMWORD [esi+1*SIZEOF_MMWORD] + jmp short .rgb_ycc_cnv + ALIGNX 16, 7 + +.columnloop: + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmG, MMWORD [esi+1*SIZEOF_MMWORD] + movq mmF, MMWORD [esi+2*SIZEOF_MMWORD] + +.rgb_ycc_cnv: + ; mmA=(00 10 20 01 11 21 02 12) + ; mmG=(22 03 13 23 04 14 24 05) + ; mmF=(15 25 06 16 26 07 17 27) + + movq mmD, mmA + psllq mmA, 4*BYTE_BIT ; mmA=(-- -- -- -- 00 10 20 01) + psrlq mmD, 4*BYTE_BIT ; mmD=(11 21 02 12 -- -- -- --) + + punpckhbw mmA, mmG ; mmA=(00 04 10 14 20 24 01 05) + psllq mmG, 4*BYTE_BIT ; mmG=(-- -- -- -- 22 03 13 23) + + punpcklbw mmD, mmF ; mmD=(11 15 21 25 02 06 12 16) + punpckhbw mmG, mmF ; mmG=(22 26 03 07 13 17 23 27) + + movq mmE, mmA + psllq mmA, 4*BYTE_BIT ; mmA=(-- -- -- -- 00 04 10 14) + psrlq mmE, 4*BYTE_BIT ; mmE=(20 24 01 05 -- -- -- --) + + punpckhbw mmA, mmD ; mmA=(00 02 04 06 10 12 14 16) + psllq mmD, 4*BYTE_BIT ; mmD=(-- -- -- -- 11 15 21 25) + + punpcklbw mmE, mmG ; mmE=(20 22 24 26 01 03 05 07) + punpckhbw mmD, mmG ; mmD=(11 13 15 17 21 23 25 27) + + pxor mmH, mmH + + movq mmC, mmA + punpcklbw mmA, mmH ; mmA=(00 02 04 06) + punpckhbw mmC, mmH ; mmC=(10 12 14 16) + + movq mmB, mmE + punpcklbw mmE, mmH ; mmE=(20 22 24 26) + punpckhbw mmB, mmH ; mmB=(01 03 05 07) + + movq mmF, mmD + punpcklbw mmD, mmH ; mmD=(11 13 15 17) + punpckhbw mmF, mmH ; mmF=(21 23 25 27) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_MMWORD/8 + jz short .column_ld2 + sub ecx, byte SIZEOF_MMWORD/8 + movd mmA, dword [esi+ecx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_MMWORD/4 + jz short .column_ld4 + sub ecx, byte SIZEOF_MMWORD/4 + movq mmF, mmA + movq mmA, MMWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld4: + test cl, SIZEOF_MMWORD/2 + mov ecx, SIZEOF_MMWORD + jz short .rgb_ycc_cnv + movq mmD, mmA + movq mmC, mmF + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmF, MMWORD [esi+1*SIZEOF_MMWORD] + jmp short .rgb_ycc_cnv + ALIGNX 16, 7 + +.columnloop: + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmF, MMWORD [esi+1*SIZEOF_MMWORD] + movq mmD, MMWORD [esi+2*SIZEOF_MMWORD] + movq mmC, MMWORD [esi+3*SIZEOF_MMWORD] + +.rgb_ycc_cnv: + ; mmA=(00 10 20 30 01 11 21 31) + ; mmF=(02 12 22 32 03 13 23 33) + ; mmD=(04 14 24 34 05 15 25 35) + ; mmC=(06 16 26 36 07 17 27 37) + + movq mmB, mmA + punpcklbw mmA, mmF ; mmA=(00 02 10 12 20 22 30 32) + punpckhbw mmB, mmF ; mmB=(01 03 11 13 21 23 31 33) + + movq mmG, mmD + punpcklbw mmD, mmC ; mmD=(04 06 14 16 24 26 34 36) + punpckhbw mmG, mmC ; mmG=(05 07 15 17 25 27 35 37) + + movq mmE, mmA + punpcklwd mmA, mmD ; mmA=(00 02 04 06 10 12 14 16) + punpckhwd mmE, mmD ; mmE=(20 22 24 26 30 32 34 36) + + movq mmH, mmB + punpcklwd mmB, mmG ; mmB=(01 03 05 07 11 13 15 17) + punpckhwd mmH, mmG ; mmH=(21 23 25 27 31 33 35 37) + + pxor mmF, mmF + + movq mmC, mmA + punpcklbw mmA, mmF ; mmA=(00 02 04 06) + punpckhbw mmC, mmF ; mmC=(10 12 14 16) + + movq mmD, mmB + punpcklbw mmB, mmF ; mmB=(01 03 05 07) + punpckhbw mmD, mmF ; mmD=(11 13 15 17) + + movq mmG, mmE + punpcklbw mmE, mmF ; mmE=(20 22 24 26) + punpckhbw mmG, mmF ; mmG=(30 32 34 36) + + punpcklbw mmF, mmH + punpckhbw mmH, mmH + psrlw mmF, BYTE_BIT ; mmF=(21 23 25 27) + psrlw mmH, BYTE_BIT ; mmH=(31 33 35 37) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; mm0=(R0 R2 R4 R6)=RE, mm2=(G0 G2 G4 G6)=GE, mm4=(B0 B2 B4 B6)=BE + ; mm1=(R1 R3 R5 R7)=RO, mm3=(G1 G3 G5 G7)=GO, mm5=(B1 B3 B5 B7)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + + movq MMWORD [wk(0)], mm0 ; wk(0)=RE + movq MMWORD [wk(1)], mm1 ; wk(1)=RO + movq MMWORD [wk(2)], mm4 ; wk(2)=BE + movq MMWORD [wk(3)], mm5 ; wk(3)=BO + + movq mm6, mm1 + punpcklwd mm1, mm3 + punpckhwd mm6, mm3 + movq mm7, mm1 + movq mm4, mm6 + pmaddwd mm1, [GOTOFF(eax,PW_F0299_F0337)] ; mm1=ROL*FIX(0.299)+GOL*FIX(0.337) + pmaddwd mm6, [GOTOFF(eax,PW_F0299_F0337)] ; mm6=ROH*FIX(0.299)+GOH*FIX(0.337) + pmaddwd mm7, [GOTOFF(eax,PW_MF016_MF033)] ; mm7=ROL*-FIX(0.168)+GOL*-FIX(0.331) + pmaddwd mm4, [GOTOFF(eax,PW_MF016_MF033)] ; mm4=ROH*-FIX(0.168)+GOH*-FIX(0.331) + + movq MMWORD [wk(4)], mm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337) + movq MMWORD [wk(5)], mm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337) + + pxor mm1, mm1 + pxor mm6, mm6 + punpcklwd mm1, mm5 ; mm1=BOL + punpckhwd mm6, mm5 ; mm6=BOH + psrld mm1, 1 ; mm1=BOL*FIX(0.500) + psrld mm6, 1 ; mm6=BOH*FIX(0.500) + + movq mm5, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm5=[PD_ONEHALFM1_CJ] + + paddd mm7, mm1 + paddd mm4, mm6 + paddd mm7, mm5 + paddd mm4, mm5 + psrld mm7, SCALEBITS ; mm7=CbOL + psrld mm4, SCALEBITS ; mm4=CbOH + packssdw mm7, mm4 ; mm7=CbO + + movq mm1, MMWORD [wk(2)] ; mm1=BE + + movq mm6, mm0 + punpcklwd mm0, mm2 + punpckhwd mm6, mm2 + movq mm5, mm0 + movq mm4, mm6 + pmaddwd mm0, [GOTOFF(eax,PW_F0299_F0337)] ; mm0=REL*FIX(0.299)+GEL*FIX(0.337) + pmaddwd mm6, [GOTOFF(eax,PW_F0299_F0337)] ; mm6=REH*FIX(0.299)+GEH*FIX(0.337) + pmaddwd mm5, [GOTOFF(eax,PW_MF016_MF033)] ; mm5=REL*-FIX(0.168)+GEL*-FIX(0.331) + pmaddwd mm4, [GOTOFF(eax,PW_MF016_MF033)] ; mm4=REH*-FIX(0.168)+GEH*-FIX(0.331) + + movq MMWORD [wk(6)], mm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337) + movq MMWORD [wk(7)], mm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337) + + pxor mm0, mm0 + pxor mm6, mm6 + punpcklwd mm0, mm1 ; mm0=BEL + punpckhwd mm6, mm1 ; mm6=BEH + psrld mm0, 1 ; mm0=BEL*FIX(0.500) + psrld mm6, 1 ; mm6=BEH*FIX(0.500) + + movq mm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ] + + paddd mm5, mm0 + paddd mm4, mm6 + paddd mm5, mm1 + paddd mm4, mm1 + psrld mm5, SCALEBITS ; mm5=CbEL + psrld mm4, SCALEBITS ; mm4=CbEH + packssdw mm5, mm4 ; mm5=CbE + + psllw mm7, BYTE_BIT + por mm5, mm7 ; mm5=Cb + movq MMWORD [ebx], mm5 ; Save Cb + + movq mm0, MMWORD [wk(3)] ; mm0=BO + movq mm6, MMWORD [wk(2)] ; mm6=BE + movq mm1, MMWORD [wk(1)] ; mm1=RO + + movq mm4, mm0 + punpcklwd mm0, mm3 + punpckhwd mm4, mm3 + movq mm7, mm0 + movq mm5, mm4 + pmaddwd mm0, [GOTOFF(eax,PW_F0114_F0250)] ; mm0=BOL*FIX(0.114)+GOL*FIX(0.250) + pmaddwd mm4, [GOTOFF(eax,PW_F0114_F0250)] ; mm4=BOH*FIX(0.114)+GOH*FIX(0.250) + pmaddwd mm7, [GOTOFF(eax,PW_MF008_MF041)] ; mm7=BOL*-FIX(0.081)+GOL*-FIX(0.418) + pmaddwd mm5, [GOTOFF(eax,PW_MF008_MF041)] ; mm5=BOH*-FIX(0.081)+GOH*-FIX(0.418) + + movq mm3, [GOTOFF(eax,PD_ONEHALF)] ; mm3=[PD_ONEHALF] + + paddd mm0, MMWORD [wk(4)] + paddd mm4, MMWORD [wk(5)] + paddd mm0, mm3 + paddd mm4, mm3 + psrld mm0, SCALEBITS ; mm0=YOL + psrld mm4, SCALEBITS ; mm4=YOH + packssdw mm0, mm4 ; mm0=YO + + pxor mm3, mm3 + pxor mm4, mm4 + punpcklwd mm3, mm1 ; mm3=ROL + punpckhwd mm4, mm1 ; mm4=ROH + psrld mm3, 1 ; mm3=ROL*FIX(0.500) + psrld mm4, 1 ; mm4=ROH*FIX(0.500) + + movq mm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm1=[PD_ONEHALFM1_CJ] + + paddd mm7, mm3 + paddd mm5, mm4 + paddd mm7, mm1 + paddd mm5, mm1 + psrld mm7, SCALEBITS ; mm7=CrOL + psrld mm5, SCALEBITS ; mm5=CrOH + packssdw mm7, mm5 ; mm7=CrO + + movq mm3, MMWORD [wk(0)] ; mm3=RE + + movq mm4, mm6 + punpcklwd mm6, mm2 + punpckhwd mm4, mm2 + movq mm1, mm6 + movq mm5, mm4 + pmaddwd mm6, [GOTOFF(eax,PW_F0114_F0250)] ; mm6=BEL*FIX(0.114)+GEL*FIX(0.250) + pmaddwd mm4, [GOTOFF(eax,PW_F0114_F0250)] ; mm4=BEH*FIX(0.114)+GEH*FIX(0.250) + pmaddwd mm1, [GOTOFF(eax,PW_MF008_MF041)] ; mm1=BEL*-FIX(0.081)+GEL*-FIX(0.418) + pmaddwd mm5, [GOTOFF(eax,PW_MF008_MF041)] ; mm5=BEH*-FIX(0.081)+GEH*-FIX(0.418) + + movq mm2, [GOTOFF(eax,PD_ONEHALF)] ; mm2=[PD_ONEHALF] + + paddd mm6, MMWORD [wk(6)] + paddd mm4, MMWORD [wk(7)] + paddd mm6, mm2 + paddd mm4, mm2 + psrld mm6, SCALEBITS ; mm6=YEL + psrld mm4, SCALEBITS ; mm4=YEH + packssdw mm6, mm4 ; mm6=YE + + psllw mm0, BYTE_BIT + por mm6, mm0 ; mm6=Y + movq MMWORD [edi], mm6 ; Save Y + + pxor mm2, mm2 + pxor mm4, mm4 + punpcklwd mm2, mm3 ; mm2=REL + punpckhwd mm4, mm3 ; mm4=REH + psrld mm2, 1 ; mm2=REL*FIX(0.500) + psrld mm4, 1 ; mm4=REH*FIX(0.500) + + movq mm0, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; mm0=[PD_ONEHALFM1_CJ] + + paddd mm1, mm2 + paddd mm5, mm4 + paddd mm1, mm0 + paddd mm5, mm0 + psrld mm1, SCALEBITS ; mm1=CrEL + psrld mm5, SCALEBITS ; mm5=CrEH + packssdw mm1, mm5 ; mm1=CrE + + psllw mm7, BYTE_BIT + por mm1, mm7 ; mm1=Cr + movq MMWORD [edx], mm1 ; Save Cr + + sub ecx, byte SIZEOF_MMWORD + add esi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; inptr + add edi, byte SIZEOF_MMWORD ; outptr0 + add ebx, byte SIZEOF_MMWORD ; outptr1 + add edx, byte SIZEOF_MMWORD ; outptr2 + cmp ecx, byte SIZEOF_MMWORD + jae near .columnloop + test ecx, ecx + jnz near .column_ld1 + + pop ecx ; col + pop esi + pop edi + pop ebx + pop edx + POPPIC eax + + add esi, byte SIZEOF_JSAMPROW ; input_buf + add edi, byte SIZEOF_JSAMPROW + add ebx, byte SIZEOF_JSAMPROW + add edx, byte SIZEOF_JSAMPROW + dec eax ; num_rows + jg near .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-sse2.asm new file mode 100644 index 0000000..1d8d5f5 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolext-sse2.asm @@ -0,0 +1,499 @@ +; +; jccolext.asm - colorspace conversion (SSE2) +; +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_ycc_convert_sse2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +%define img_width(b) (b) + 8 ; JDIMENSION img_width +%define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf +%define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf +%define output_row(b) (b) + 20 ; JDIMENSION output_row +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 8 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_ycc_convert_sse2) + +EXTN(jsimd_rgb_ycc_convert_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [img_width(eax)] + test ecx, ecx + jz near .return + + push ecx + + mov esi, JSAMPIMAGE [output_buf(eax)] + mov ecx, JDIMENSION [output_row(eax)] + mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [esi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [esi+2*SIZEOF_JSAMPARRAY] + lea edi, [edi+ecx*SIZEOF_JSAMPROW] + lea ebx, [ebx+ecx*SIZEOF_JSAMPROW] + lea edx, [edx+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov esi, JSAMPARRAY [input_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + PUSHPIC eax + push edx + push ebx + push edi + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr0 + mov ebx, JSAMPROW [ebx] ; outptr1 + mov edx, JSAMPROW [edx] ; outptr2 + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + cmp ecx, byte SIZEOF_XMMWORD + jae near .columnloop + ALIGNX 16, 7 + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push eax + push edx + lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub ecx, byte SIZEOF_BYTE + movzx eax, byte [esi+ecx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub ecx, byte SIZEOF_WORD + movzx edx, word [esi+ecx] + shl eax, WORD_BIT + or eax, edx +.column_ld4: + movd xmmA, eax + pop edx + pop eax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub ecx, byte SIZEOF_DWORD + movd xmmF, XMM_DWORD [esi+ecx] + pslldq xmmA, SIZEOF_DWORD + por xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub ecx, byte SIZEOF_MMWORD + movq xmmB, XMM_MMWORD [esi+ecx] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + movdqa xmmF, xmmA + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + mov ecx, SIZEOF_XMMWORD + jmp short .rgb_ycc_cnv +.column_ld32: + test cl, 2*SIZEOF_XMMWORD + mov ecx, SIZEOF_XMMWORD + jz short .rgb_ycc_cnv + movdqa xmmB, xmmA + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD] + jmp short .rgb_ycc_cnv + ALIGNX 16, 7 + +.columnloop: + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD] + movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD] + +.rgb_ycc_cnv: + ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + movdqa xmmG, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12) + psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A) + pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27) + + punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D) + punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F) + + movdqa xmmD, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09) + psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D) + pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B) + + punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E) + punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F) + + movdqa xmmE, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C) + psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D) + + punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F) + + pxor xmmH, xmmH + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmB, xmmE + punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F) + + movdqa xmmF, xmmD + punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F) + punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub ecx, byte SIZEOF_XMMWORD/16 + movd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub ecx, byte SIZEOF_XMMWORD/8 + movq xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmE +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub ecx, byte SIZEOF_XMMWORD/4 + movdqa xmmE, xmmA + movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + mov ecx, SIZEOF_XMMWORD + jz short .rgb_ycc_cnv + movdqa xmmF, xmmA + movdqa xmmH, xmmE + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD] + jmp short .rgb_ycc_cnv + ALIGNX 16, 7 + +.columnloop: + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD] + movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD] + +.rgb_ycc_cnv: + ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35) + punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37) + + movdqa xmmC, xmmF + punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D) + punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F) + + movdqa xmmB, xmmA + punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C) + punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D) + + movdqa xmmG, xmmD + punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E) + punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F) + + movdqa xmmE, xmmA + punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E) + + movdqa xmmH, xmmB + punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F) + punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F) + + pxor xmmF, xmmF + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmD, xmmB + punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F) + + movdqa xmmG, xmmE + punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E) + + punpcklbw xmmF, xmmH + punpckhbw xmmH, xmmH + psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F) + psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE + ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE + movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO + movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE + movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO + + movdqa xmm6, xmm1 + punpcklwd xmm1, xmm3 + punpckhwd xmm6, xmm3 + movdqa xmm7, xmm1 + movdqa xmm4, xmm6 + pmaddwd xmm1, [GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337) + pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337) + pmaddwd xmm7, [GOTOFF(eax,PW_MF016_MF033)] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331) + pmaddwd xmm4, [GOTOFF(eax,PW_MF016_MF033)] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331) + + movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337) + movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337) + + pxor xmm1, xmm1 + pxor xmm6, xmm6 + punpcklwd xmm1, xmm5 ; xmm1=BOL + punpckhwd xmm6, xmm5 ; xmm6=BOH + psrld xmm1, 1 ; xmm1=BOL*FIX(0.500) + psrld xmm6, 1 ; xmm6=BOH*FIX(0.500) + + movdqa xmm5, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm5=[PD_ONEHALFM1_CJ] + + paddd xmm7, xmm1 + paddd xmm4, xmm6 + paddd xmm7, xmm5 + paddd xmm4, xmm5 + psrld xmm7, SCALEBITS ; xmm7=CbOL + psrld xmm4, SCALEBITS ; xmm4=CbOH + packssdw xmm7, xmm4 ; xmm7=CbO + + movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE + + movdqa xmm6, xmm0 + punpcklwd xmm0, xmm2 + punpckhwd xmm6, xmm2 + movdqa xmm5, xmm0 + movdqa xmm4, xmm6 + pmaddwd xmm0, [GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337) + pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337) + pmaddwd xmm5, [GOTOFF(eax,PW_MF016_MF033)] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331) + pmaddwd xmm4, [GOTOFF(eax,PW_MF016_MF033)] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331) + + movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337) + movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337) + + pxor xmm0, xmm0 + pxor xmm6, xmm6 + punpcklwd xmm0, xmm1 ; xmm0=BEL + punpckhwd xmm6, xmm1 ; xmm6=BEH + psrld xmm0, 1 ; xmm0=BEL*FIX(0.500) + psrld xmm6, 1 ; xmm6=BEH*FIX(0.500) + + movdqa xmm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ] + + paddd xmm5, xmm0 + paddd xmm4, xmm6 + paddd xmm5, xmm1 + paddd xmm4, xmm1 + psrld xmm5, SCALEBITS ; xmm5=CbEL + psrld xmm4, SCALEBITS ; xmm4=CbEH + packssdw xmm5, xmm4 ; xmm5=CbE + + psllw xmm7, BYTE_BIT + por xmm5, xmm7 ; xmm5=Cb + movdqa XMMWORD [ebx], xmm5 ; Save Cb + + movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO + movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm3 + punpckhwd xmm4, xmm3 + movdqa xmm7, xmm0 + movdqa xmm5, xmm4 + pmaddwd xmm0, [GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250) + pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250) + pmaddwd xmm7, [GOTOFF(eax,PW_MF008_MF041)] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418) + pmaddwd xmm5, [GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418) + + movdqa xmm3, [GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF] + + paddd xmm0, XMMWORD [wk(4)] + paddd xmm4, XMMWORD [wk(5)] + paddd xmm0, xmm3 + paddd xmm4, xmm3 + psrld xmm0, SCALEBITS ; xmm0=YOL + psrld xmm4, SCALEBITS ; xmm4=YOH + packssdw xmm0, xmm4 ; xmm0=YO + + pxor xmm3, xmm3 + pxor xmm4, xmm4 + punpcklwd xmm3, xmm1 ; xmm3=ROL + punpckhwd xmm4, xmm1 ; xmm4=ROH + psrld xmm3, 1 ; xmm3=ROL*FIX(0.500) + psrld xmm4, 1 ; xmm4=ROH*FIX(0.500) + + movdqa xmm1, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm1=[PD_ONEHALFM1_CJ] + + paddd xmm7, xmm3 + paddd xmm5, xmm4 + paddd xmm7, xmm1 + paddd xmm5, xmm1 + psrld xmm7, SCALEBITS ; xmm7=CrOL + psrld xmm5, SCALEBITS ; xmm5=CrOH + packssdw xmm7, xmm5 ; xmm7=CrO + + movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE + + movdqa xmm4, xmm6 + punpcklwd xmm6, xmm2 + punpckhwd xmm4, xmm2 + movdqa xmm1, xmm6 + movdqa xmm5, xmm4 + pmaddwd xmm6, [GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250) + pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250) + pmaddwd xmm1, [GOTOFF(eax,PW_MF008_MF041)] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418) + pmaddwd xmm5, [GOTOFF(eax,PW_MF008_MF041)] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418) + + movdqa xmm2, [GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF] + + paddd xmm6, XMMWORD [wk(6)] + paddd xmm4, XMMWORD [wk(7)] + paddd xmm6, xmm2 + paddd xmm4, xmm2 + psrld xmm6, SCALEBITS ; xmm6=YEL + psrld xmm4, SCALEBITS ; xmm4=YEH + packssdw xmm6, xmm4 ; xmm6=YE + + psllw xmm0, BYTE_BIT + por xmm6, xmm0 ; xmm6=Y + movdqa XMMWORD [edi], xmm6 ; Save Y + + pxor xmm2, xmm2 + pxor xmm4, xmm4 + punpcklwd xmm2, xmm3 ; xmm2=REL + punpckhwd xmm4, xmm3 ; xmm4=REH + psrld xmm2, 1 ; xmm2=REL*FIX(0.500) + psrld xmm4, 1 ; xmm4=REH*FIX(0.500) + + movdqa xmm0, [GOTOFF(eax,PD_ONEHALFM1_CJ)] ; xmm0=[PD_ONEHALFM1_CJ] + + paddd xmm1, xmm2 + paddd xmm5, xmm4 + paddd xmm1, xmm0 + paddd xmm5, xmm0 + psrld xmm1, SCALEBITS ; xmm1=CrEL + psrld xmm5, SCALEBITS ; xmm5=CrEH + packssdw xmm1, xmm5 ; xmm1=CrE + + psllw xmm7, BYTE_BIT + por xmm1, xmm7 ; xmm1=Cr + movdqa XMMWORD [edx], xmm1 ; Save Cr + + sub ecx, byte SIZEOF_XMMWORD + add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr + add edi, byte SIZEOF_XMMWORD ; outptr0 + add ebx, byte SIZEOF_XMMWORD ; outptr1 + add edx, byte SIZEOF_XMMWORD ; outptr2 + cmp ecx, byte SIZEOF_XMMWORD + jae near .columnloop + test ecx, ecx + jnz near .column_ld1 + + pop ecx ; col + pop esi + pop edi + pop ebx + pop edx + POPPIC eax + + add esi, byte SIZEOF_JSAMPROW ; input_buf + add edi, byte SIZEOF_JSAMPROW + add ebx, byte SIZEOF_JSAMPROW + add edx, byte SIZEOF_JSAMPROW + dec eax ; num_rows + jg near .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-avx2.asm new file mode 100644 index 0000000..9ad5ea9 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-avx2.asm @@ -0,0 +1,117 @@ +; +; jccolor.asm - colorspace conversion (AVX2) +; +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_081 equ 5329 ; FIX(0.08131) +F_0_114 equ 7471 ; FIX(0.11400) +F_0_168 equ 11059 ; FIX(0.16874) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_331 equ 21709 ; FIX(0.33126) +F_0_418 equ 27439 ; FIX(0.41869) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_ycc_convert_avx2) + +EXTN(jconst_rgb_ycc_convert_avx2): + +PW_F0299_F0337 times 8 dw F_0_299, F_0_337 +PW_F0114_F0250 times 8 dw F_0_114, F_0_250 +PW_MF016_MF033 times 8 dw -F_0_168, -F_0_331 +PW_MF008_MF041 times 8 dw -F_0_081, -F_0_418 +PD_ONEHALFM1_CJ times 8 dd (1 << (SCALEBITS - 1)) - 1 + \ + (CENTERJSAMPLE << SCALEBITS) +PD_ONEHALF times 8 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgb_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgbx_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgr_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgrx_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extxbgr_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extxrgb_ycc_convert_avx2 +%include "jccolext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-mmx.asm new file mode 100644 index 0000000..0dbec54 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-mmx.asm @@ -0,0 +1,117 @@ +; +; jccolor.asm - colorspace conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_081 equ 5329 ; FIX(0.08131) +F_0_114 equ 7471 ; FIX(0.11400) +F_0_168 equ 11059 ; FIX(0.16874) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_331 equ 21709 ; FIX(0.33126) +F_0_418 equ 27439 ; FIX(0.41869) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_ycc_convert_mmx) + +EXTN(jconst_rgb_ycc_convert_mmx): + +PW_F0299_F0337 times 2 dw F_0_299, F_0_337 +PW_F0114_F0250 times 2 dw F_0_114, F_0_250 +PW_MF016_MF033 times 2 dw -F_0_168, -F_0_331 +PW_MF008_MF041 times 2 dw -F_0_081, -F_0_418 +PD_ONEHALFM1_CJ times 2 dd (1 << (SCALEBITS - 1)) - 1 + \ + (CENTERJSAMPLE << SCALEBITS) +PD_ONEHALF times 2 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jccolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_mmx jsimd_extrgb_ycc_convert_mmx +%include "jccolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_ycc_convert_mmx jsimd_extrgbx_ycc_convert_mmx +%include "jccolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_mmx jsimd_extbgr_ycc_convert_mmx +%include "jccolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_ycc_convert_mmx jsimd_extbgrx_ycc_convert_mmx +%include "jccolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_mmx jsimd_extxbgr_ycc_convert_mmx +%include "jccolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_mmx jsimd_extxrgb_ycc_convert_mmx +%include "jccolext-mmx.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-sse2.asm new file mode 100644 index 0000000..678306a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jccolor-sse2.asm @@ -0,0 +1,116 @@ +; +; jccolor.asm - colorspace conversion (SSE2) +; +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_081 equ 5329 ; FIX(0.08131) +F_0_114 equ 7471 ; FIX(0.11400) +F_0_168 equ 11059 ; FIX(0.16874) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_331 equ 21709 ; FIX(0.33126) +F_0_418 equ 27439 ; FIX(0.41869) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_ycc_convert_sse2) + +EXTN(jconst_rgb_ycc_convert_sse2): + +PW_F0299_F0337 times 4 dw F_0_299, F_0_337 +PW_F0114_F0250 times 4 dw F_0_114, F_0_250 +PW_MF016_MF033 times 4 dw -F_0_168, -F_0_331 +PW_MF008_MF041 times 4 dw -F_0_081, -F_0_418 +PD_ONEHALFM1_CJ times 4 dd (1 << (SCALEBITS - 1)) - 1 + \ + (CENTERJSAMPLE << SCALEBITS) +PD_ONEHALF times 4 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgb_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgbx_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgr_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgrx_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extxbgr_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extxrgb_ycc_convert_sse2 +%include "jccolext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-avx2.asm new file mode 100644 index 0000000..ded3956 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-avx2.asm @@ -0,0 +1,109 @@ +; +; jcgray.asm - grayscale colorspace conversion (AVX2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_114 equ 7471 ; FIX(0.11400) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_gray_convert_avx2) + +EXTN(jconst_rgb_gray_convert_avx2): + +PW_F0299_F0337 times 8 dw F_0_299, F_0_337 +PW_F0114_F0250 times 8 dw F_0_114, F_0_250 +PD_ONEHALF times 8 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extrgb_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extrgbx_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extbgr_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extbgrx_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extxbgr_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extxrgb_gray_convert_avx2 +%include "jcgryext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-mmx.asm new file mode 100644 index 0000000..d6f0318 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-mmx.asm @@ -0,0 +1,109 @@ +; +; jcgray.asm - grayscale colorspace conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_114 equ 7471 ; FIX(0.11400) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_gray_convert_mmx) + +EXTN(jconst_rgb_gray_convert_mmx): + +PW_F0299_F0337 times 2 dw F_0_299, F_0_337 +PW_F0114_F0250 times 2 dw F_0_114, F_0_250 +PD_ONEHALF times 2 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jcgryext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_gray_convert_mmx jsimd_extrgb_gray_convert_mmx +%include "jcgryext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_gray_convert_mmx jsimd_extrgbx_gray_convert_mmx +%include "jcgryext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_gray_convert_mmx jsimd_extbgr_gray_convert_mmx +%include "jcgryext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_gray_convert_mmx jsimd_extbgrx_gray_convert_mmx +%include "jcgryext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_gray_convert_mmx jsimd_extxbgr_gray_convert_mmx +%include "jcgryext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_gray_convert_mmx jsimd_extxrgb_gray_convert_mmx +%include "jcgryext-mmx.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-sse2.asm new file mode 100644 index 0000000..ecc7fa0 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgray-sse2.asm @@ -0,0 +1,108 @@ +; +; jcgray.asm - grayscale colorspace conversion (SSE2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_114 equ 7471 ; FIX(0.11400) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_gray_convert_sse2) + +EXTN(jconst_rgb_gray_convert_sse2): + +PW_F0299_F0337 times 4 dw F_0_299, F_0_337 +PW_F0114_F0250 times 4 dw F_0_114, F_0_250 +PD_ONEHALF times 4 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extrgb_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extrgbx_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extbgr_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extbgrx_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extxbgr_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extxrgb_gray_convert_sse2 +%include "jcgryext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-avx2.asm new file mode 100644 index 0000000..70df8f8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-avx2.asm @@ -0,0 +1,453 @@ +; +; jcgryext.asm - grayscale colorspace conversion (AVX2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_gray_convert_avx2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +%define img_width(b) (b) + 8 ; JDIMENSION img_width +%define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf +%define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf +%define output_row(b) (b) + 20 ; JDIMENSION output_row +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_YMMWORD + ; ymmword wk[WK_NUM] +%define WK_NUM 2 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_gray_convert_avx2) + +EXTN(jsimd_rgb_gray_convert_avx2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [img_width(eax)] + test ecx, ecx + jz near .return + + push ecx + + mov esi, JSAMPIMAGE [output_buf(eax)] + mov ecx, JDIMENSION [output_row(eax)] + mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY] + lea edi, [edi+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov esi, JSAMPARRAY [input_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + PUSHPIC eax + push edi + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr0 + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + cmp ecx, byte SIZEOF_YMMWORD + jae near .columnloop + ALIGNX 16, 7 + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push eax + push edx + lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub ecx, byte SIZEOF_BYTE + movzx eax, byte [esi+ecx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub ecx, byte SIZEOF_WORD + movzx edx, word [esi+ecx] + shl eax, WORD_BIT + or eax, edx +.column_ld4: + vmovd xmmA, eax + pop edx + pop eax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub ecx, byte SIZEOF_DWORD + vmovd xmmF, XMM_DWORD [esi+ecx] + vpslldq xmmA, xmmA, SIZEOF_DWORD + vpor xmmA, xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub ecx, byte SIZEOF_MMWORD + vmovq xmmB, XMM_MMWORD [esi+ecx] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + sub ecx, byte SIZEOF_XMMWORD + vmovdqu xmmB, XMM_MMWORD [esi+ecx] + vperm2i128 ymmA, ymmA, ymmA, 1 + vpor ymmA, ymmB +.column_ld32: + test cl, SIZEOF_YMMWORD + jz short .column_ld64 + sub ecx, byte SIZEOF_YMMWORD + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] +.column_ld64: + test cl, 2*SIZEOF_YMMWORD + mov ecx, SIZEOF_YMMWORD + jz short .rgb_gray_cnv + vmovdqa ymmB, ymmA + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + jmp short .rgb_gray_cnv + ALIGNX 16, 7 + +.columnloop: + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + vmovdqu ymmB, YMMWORD [esi+2*SIZEOF_YMMWORD] + +.rgb_gray_cnv: + ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vmovdqu ymmC, ymmA + vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + + vmovdqa ymmG, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12 + ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I) + vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I + ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A + ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q) + vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27 + ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N) + + vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D + ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T) + vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F + ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V) + + vmovdqa ymmD, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09 + ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P) + vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P + ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T) + vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B + ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R) + + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E + ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F + ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V) + + vmovdqa ymmE, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C + ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S) + vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S + ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D + ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T) + + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F + ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F + ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V) + + vpxor ymmH, ymmH, ymmH + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmB, ymmE + vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + + vmovdqa ymmF, ymmD + vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub ecx, byte SIZEOF_XMMWORD/16 + vmovd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub ecx, byte SIZEOF_XMMWORD/8 + vmovq xmmF, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmF +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub ecx, byte SIZEOF_XMMWORD/4 + vmovdqa xmmF, xmmA + vperm2i128 ymmF, ymmF, ymmF, 1 + vmovdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE] + vpor ymmA, ymmA, ymmF +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + jz short .column_ld16 + sub ecx, byte SIZEOF_XMMWORD/2 + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld16: + test cl, SIZEOF_XMMWORD + mov ecx, SIZEOF_YMMWORD + jz short .rgb_gray_cnv + vmovdqa ymmE, ymmA + vmovdqa ymmH, ymmF + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + jmp short .rgb_gray_cnv + ALIGNX 16, 7 + +.columnloop: + vmovdqu ymmA, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [esi+1*SIZEOF_YMMWORD] + vmovdqu ymmE, YMMWORD [esi+2*SIZEOF_YMMWORD] + vmovdqu ymmH, YMMWORD [esi+3*SIZEOF_YMMWORD] + +.rgb_gray_cnv: + ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmB, ymmA + vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + + vmovdqa ymmB, ymmF + vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmD, ymmA + vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35 + ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L) + vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37 + ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N) + + vmovdqa ymmC, ymmF + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D + ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F + ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V) + + vmovdqa ymmB, ymmA + vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S) + vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D + ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T) + + vmovdqa ymmG, ymmD + vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E + ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U) + vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F + ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V) + + vmovdqa ymmE, ymmA + vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E + ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U) + + vmovdqa ymmH, ymmB + vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F + ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F + ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V) + + vpxor ymmF, ymmF, ymmF + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmD, ymmB + vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + + vmovdqa ymmG, ymmE + vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U) + + vpunpcklbw ymmF, ymmF, ymmH + vpunpckhbw ymmH, ymmH, ymmH + vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE + ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + + vmovdqa ymm6, ymm1 + vpunpcklwd ymm1, ymm1, ymm3 + vpunpckhwd ymm6, ymm6, ymm3 + vpmaddwd ymm1, ymm1, [GOTOFF(eax,PW_F0299_F0337)] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337) + vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337) + + vmovdqa ymm7, ymm6 ; ymm7=ROH*FIX(0.299)+GOH*FIX(0.337) + + vmovdqa ymm6, ymm0 + vpunpcklwd ymm0, ymm0, ymm2 + vpunpckhwd ymm6, ymm6, ymm2 + vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0299_F0337)] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337) + vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0299_F0337)] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337) + + vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337) + vmovdqa YMMWORD [wk(1)], ymm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337) + + vmovdqa ymm0, ymm5 ; ymm0=BO + vmovdqa ymm6, ymm4 ; ymm6=BE + + vmovdqa ymm4, ymm0 + vpunpcklwd ymm0, ymm0, ymm3 + vpunpckhwd ymm4, ymm4, ymm3 + vpmaddwd ymm0, ymm0, [GOTOFF(eax,PW_F0114_F0250)] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250) + vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250) + + vmovdqa ymm3, [GOTOFF(eax,PD_ONEHALF)] ; ymm3=[PD_ONEHALF] + + vpaddd ymm0, ymm0, ymm1 + vpaddd ymm4, ymm4, ymm7 + vpaddd ymm0, ymm0, ymm3 + vpaddd ymm4, ymm4, ymm3 + vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH + vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO + + vmovdqa ymm4, ymm6 + vpunpcklwd ymm6, ymm6, ymm2 + vpunpckhwd ymm4, ymm4, ymm2 + vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_F0114_F0250)] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250) + vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_F0114_F0250)] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250) + + vmovdqa ymm2, [GOTOFF(eax,PD_ONEHALF)] ; ymm2=[PD_ONEHALF] + + vpaddd ymm6, ymm6, YMMWORD [wk(0)] + vpaddd ymm4, ymm4, YMMWORD [wk(1)] + vpaddd ymm6, ymm6, ymm2 + vpaddd ymm4, ymm4, ymm2 + vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH + vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE + + vpsllw ymm0, ymm0, BYTE_BIT + vpor ymm6, ymm6, ymm0 ; ymm6=Y + vmovdqu YMMWORD [edi], ymm6 ; Save Y + + sub ecx, byte SIZEOF_YMMWORD + add esi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr + add edi, byte SIZEOF_YMMWORD ; outptr0 + cmp ecx, byte SIZEOF_YMMWORD + jae near .columnloop + test ecx, ecx + jnz near .column_ld1 + + pop ecx ; col + pop esi + pop edi + POPPIC eax + + add esi, byte SIZEOF_JSAMPROW ; input_buf + add edi, byte SIZEOF_JSAMPROW + dec eax ; num_rows + jg near .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-mmx.asm new file mode 100644 index 0000000..dd90c3d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-mmx.asm @@ -0,0 +1,351 @@ +; +; jcgryext.asm - grayscale colorspace conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_gray_convert_mmx(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +%define img_width(b) (b) + 8 ; JDIMENSION img_width +%define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf +%define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf +%define output_row(b) (b) + 20 ; JDIMENSION output_row +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD + ; mmword wk[WK_NUM] +%define WK_NUM 2 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_gray_convert_mmx) + +EXTN(jsimd_rgb_gray_convert_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [img_width(eax)] ; num_cols + test ecx, ecx + jz near .return + + push ecx + + mov esi, JSAMPIMAGE [output_buf(eax)] + mov ecx, JDIMENSION [output_row(eax)] + mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY] + lea edi, [edi+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov esi, JSAMPARRAY [input_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + PUSHPIC eax + push edi + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr0 + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + cmp ecx, byte SIZEOF_MMWORD + jae short .columnloop + ALIGNX 16, 7 + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push eax + push edx + lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub ecx, byte SIZEOF_BYTE + xor eax, eax + mov al, byte [esi+ecx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub ecx, byte SIZEOF_WORD + xor edx, edx + mov dx, word [esi+ecx] + shl eax, WORD_BIT + or eax, edx +.column_ld4: + movd mmA, eax + pop edx + pop eax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub ecx, byte SIZEOF_DWORD + movd mmG, dword [esi+ecx] + psllq mmA, DWORD_BIT + por mmA, mmG +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + movq mmG, mmA + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + mov ecx, SIZEOF_MMWORD + jmp short .rgb_gray_cnv +.column_ld16: + test cl, 2*SIZEOF_MMWORD + mov ecx, SIZEOF_MMWORD + jz short .rgb_gray_cnv + movq mmF, mmA + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmG, MMWORD [esi+1*SIZEOF_MMWORD] + jmp short .rgb_gray_cnv + ALIGNX 16, 7 + +.columnloop: + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmG, MMWORD [esi+1*SIZEOF_MMWORD] + movq mmF, MMWORD [esi+2*SIZEOF_MMWORD] + +.rgb_gray_cnv: + ; mmA=(00 10 20 01 11 21 02 12) + ; mmG=(22 03 13 23 04 14 24 05) + ; mmF=(15 25 06 16 26 07 17 27) + + movq mmD, mmA + psllq mmA, 4*BYTE_BIT ; mmA=(-- -- -- -- 00 10 20 01) + psrlq mmD, 4*BYTE_BIT ; mmD=(11 21 02 12 -- -- -- --) + + punpckhbw mmA, mmG ; mmA=(00 04 10 14 20 24 01 05) + psllq mmG, 4*BYTE_BIT ; mmG=(-- -- -- -- 22 03 13 23) + + punpcklbw mmD, mmF ; mmD=(11 15 21 25 02 06 12 16) + punpckhbw mmG, mmF ; mmG=(22 26 03 07 13 17 23 27) + + movq mmE, mmA + psllq mmA, 4*BYTE_BIT ; mmA=(-- -- -- -- 00 04 10 14) + psrlq mmE, 4*BYTE_BIT ; mmE=(20 24 01 05 -- -- -- --) + + punpckhbw mmA, mmD ; mmA=(00 02 04 06 10 12 14 16) + psllq mmD, 4*BYTE_BIT ; mmD=(-- -- -- -- 11 15 21 25) + + punpcklbw mmE, mmG ; mmE=(20 22 24 26 01 03 05 07) + punpckhbw mmD, mmG ; mmD=(11 13 15 17 21 23 25 27) + + pxor mmH, mmH + + movq mmC, mmA + punpcklbw mmA, mmH ; mmA=(00 02 04 06) + punpckhbw mmC, mmH ; mmC=(10 12 14 16) + + movq mmB, mmE + punpcklbw mmE, mmH ; mmE=(20 22 24 26) + punpckhbw mmB, mmH ; mmB=(01 03 05 07) + + movq mmF, mmD + punpcklbw mmD, mmH ; mmD=(11 13 15 17) + punpckhbw mmF, mmH ; mmF=(21 23 25 27) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_MMWORD/8 + jz short .column_ld2 + sub ecx, byte SIZEOF_MMWORD/8 + movd mmA, dword [esi+ecx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_MMWORD/4 + jz short .column_ld4 + sub ecx, byte SIZEOF_MMWORD/4 + movq mmF, mmA + movq mmA, MMWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld4: + test cl, SIZEOF_MMWORD/2 + mov ecx, SIZEOF_MMWORD + jz short .rgb_gray_cnv + movq mmD, mmA + movq mmC, mmF + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmF, MMWORD [esi+1*SIZEOF_MMWORD] + jmp short .rgb_gray_cnv + ALIGNX 16, 7 + +.columnloop: + movq mmA, MMWORD [esi+0*SIZEOF_MMWORD] + movq mmF, MMWORD [esi+1*SIZEOF_MMWORD] + movq mmD, MMWORD [esi+2*SIZEOF_MMWORD] + movq mmC, MMWORD [esi+3*SIZEOF_MMWORD] + +.rgb_gray_cnv: + ; mmA=(00 10 20 30 01 11 21 31) + ; mmF=(02 12 22 32 03 13 23 33) + ; mmD=(04 14 24 34 05 15 25 35) + ; mmC=(06 16 26 36 07 17 27 37) + + movq mmB, mmA + punpcklbw mmA, mmF ; mmA=(00 02 10 12 20 22 30 32) + punpckhbw mmB, mmF ; mmB=(01 03 11 13 21 23 31 33) + + movq mmG, mmD + punpcklbw mmD, mmC ; mmD=(04 06 14 16 24 26 34 36) + punpckhbw mmG, mmC ; mmG=(05 07 15 17 25 27 35 37) + + movq mmE, mmA + punpcklwd mmA, mmD ; mmA=(00 02 04 06 10 12 14 16) + punpckhwd mmE, mmD ; mmE=(20 22 24 26 30 32 34 36) + + movq mmH, mmB + punpcklwd mmB, mmG ; mmB=(01 03 05 07 11 13 15 17) + punpckhwd mmH, mmG ; mmH=(21 23 25 27 31 33 35 37) + + pxor mmF, mmF + + movq mmC, mmA + punpcklbw mmA, mmF ; mmA=(00 02 04 06) + punpckhbw mmC, mmF ; mmC=(10 12 14 16) + + movq mmD, mmB + punpcklbw mmB, mmF ; mmB=(01 03 05 07) + punpckhbw mmD, mmF ; mmD=(11 13 15 17) + + movq mmG, mmE + punpcklbw mmE, mmF ; mmE=(20 22 24 26) + punpckhbw mmG, mmF ; mmG=(30 32 34 36) + + punpcklbw mmF, mmH + punpckhbw mmH, mmH + psrlw mmF, BYTE_BIT ; mmF=(21 23 25 27) + psrlw mmH, BYTE_BIT ; mmH=(31 33 35 37) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; mm0=(R0 R2 R4 R6)=RE, mm2=(G0 G2 G4 G6)=GE, mm4=(B0 B2 B4 B6)=BE + ; mm1=(R1 R3 R5 R7)=RO, mm3=(G1 G3 G5 G7)=GO, mm5=(B1 B3 B5 B7)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + + movq mm6, mm1 + punpcklwd mm1, mm3 + punpckhwd mm6, mm3 + pmaddwd mm1, [GOTOFF(eax,PW_F0299_F0337)] ; mm1=ROL*FIX(0.299)+GOL*FIX(0.337) + pmaddwd mm6, [GOTOFF(eax,PW_F0299_F0337)] ; mm6=ROH*FIX(0.299)+GOH*FIX(0.337) + + movq mm7, mm6 ; mm7=ROH*FIX(0.299)+GOH*FIX(0.337) + + movq mm6, mm0 + punpcklwd mm0, mm2 + punpckhwd mm6, mm2 + pmaddwd mm0, [GOTOFF(eax,PW_F0299_F0337)] ; mm0=REL*FIX(0.299)+GEL*FIX(0.337) + pmaddwd mm6, [GOTOFF(eax,PW_F0299_F0337)] ; mm6=REH*FIX(0.299)+GEH*FIX(0.337) + + movq MMWORD [wk(0)], mm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337) + movq MMWORD [wk(1)], mm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337) + + movq mm0, mm5 ; mm0=BO + movq mm6, mm4 ; mm6=BE + + movq mm4, mm0 + punpcklwd mm0, mm3 + punpckhwd mm4, mm3 + pmaddwd mm0, [GOTOFF(eax,PW_F0114_F0250)] ; mm0=BOL*FIX(0.114)+GOL*FIX(0.250) + pmaddwd mm4, [GOTOFF(eax,PW_F0114_F0250)] ; mm4=BOH*FIX(0.114)+GOH*FIX(0.250) + + movq mm3, [GOTOFF(eax,PD_ONEHALF)] ; mm3=[PD_ONEHALF] + + paddd mm0, mm1 + paddd mm4, mm7 + paddd mm0, mm3 + paddd mm4, mm3 + psrld mm0, SCALEBITS ; mm0=YOL + psrld mm4, SCALEBITS ; mm4=YOH + packssdw mm0, mm4 ; mm0=YO + + movq mm4, mm6 + punpcklwd mm6, mm2 + punpckhwd mm4, mm2 + pmaddwd mm6, [GOTOFF(eax,PW_F0114_F0250)] ; mm6=BEL*FIX(0.114)+GEL*FIX(0.250) + pmaddwd mm4, [GOTOFF(eax,PW_F0114_F0250)] ; mm4=BEH*FIX(0.114)+GEH*FIX(0.250) + + movq mm2, [GOTOFF(eax,PD_ONEHALF)] ; mm2=[PD_ONEHALF] + + paddd mm6, MMWORD [wk(0)] + paddd mm4, MMWORD [wk(1)] + paddd mm6, mm2 + paddd mm4, mm2 + psrld mm6, SCALEBITS ; mm6=YEL + psrld mm4, SCALEBITS ; mm4=YEH + packssdw mm6, mm4 ; mm6=YE + + psllw mm0, BYTE_BIT + por mm6, mm0 ; mm6=Y + movq MMWORD [edi], mm6 ; Save Y + + sub ecx, byte SIZEOF_MMWORD + add esi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; inptr + add edi, byte SIZEOF_MMWORD ; outptr0 + cmp ecx, byte SIZEOF_MMWORD + jae near .columnloop + test ecx, ecx + jnz near .column_ld1 + + pop ecx ; col + pop esi + pop edi + POPPIC eax + + add esi, byte SIZEOF_JSAMPROW ; input_buf + add edi, byte SIZEOF_JSAMPROW + dec eax ; num_rows + jg near .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-sse2.asm new file mode 100644 index 0000000..227295f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcgryext-sse2.asm @@ -0,0 +1,378 @@ +; +; jcgryext.asm - grayscale colorspace conversion (SSE2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_gray_convert_sse2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +%define img_width(b) (b) + 8 ; JDIMENSION img_width +%define input_buf(b) (b) + 12 ; JSAMPARRAY input_buf +%define output_buf(b) (b) + 16 ; JSAMPIMAGE output_buf +%define output_row(b) (b) + 20 ; JDIMENSION output_row +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_gray_convert_sse2) + +EXTN(jsimd_rgb_gray_convert_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [img_width(eax)] + test ecx, ecx + jz near .return + + push ecx + + mov esi, JSAMPIMAGE [output_buf(eax)] + mov ecx, JDIMENSION [output_row(eax)] + mov edi, JSAMPARRAY [esi+0*SIZEOF_JSAMPARRAY] + lea edi, [edi+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov esi, JSAMPARRAY [input_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + PUSHPIC eax + push edi + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr0 + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + cmp ecx, byte SIZEOF_XMMWORD + jae near .columnloop + ALIGNX 16, 7 + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push eax + push edx + lea ecx, [ecx+ecx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub ecx, byte SIZEOF_BYTE + movzx eax, byte [esi+ecx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub ecx, byte SIZEOF_WORD + movzx edx, word [esi+ecx] + shl eax, WORD_BIT + or eax, edx +.column_ld4: + movd xmmA, eax + pop edx + pop eax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub ecx, byte SIZEOF_DWORD + movd xmmF, XMM_DWORD [esi+ecx] + pslldq xmmA, SIZEOF_DWORD + por xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub ecx, byte SIZEOF_MMWORD + movq xmmB, XMM_MMWORD [esi+ecx] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + movdqa xmmF, xmmA + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + mov ecx, SIZEOF_XMMWORD + jmp short .rgb_gray_cnv +.column_ld32: + test cl, 2*SIZEOF_XMMWORD + mov ecx, SIZEOF_XMMWORD + jz short .rgb_gray_cnv + movdqa xmmB, xmmA + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD] + jmp short .rgb_gray_cnv + ALIGNX 16, 7 + +.columnloop: + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [esi+1*SIZEOF_XMMWORD] + movdqu xmmB, XMMWORD [esi+2*SIZEOF_XMMWORD] + +.rgb_gray_cnv: + ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + movdqa xmmG, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12) + psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A) + pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27) + + punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D) + punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F) + + movdqa xmmD, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09) + psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D) + pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B) + + punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E) + punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F) + + movdqa xmmE, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C) + psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D) + + punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F) + + pxor xmmH, xmmH + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmB, xmmE + punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F) + + movdqa xmmF, xmmD + punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F) + punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub ecx, byte SIZEOF_XMMWORD/16 + movd xmmA, XMM_DWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub ecx, byte SIZEOF_XMMWORD/8 + movq xmmE, XMM_MMWORD [esi+ecx*RGB_PIXELSIZE] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmE +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub ecx, byte SIZEOF_XMMWORD/4 + movdqa xmmE, xmmA + movdqu xmmA, XMMWORD [esi+ecx*RGB_PIXELSIZE] +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + mov ecx, SIZEOF_XMMWORD + jz short .rgb_gray_cnv + movdqa xmmF, xmmA + movdqa xmmH, xmmE + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD] + jmp short .rgb_gray_cnv + ALIGNX 16, 7 + +.columnloop: + movdqu xmmA, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [esi+1*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [esi+2*SIZEOF_XMMWORD] + movdqu xmmH, XMMWORD [esi+3*SIZEOF_XMMWORD] + +.rgb_gray_cnv: + ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35) + punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37) + + movdqa xmmC, xmmF + punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D) + punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F) + + movdqa xmmB, xmmA + punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C) + punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D) + + movdqa xmmG, xmmD + punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E) + punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F) + + movdqa xmmE, xmmA + punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E) + + movdqa xmmH, xmmB + punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F) + punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F) + + pxor xmmF, xmmF + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmD, xmmB + punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F) + + movdqa xmmG, xmmE + punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E) + + punpcklbw xmmF, xmmH + punpckhbw xmmH, xmmH + psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F) + psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE + ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + + movdqa xmm6, xmm1 + punpcklwd xmm1, xmm3 + punpckhwd xmm6, xmm3 + pmaddwd xmm1, [GOTOFF(eax,PW_F0299_F0337)] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337) + pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337) + + movdqa xmm7, xmm6 ; xmm7=ROH*FIX(0.299)+GOH*FIX(0.337) + + movdqa xmm6, xmm0 + punpcklwd xmm0, xmm2 + punpckhwd xmm6, xmm2 + pmaddwd xmm0, [GOTOFF(eax,PW_F0299_F0337)] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337) + pmaddwd xmm6, [GOTOFF(eax,PW_F0299_F0337)] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337) + + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337) + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337) + + movdqa xmm0, xmm5 ; xmm0=BO + movdqa xmm6, xmm4 ; xmm6=BE + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm3 + punpckhwd xmm4, xmm3 + pmaddwd xmm0, [GOTOFF(eax,PW_F0114_F0250)] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250) + pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250) + + movdqa xmm3, [GOTOFF(eax,PD_ONEHALF)] ; xmm3=[PD_ONEHALF] + + paddd xmm0, xmm1 + paddd xmm4, xmm7 + paddd xmm0, xmm3 + paddd xmm4, xmm3 + psrld xmm0, SCALEBITS ; xmm0=YOL + psrld xmm4, SCALEBITS ; xmm4=YOH + packssdw xmm0, xmm4 ; xmm0=YO + + movdqa xmm4, xmm6 + punpcklwd xmm6, xmm2 + punpckhwd xmm4, xmm2 + pmaddwd xmm6, [GOTOFF(eax,PW_F0114_F0250)] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250) + pmaddwd xmm4, [GOTOFF(eax,PW_F0114_F0250)] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250) + + movdqa xmm2, [GOTOFF(eax,PD_ONEHALF)] ; xmm2=[PD_ONEHALF] + + paddd xmm6, XMMWORD [wk(0)] + paddd xmm4, XMMWORD [wk(1)] + paddd xmm6, xmm2 + paddd xmm4, xmm2 + psrld xmm6, SCALEBITS ; xmm6=YEL + psrld xmm4, SCALEBITS ; xmm4=YEH + packssdw xmm6, xmm4 ; xmm6=YE + + psllw xmm0, BYTE_BIT + por xmm6, xmm0 ; xmm6=Y + movdqa XMMWORD [edi], xmm6 ; Save Y + + sub ecx, byte SIZEOF_XMMWORD + add esi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr + add edi, byte SIZEOF_XMMWORD ; outptr0 + cmp ecx, byte SIZEOF_XMMWORD + jae near .columnloop + test ecx, ecx + jnz near .column_ld1 + + pop ecx ; col + pop esi + pop edi + POPPIC eax + + add esi, byte SIZEOF_JSAMPROW ; input_buf + add edi, byte SIZEOF_JSAMPROW + dec eax ; num_rows + jg near .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jchuff-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jchuff-sse2.asm new file mode 100644 index 0000000..ed194dd --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jchuff-sse2.asm @@ -0,0 +1,758 @@ +; +; jchuff-sse2.asm - Huffman entropy encoding (SSE2) +; +; Copyright (C) 2009-2011, 2014-2017, 2019, 2024, D. R. Commander. +; Copyright (C) 2015, Matthieu Darbois. +; Copyright (C) 2018, Matthias Räncker. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains an SSE2 implementation for Huffman coding of one block. +; The following code is based on jchuff.c; see jchuff.c for more details. + +%include "jsimdext.inc" + +struc working_state +.next_output_byte: resp 1 ; => next byte to write in buffer +.free_in_buffer: resp 1 ; # of byte spaces remaining in buffer +.cur.put_buffer.simd resq 1 ; current bit accumulation buffer +.cur.free_bits resd 1 ; # of bits available in it +.cur.last_dc_val resd 4 ; last DC coef for each component +.cinfo: resp 1 ; dump_buffer needs access to this +endstruc + +struc c_derived_tbl +.ehufco: resd 256 ; code for each symbol +.ehufsi: resb 256 ; length of code for each symbol +; If no code has been allocated for a symbol S, ehufsi[S] contains 0 +endstruc + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + GLOBAL_DATA(jconst_huff_encode_one_block) + +EXTN(jconst_huff_encode_one_block): + + ALIGNZ 32 + +jpeg_mask_bits dq 0x0000, 0x0001, 0x0003, 0x0007 + dq 0x000f, 0x001f, 0x003f, 0x007f + dq 0x00ff, 0x01ff, 0x03ff, 0x07ff + dq 0x0fff, 0x1fff, 0x3fff, 0x7fff + +times 1 << 14 db 15 +times 1 << 13 db 14 +times 1 << 12 db 13 +times 1 << 11 db 12 +times 1 << 10 db 11 +times 1 << 9 db 10 +times 1 << 8 db 9 +times 1 << 7 db 8 +times 1 << 6 db 7 +times 1 << 5 db 6 +times 1 << 4 db 5 +times 1 << 3 db 4 +times 1 << 2 db 3 +times 1 << 1 db 2 +times 1 << 0 db 1 +times 1 db 0 +GLOBAL_DATA(jpeg_nbits_table) +EXTN(jpeg_nbits_table): +times 1 db 0 +times 1 << 0 db 1 +times 1 << 1 db 2 +times 1 << 2 db 3 +times 1 << 3 db 4 +times 1 << 4 db 5 +times 1 << 5 db 6 +times 1 << 6 db 7 +times 1 << 7 db 8 +times 1 << 8 db 9 +times 1 << 9 db 10 +times 1 << 10 db 11 +times 1 << 11 db 12 +times 1 << 12 db 13 +times 1 << 13 db 14 +times 1 << 14 db 15 + + ALIGNZ 32 + +%ifdef PIC +%define NBITS(x) nbits_base + x +%else +%define NBITS(x) EXTN(jpeg_nbits_table) + x +%endif +%define MASK_BITS(x) NBITS((x) * 8) + (jpeg_mask_bits - EXTN(jpeg_nbits_table)) + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%define mm_put_buffer mm0 +%define mm_all_0xff mm1 +%define mm_temp mm2 +%define mm_nbits mm3 +%define mm_code_bits mm3 +%define mm_code mm4 +%define mm_overflow_bits mm5 +%define mm_save_nbits mm6 + +; Shorthand used to describe SIMD operations: +; wN: xmmN treated as eight signed 16-bit values +; wN[i]: perform the same operation on all eight signed 16-bit values, i=0..7 +; bN: xmmN treated as 16 unsigned 8-bit values, or +; mmN treated as eight unsigned 8-bit values +; bN[i]: perform the same operation on all unsigned 8-bit values, +; i=0..15 (SSE register) or i=0..7 (MMX register) +; Contents of SIMD registers are shown in memory order. + +; Fill the bit buffer to capacity with the leading bits from code, then output +; the bit buffer and put the remaining bits from code into the bit buffer. +; +; Usage: +; code - contains the bits to shift into the bit buffer (LSB-aligned) +; %1 - temp register +; %2 - low byte of temp register +; %3 - second byte of temp register +; %4-%8 (optional) - extra instructions to execute before the macro completes +; %9 - the label to which to jump when the macro completes +; +; Upon completion, free_bits will be set to the number of remaining bits from +; code, and put_buffer will contain those remaining bits. temp and code will +; be clobbered. +; +; This macro encodes any 0xFF bytes as 0xFF 0x00, as does the EMIT_BYTE() +; macro in jchuff.c. + +%macro EMIT_QWORD 9 +%define %%temp %1 +%define %%tempb %2 +%define %%temph %3 + add nbits, free_bits ; nbits += free_bits; + neg free_bits ; free_bits = -free_bits; + movq mm_temp, mm_code ; temp = code; + movd mm_nbits, nbits ; nbits --> MMX register + movd mm_overflow_bits, free_bits ; overflow_bits (temp register) = free_bits; + neg free_bits ; free_bits = -free_bits; + psllq mm_put_buffer, mm_nbits ; put_buffer <<= nbits; + psrlq mm_temp, mm_overflow_bits ; temp >>= overflow_bits; + add free_bits, 64 ; free_bits += 64; + por mm_temp, mm_put_buffer ; temp |= put_buffer; +%ifidn %%temp, nbits_base + movd mm_save_nbits, nbits_base ; save nbits_base +%endif + movq mm_code_bits, mm_temp ; code_bits (temp register) = temp; + movq mm_put_buffer, mm_code ; put_buffer = code; + pcmpeqb mm_temp, mm_all_0xff ; b_temp[i] = (b_temp[i] == 0xFF ? 0xFF : 0); + movq mm_code, mm_code_bits ; code = code_bits; + psrlq mm_code_bits, 32 ; code_bits >>= 32; + pmovmskb nbits, mm_temp ; nbits = 0; nbits |= ((b_temp[i] >> 7) << i); + movd %%temp, mm_code_bits ; temp = code_bits; + bswap %%temp ; temp = htonl(temp); + test nbits, nbits ; if (nbits != 0) /* Some 0xFF bytes */ + jnz %%.SLOW ; goto %%.SLOW + mov dword [buffer], %%temp ; *(uint32_t)buffer = temp; +%ifidn %%temp, nbits_base + movd nbits_base, mm_save_nbits ; restore nbits_base +%endif + %4 + movd nbits, mm_code ; nbits = (uint32_t)(code); + %5 + bswap nbits ; nbits = htonl(nbits); + mov dword [buffer + 4], nbits ; *(uint32_t)(buffer + 4) = nbits; + lea buffer, [buffer + 8] ; buffer += 8; + %6 + %7 + %8 + jmp %9 ; return +%%.SLOW: + ; Execute the equivalent of the EMIT_BYTE() macro in jchuff.c for all 8 + ; bytes in the qword. + mov byte [buffer], %%tempb ; buffer[0] = temp[0]; + cmp %%tempb, 0xFF ; Set CF if temp[0] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (temp[0] < 0xFF ? 1 : 0)); + mov byte [buffer], %%temph ; buffer[0] = temp[1]; + cmp %%temph, 0xFF ; Set CF if temp[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (temp[1] < 0xFF ? 1 : 0)); + shr %%temp, 16 ; temp >>= 16; + mov byte [buffer], %%tempb ; buffer[0] = temp[0]; + cmp %%tempb, 0xFF ; Set CF if temp[0] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (temp[0] < 0xFF ? 1 : 0)); + mov byte [buffer], %%temph ; buffer[0] = temp[1]; + cmp %%temph, 0xFF ; Set CF if temp[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (temp[1] < 0xFF ? 1 : 0)); + movd nbits, mm_code ; nbits (temp register) = (uint32_t)(code) +%ifidn %%temp, nbits_base + movd nbits_base, mm_save_nbits ; restore nbits_base +%endif + bswap nbits ; nbits = htonl(nbits) + mov byte [buffer], nbitsb ; buffer[0] = nbits[0]; + cmp nbitsb, 0xFF ; Set CF if nbits[0] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (nbits[0] < 0xFF ? 1 : 0)); + mov byte [buffer], nbitsh ; buffer[0] = nbits[1]; + cmp nbitsh, 0xFF ; Set CF if nbits[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (nbits[1] < 0xFF ? 1 : 0)); + shr nbits, 16 ; nbits >>= 16; + mov byte [buffer], nbitsb ; buffer[0] = nbits[0]; + cmp nbitsb, 0xFF ; Set CF if nbits[0] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (nbits[0] < 0xFF ? 1 : 0)); + mov byte [buffer], nbitsh ; buffer[0] = nbits[1]; + %4 + cmp nbitsh, 0xFF ; Set CF if nbits[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb buffer, -2 ; buffer -= (-2 + (nbits[1] < 0xFF ? 1 : 0)); + %5 + %6 + %7 + %8 + jmp %9 ; return; +%endmacro + +%macro PUSH 1 + push %1 +%assign stack_offset stack_offset + 4 +%endmacro + +%macro POP 1 + pop %1 +%assign stack_offset stack_offset - 4 +%endmacro + +; If PIC is defined, load the address of a symbol defined in this file into a +; register. Equivalent to +; GET_GOT %1 +; lea %1, [GOTOFF(%1, %2)] +; without using the GOT. +; +; Usage: +; %1 - register into which to load the address of the symbol +; %2 - symbol whose address should be loaded +; %3 - optional multi-line macro to execute before the symbol address is loaded +; %4 - optional multi-line macro to execute after the symbol address is loaded +; +; If PIC is not defined, then %3 and %4 are executed in order. + +%macro GET_SYM 2-4 +%ifdef PIC + call %%.geteip +%%.ref: + %4 + add %1, %2 - %%.ref + jmp short %%.done + align 32 +%%.geteip: + %3 4 ; must adjust stack pointer because of call + mov %1, POINTER [esp] + ret + align 32 +%%.done: +%else + %3 0 + %4 +%endif +%endmacro + +; +; Encode a single block's worth of coefficients. +; +; GLOBAL(JOCTET *) +; jsimd_huff_encode_one_block_sse2(working_state *state, JOCTET *buffer, +; JCOEFPTR block, int last_dc_val, +; c_derived_tbl *dctbl, c_derived_tbl *actbl) +; +; Stack layout: +; Function args +; Return address +; Saved ebx +; Saved ebp +; Saved esi +; Saved edi <-- esp_save +; ... +; esp_save +; t_ 64*2 bytes (aligned to 128 bytes) +; +; esp is used (as t) to point into t_ (data in lower indices is not used once +; esp passes over them, so this is signal-safe.) Aligning to 128 bytes allows +; us to find the rest of the data again. +; +; NOTES: +; When shuffling data, we try to avoid pinsrw as much as possible, since it is +; slow on many CPUs. Its reciprocal throughput (issue latency) is 1 even on +; modern CPUs, so chains of pinsrw instructions (even with different outputs) +; can limit performance. pinsrw is a VectorPath instruction on AMD K8 and +; requires 2 µops (with memory operand) on Intel. In either case, only one +; pinsrw instruction can be decoded per cycle (and nothing else if they are +; back-to-back), so out-of-order execution cannot be used to work around long +; pinsrw chains (though for Sandy Bridge and later, this may be less of a +; problem if the code runs from the µop cache.) +; +; We use tzcnt instead of bsf without checking for support. The instruction is +; executed as bsf on CPUs that don't support tzcnt (encoding is equivalent to +; rep bsf.) The destination (first) operand of bsf (and tzcnt on some CPUs) is +; an input dependency (although the behavior is not formally defined, Intel +; CPUs usually leave the destination unmodified if the source is zero.) This +; can prevent out-of-order execution, so we clear the destination before +; invoking tzcnt. +; +; Initial register allocation +; eax - frame --> buffer +; ebx - nbits_base (PIC) / emit_temp +; ecx - dctbl --> size --> state +; edx - block --> nbits +; esi - code_temp --> state --> actbl +; edi - index_temp --> free_bits +; esp - t +; ebp - index + +%define frame eax +%ifdef PIC +%define nbits_base ebx +%endif +%define emit_temp ebx +%define emit_tempb bl +%define emit_temph bh +%define dctbl ecx +%define block edx +%define code_temp esi +%define index_temp edi +%define t esp +%define index ebp + +%assign save_frame DCTSIZE2 * SIZEOF_WORD + +; Step 1: Re-arrange input data according to jpeg_natural_order +; xx 01 02 03 04 05 06 07 xx 01 08 16 09 02 03 10 +; 08 09 10 11 12 13 14 15 17 24 32 25 18 11 04 05 +; 16 17 18 19 20 21 22 23 12 19 26 33 40 48 41 34 +; 24 25 26 27 28 29 30 31 ==> 27 20 13 06 07 14 21 28 +; 32 33 34 35 36 37 38 39 35 42 49 56 57 50 43 36 +; 40 41 42 43 44 45 46 47 29 22 15 23 30 37 44 51 +; 48 49 50 51 52 53 54 55 58 59 52 45 38 31 39 46 +; 56 57 58 59 60 61 62 63 53 60 61 54 47 55 62 63 + + align 32 + GLOBAL_FUNCTION(jsimd_huff_encode_one_block_sse2) + +EXTN(jsimd_huff_encode_one_block_sse2): + +%assign stack_offset 0 +%define arg_state 4 + stack_offset +%define arg_buffer 8 + stack_offset +%define arg_block 12 + stack_offset +%define arg_last_dc_val 16 + stack_offset +%define arg_dctbl 20 + stack_offset +%define arg_actbl 24 + stack_offset + + ;X: X = code stream + mov block, [esp + arg_block] + PUSH ebx + PUSH ebp + movups xmm3, XMMWORD [block + 0 * SIZEOF_WORD] ;D: w3 = xx 01 02 03 04 05 06 07 + PUSH esi + PUSH edi + movdqa xmm0, xmm3 ;A: w0 = xx 01 02 03 04 05 06 07 + mov frame, esp + lea t, [frame - (save_frame + 4)] + movups xmm1, XMMWORD [block + 8 * SIZEOF_WORD] ;B: w1 = 08 09 10 11 12 13 14 15 + and t, -DCTSIZE2 * SIZEOF_WORD ; t = &t_[0] + mov [t + save_frame], frame + pxor xmm4, xmm4 ;A: w4[i] = 0; + punpckldq xmm0, xmm1 ;A: w0 = xx 01 08 09 02 03 10 11 + pshuflw xmm0, xmm0, 11001001b ;A: w0 = 01 08 xx 09 02 03 10 11 + pinsrw xmm0, word [block + 16 * SIZEOF_WORD], 2 ;A: w0 = 01 08 16 09 02 03 10 11 + punpckhdq xmm3, xmm1 ;D: w3 = 04 05 12 13 06 07 14 15 + punpcklqdq xmm1, xmm3 ;B: w1 = 08 09 10 11 04 05 12 13 + pinsrw xmm0, word [block + 17 * SIZEOF_WORD], 7 ;A: w0 = 01 08 16 09 02 03 10 17 + ;A: (Row 0, offset 1) + pcmpgtw xmm4, xmm0 ;A: w4[i] = (w0[i] < 0 ? -1 : 0); + paddw xmm0, xmm4 ;A: w0[i] += w4[i]; + movaps XMMWORD [t + 0 * SIZEOF_WORD], xmm0 ;A: t[i] = w0[i]; + + movq xmm2, qword [block + 24 * SIZEOF_WORD] ;B: w2 = 24 25 26 27 -- -- -- -- + pshuflw xmm2, xmm2, 11011000b ;B: w2 = 24 26 25 27 -- -- -- -- + pslldq xmm1, 1 * SIZEOF_WORD ;B: w1 = -- 08 09 10 11 04 05 12 + movups xmm5, XMMWORD [block + 48 * SIZEOF_WORD] ;H: w5 = 48 49 50 51 52 53 54 55 + movsd xmm1, xmm2 ;B: w1 = 24 26 25 27 11 04 05 12 + punpcklqdq xmm2, xmm5 ;C: w2 = 24 26 25 27 48 49 50 51 + pinsrw xmm1, word [block + 32 * SIZEOF_WORD], 1 ;B: w1 = 24 32 25 27 11 04 05 12 + pxor xmm4, xmm4 ;A: w4[i] = 0; + psrldq xmm3, 2 * SIZEOF_WORD ;D: w3 = 12 13 06 07 14 15 -- -- + pcmpeqw xmm0, xmm4 ;A: w0[i] = (w0[i] == 0 ? -1 : 0); + pinsrw xmm1, word [block + 18 * SIZEOF_WORD], 3 ;B: w1 = 24 32 25 18 11 04 05 12 + ; (Row 1, offset 1) + pcmpgtw xmm4, xmm1 ;B: w4[i] = (w1[i] < 0 ? -1 : 0); + paddw xmm1, xmm4 ;B: w1[i] += w4[i]; + movaps XMMWORD [t + 8 * SIZEOF_WORD], xmm1 ;B: t[i+8] = w1[i]; + pxor xmm4, xmm4 ;B: w4[i] = 0; + pcmpeqw xmm1, xmm4 ;B: w1[i] = (w1[i] == 0 ? -1 : 0); + + packsswb xmm0, xmm1 ;AB: b0[i] = w0[i], b0[i+8] = w1[i] + ; w/ signed saturation + + pinsrw xmm3, word [block + 20 * SIZEOF_WORD], 0 ;D: w3 = 20 13 06 07 14 15 -- -- + pinsrw xmm3, word [block + 21 * SIZEOF_WORD], 5 ;D: w3 = 20 13 06 07 14 21 -- -- + pinsrw xmm3, word [block + 28 * SIZEOF_WORD], 6 ;D: w3 = 20 13 06 07 14 21 28 -- + pinsrw xmm3, word [block + 35 * SIZEOF_WORD], 7 ;D: w3 = 20 13 06 07 14 21 28 35 + ; (Row 3, offset 1) + pcmpgtw xmm4, xmm3 ;D: w4[i] = (w3[i] < 0 ? -1 : 0); + paddw xmm3, xmm4 ;D: w3[i] += w4[i]; + movaps XMMWORD [t + 24 * SIZEOF_WORD], xmm3 ;D: t[i+24] = w3[i]; + pxor xmm4, xmm4 ;D: w4[i] = 0; + pcmpeqw xmm3, xmm4 ;D: w3[i] = (w3[i] == 0 ? -1 : 0); + + pinsrw xmm2, word [block + 19 * SIZEOF_WORD], 0 ;C: w2 = 19 26 25 27 48 49 50 51 + pinsrw xmm2, word [block + 33 * SIZEOF_WORD], 2 ;C: w2 = 19 26 33 27 48 49 50 51 + pinsrw xmm2, word [block + 40 * SIZEOF_WORD], 3 ;C: w2 = 19 26 33 40 48 49 50 51 + pinsrw xmm2, word [block + 41 * SIZEOF_WORD], 5 ;C: w2 = 19 26 33 40 48 41 50 51 + pinsrw xmm2, word [block + 34 * SIZEOF_WORD], 6 ;C: w2 = 19 26 33 40 48 41 34 51 + pinsrw xmm2, word [block + 27 * SIZEOF_WORD], 7 ;C: w2 = 19 26 33 40 48 41 34 27 + ; (Row 2, offset 1) + pcmpgtw xmm4, xmm2 ;C: w4[i] = (w2[i] < 0 ? -1 : 0); + paddw xmm2, xmm4 ;C: w2[i] += w4[i]; + movsx code_temp, word [block] ;Z: code_temp = block[0]; + +; %1 - stack pointer adjustment +%macro GET_SYM_BEFORE 1 + movaps XMMWORD [t + 16 * SIZEOF_WORD + %1], xmm2 + ;C: t[i+16] = w2[i]; + pxor xmm4, xmm4 ;C: w4[i] = 0; + pcmpeqw xmm2, xmm4 ;C: w2[i] = (w2[i] == 0 ? -1 : 0); + sub code_temp, [frame + arg_last_dc_val] ;Z: code_temp -= last_dc_val; + + packsswb xmm2, xmm3 ;CD: b2[i] = w2[i], b2[i+8] = w3[i] + ; w/ signed saturation + + movdqa xmm3, xmm5 ;H: w3 = 48 49 50 51 52 53 54 55 + pmovmskb index_temp, xmm2 ;Z: index_temp = 0; index_temp |= ((b2[i] >> 7) << i); + pmovmskb index, xmm0 ;Z: index = 0; index |= ((b0[i] >> 7) << i); + movups xmm0, XMMWORD [block + 56 * SIZEOF_WORD] ;H: w0 = 56 57 58 59 60 61 62 63 + punpckhdq xmm3, xmm0 ;H: w3 = 52 53 60 61 54 55 62 63 + shl index_temp, 16 ;Z: index_temp <<= 16; + psrldq xmm3, 1 * SIZEOF_WORD ;H: w3 = 53 60 61 54 55 62 63 -- + pxor xmm2, xmm2 ;H: w2[i] = 0; + pshuflw xmm3, xmm3, 00111001b ;H: w3 = 60 61 54 53 55 62 63 -- + or index, index_temp ;Z: index |= index_temp; +%undef index_temp +%define free_bits edi +%endmacro + +%macro GET_SYM_AFTER 0 + movq xmm1, qword [block + 44 * SIZEOF_WORD] ;G: w1 = 44 45 46 47 -- -- -- -- + unpcklps xmm5, xmm0 ;E: w5 = 48 49 56 57 50 51 58 59 + pxor xmm0, xmm0 ;H: w0[i] = 0; + not index ;Z: index = ~index; + pinsrw xmm3, word [block + 47 * SIZEOF_WORD], 3 ;H: w3 = 60 61 54 47 55 62 63 -- + ; (Row 7, offset 1) + pcmpgtw xmm2, xmm3 ;H: w2[i] = (w3[i] < 0 ? -1 : 0); + mov dctbl, [frame + arg_dctbl] + paddw xmm3, xmm2 ;H: w3[i] += w2[i]; + movaps XMMWORD [t + 56 * SIZEOF_WORD], xmm3 ;H: t[i+56] = w3[i]; + movq xmm4, qword [block + 36 * SIZEOF_WORD] ;G: w4 = 36 37 38 39 -- -- -- -- + pcmpeqw xmm3, xmm0 ;H: w3[i] = (w3[i] == 0 ? -1 : 0); + punpckldq xmm4, xmm1 ;G: w4 = 36 37 44 45 38 39 46 47 + movdqa xmm1, xmm4 ;F: w1 = 36 37 44 45 38 39 46 47 + pcmpeqw mm_all_0xff, mm_all_0xff ;Z: all_0xff[i] = 0xFF; +%endmacro + + GET_SYM nbits_base, EXTN(jpeg_nbits_table), GET_SYM_BEFORE, GET_SYM_AFTER + + psrldq xmm4, 1 * SIZEOF_WORD ;G: w4 = 37 44 45 38 39 46 47 -- + shufpd xmm1, xmm5, 10b ;F: w1 = 36 37 44 45 50 51 58 59 + pshufhw xmm4, xmm4, 11010011b ;G: w4 = 37 44 45 38 -- 39 46 -- + pslldq xmm1, 1 * SIZEOF_WORD ;F: w1 = -- 36 37 44 45 50 51 58 + pinsrw xmm4, word [block + 59 * SIZEOF_WORD], 0 ;G: w4 = 59 44 45 38 -- 39 46 -- + pshufd xmm1, xmm1, 11011000b ;F: w1 = -- 36 45 50 37 44 51 58 + cmp code_temp, 1 << 31 ;Z: Set CF if code_temp < 0x80000000, + ;Z: i.e. if code_temp is positive + pinsrw xmm4, word [block + 52 * SIZEOF_WORD], 1 ;G: w4 = 59 52 45 38 -- 39 46 -- + movlps xmm1, qword [block + 20 * SIZEOF_WORD] ;F: w1 = 20 21 22 23 37 44 51 58 + pinsrw xmm4, word [block + 31 * SIZEOF_WORD], 4 ;G: w4 = 59 52 45 38 31 39 46 -- + pshuflw xmm1, xmm1, 01110010b ;F: w1 = 22 20 23 21 37 44 51 58 + pinsrw xmm4, word [block + 53 * SIZEOF_WORD], 7 ;G: w4 = 59 52 45 38 31 39 46 53 + ; (Row 6, offset 1) + adc code_temp, -1 ;Z: code_temp += -1 + (code_temp >= 0 ? 1 : 0); + pxor xmm2, xmm2 ;G: w2[i] = 0; + pcmpgtw xmm0, xmm4 ;G: w0[i] = (w4[i] < 0 ? -1 : 0); + pinsrw xmm1, word [block + 15 * SIZEOF_WORD], 1 ;F: w1 = 22 15 23 21 37 44 51 58 + paddw xmm4, xmm0 ;G: w4[i] += w0[i]; + movaps XMMWORD [t + 48 * SIZEOF_WORD], xmm4 ;G: t[48+i] = w4[i]; + movd mm_temp, code_temp ;Z: temp = code_temp + pinsrw xmm1, word [block + 30 * SIZEOF_WORD], 3 ;F: w1 = 22 15 23 30 37 44 51 58 + ; (Row 5, offset 1) + pcmpeqw xmm4, xmm2 ;G: w4[i] = (w4[i] == 0 ? -1 : 0); + + packsswb xmm4, xmm3 ;GH: b4[i] = w4[i], b4[i+8] = w3[i] + ; w/ signed saturation + + lea t, [t - SIZEOF_WORD] ;Z: t = &t[-1] + pxor xmm0, xmm0 ;F: w0[i] = 0; + pcmpgtw xmm2, xmm1 ;F: w2[i] = (w1[i] < 0 ? -1 : 0); + paddw xmm1, xmm2 ;F: w1[i] += w2[i]; + movaps XMMWORD [t + (40+1) * SIZEOF_WORD], xmm1 ;F: t[40+i] = w1[i]; + pcmpeqw xmm1, xmm0 ;F: w1[i] = (w1[i] == 0 ? -1 : 0); + pinsrw xmm5, word [block + 42 * SIZEOF_WORD], 0 ;E: w5 = 42 49 56 57 50 51 58 59 + pinsrw xmm5, word [block + 43 * SIZEOF_WORD], 5 ;E: w5 = 42 49 56 57 50 43 58 59 + pinsrw xmm5, word [block + 36 * SIZEOF_WORD], 6 ;E: w5 = 42 49 56 57 50 43 36 59 + pinsrw xmm5, word [block + 29 * SIZEOF_WORD], 7 ;E: w5 = 42 49 56 57 50 43 36 29 + ; (Row 4, offset 1) +%undef block +%define nbits edx +%define nbitsb dl +%define nbitsh dh + movzx nbits, byte [NBITS(code_temp)] ;Z: nbits = JPEG_NBITS(code_temp); +%undef code_temp +%define state esi + pxor xmm2, xmm2 ;E: w2[i] = 0; + mov state, [frame + arg_state] + movd mm_nbits, nbits ;Z: nbits --> MMX register + pcmpgtw xmm0, xmm5 ;E: w0[i] = (w5[i] < 0 ? -1 : 0); + movd mm_code, dword [dctbl + c_derived_tbl.ehufco + nbits * 4] + ;Z: code = dctbl->ehufco[nbits]; +%define size ecx +%define sizeb cl +%define sizeh ch + paddw xmm5, xmm0 ;E: w5[i] += w0[i]; + movaps XMMWORD [t + (32+1) * SIZEOF_WORD], xmm5 ;E: t[32+i] = w5[i]; + movzx size, byte [dctbl + c_derived_tbl.ehufsi + nbits] + ;Z: size = dctbl->ehufsi[nbits]; +%undef dctbl + pcmpeqw xmm5, xmm2 ;E: w5[i] = (w5[i] == 0 ? -1 : 0); + + packsswb xmm5, xmm1 ;EF: b5[i] = w5[i], b5[i+8] = w1[i] + ; w/ signed saturation + + movq mm_put_buffer, [state + working_state.cur.put_buffer.simd] + ;Z: put_buffer = state->cur.put_buffer.simd; + mov free_bits, [state + working_state.cur.free_bits] + ;Z: free_bits = state->cur.free_bits; +%undef state +%define actbl esi + mov actbl, [frame + arg_actbl] +%define buffer eax + mov buffer, [frame + arg_buffer] +%undef frame + jmp .BEGIN + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +; size <= 32, so this is not really a loop +.BRLOOP1: ; .BRLOOP1: + movzx nbits, byte [actbl + c_derived_tbl.ehufsi + 0xf0] + ; nbits = actbl->ehufsi[0xf0]; + movd mm_code, dword [actbl + c_derived_tbl.ehufco + 0xf0 * 4] + ; code = actbl->ehufco[0xf0]; + and index, 0x7ffffff ; clear index if size == 32 + sub size, 16 ; size -= 16; + sub free_bits, nbits ; if ((free_bits -= nbits) <= 0) + jle .EMIT_BRLOOP1 ; goto .EMIT_BRLOOP1; + movd mm_nbits, nbits ; nbits --> MMX register + psllq mm_put_buffer, mm_nbits ; put_buffer <<= nbits; + por mm_put_buffer, mm_code ; put_buffer |= code; + jmp .ERLOOP1 ; goto .ERLOOP1; + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +%ifdef PIC + times 6 nop +%else + times 2 nop +%endif +.BLOOP1: ; do { /* size = # of zero bits/elements to skip */ +; if size == 32, index remains unchanged. Correct in .BRLOOP. + shr index, sizeb ; index >>= size; + lea t, [t + size * SIZEOF_WORD] ; t += size; + cmp size, 16 ; if (size > 16) + jg .BRLOOP1 ; goto .BRLOOP1; +.ERLOOP1: ; .ERLOOP1: + movsx nbits, word [t] ; nbits = *t; +%ifdef PIC + add size, size ; size += size; +%else + lea size, [size * 2] ; size += size; +%endif + movd mm_temp, nbits ; temp = nbits; + movzx nbits, byte [NBITS(nbits)] ; nbits = JPEG_NBITS(nbits); + lea size, [size * 8 + nbits] ; size = size * 8 + nbits; + movd mm_nbits, nbits ; nbits --> MMX register + movd mm_code, dword [actbl + c_derived_tbl.ehufco + (size - 16) * 4] + ; code = actbl->ehufco[size-16]; + movzx size, byte [actbl + c_derived_tbl.ehufsi + (size - 16)] + ; size = actbl->ehufsi[size-16]; +.BEGIN: ; .BEGIN: + pand mm_temp, [MASK_BITS(nbits)] ; temp &= (1 << nbits) - 1; + psllq mm_code, mm_nbits ; code <<= nbits; + add nbits, size ; nbits += size; + por mm_code, mm_temp ; code |= temp; + sub free_bits, nbits ; if ((free_bits -= nbits) <= 0) + jle .EMIT_ERLOOP1 ; insert code, flush buffer, init size, goto .BLOOP1 + xor size, size ; size = 0; /* kill tzcnt input dependency */ + tzcnt size, index ; size = # of trailing 0 bits in index + movd mm_nbits, nbits ; nbits --> MMX register + psllq mm_put_buffer, mm_nbits ; put_buffer <<= nbits; + inc size ; ++size; + por mm_put_buffer, mm_code ; put_buffer |= code; + test index, index + jnz .BLOOP1 ; } while (index != 0); +; Round 2 +; t points to the last used word, possibly below t_ if the previous index had 32 zero bits. +.ELOOP1: ; .ELOOP1: + pmovmskb size, xmm4 ; size = 0; size |= ((b4[i] >> 7) << i); + pmovmskb index, xmm5 ; index = 0; index |= ((b5[i] >> 7) << i); + shl size, 16 ; size <<= 16; + or index, size ; index |= size; + not index ; index = ~index; + lea nbits, [t + (1 + DCTSIZE2) * SIZEOF_WORD] + ; nbits = t + 1 + 64; + and nbits, -DCTSIZE2 * SIZEOF_WORD ; nbits &= -128; /* now points to &t_[64] */ + sub nbits, t ; nbits -= t; + shr nbits, 1 ; nbits >>= 1; /* # of leading 0 bits in old index + 33 */ + tzcnt size, index ; size = # of trailing 0 bits in index + inc size ; ++size; + test index, index ; if (index == 0) + jz .ELOOP2 ; goto .ELOOP2; +; NOTE: size == 32 cannot happen, since the last element is always 0. + shr index, sizeb ; index >>= size; + lea size, [size + nbits - 33] ; size = size + nbits - 33; + lea t, [t + size * SIZEOF_WORD] ; t += size; + cmp size, 16 ; if (size <= 16) + jle .ERLOOP2 ; goto .ERLOOP2; +.BRLOOP2: ; do { + movzx nbits, byte [actbl + c_derived_tbl.ehufsi + 0xf0] + ; nbits = actbl->ehufsi[0xf0]; + sub size, 16 ; size -= 16; + movd mm_code, dword [actbl + c_derived_tbl.ehufco + 0xf0 * 4] + ; code = actbl->ehufco[0xf0]; + sub free_bits, nbits ; if ((free_bits -= nbits) <= 0) + jle .EMIT_BRLOOP2 ; insert code and flush put_buffer + movd mm_nbits, nbits ; else { nbits --> MMX register + psllq mm_put_buffer, mm_nbits ; put_buffer <<= nbits; + por mm_put_buffer, mm_code ; put_buffer |= code; + cmp size, 16 ; if (size <= 16) + jle .ERLOOP2 ; goto .ERLOOP2; + jmp .BRLOOP2 ; } while (1); + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +.BLOOP2: ; do { /* size = # of zero bits/elements to skip */ + shr index, sizeb ; index >>= size; + lea t, [t + size * SIZEOF_WORD] ; t += size; + cmp size, 16 ; if (size > 16) + jg .BRLOOP2 ; goto .BRLOOP2; +.ERLOOP2: ; .ERLOOP2: + movsx nbits, word [t] ; nbits = *t; + add size, size ; size += size; + movd mm_temp, nbits ; temp = nbits; + movzx nbits, byte [NBITS(nbits)] ; nbits = JPEG_NBITS(nbits); + movd mm_nbits, nbits ; nbits --> MMX register + lea size, [size * 8 + nbits] ; size = size * 8 + nbits; + movd mm_code, dword [actbl + c_derived_tbl.ehufco + (size - 16) * 4] + ; code = actbl->ehufco[size-16]; + movzx size, byte [actbl + c_derived_tbl.ehufsi + (size - 16)] + ; size = actbl->ehufsi[size-16]; + psllq mm_code, mm_nbits ; code <<= nbits; + pand mm_temp, [MASK_BITS(nbits)] ; temp &= (1 << nbits) - 1; + lea nbits, [nbits + size] ; nbits += size; + por mm_code, mm_temp ; code |= temp; + xor size, size ; size = 0; /* kill tzcnt input dependency */ + sub free_bits, nbits ; if ((free_bits -= nbits) <= 0) + jle .EMIT_ERLOOP2 ; insert code, flush buffer, init size, goto .BLOOP2 + tzcnt size, index ; size = # of trailing 0 bits in index + movd mm_nbits, nbits ; nbits --> MMX register + psllq mm_put_buffer, mm_nbits ; put_buffer <<= nbits; + inc size ; ++size; + por mm_put_buffer, mm_code ; put_buffer |= code; + test index, index + jnz .BLOOP2 ; } while (index != 0); +.ELOOP2: ; .ELOOP2: + mov nbits, t ; nbits = t; + lea t, [t + SIZEOF_WORD] ; t = &t[1]; + and nbits, DCTSIZE2 * SIZEOF_WORD - 1 ; nbits &= 127; + and t, -DCTSIZE2 * SIZEOF_WORD ; t &= -128; /* t = &t_[0]; */ + cmp nbits, (DCTSIZE2 - 2) * SIZEOF_WORD ; if (nbits != 62 * 2) + je .EFN ; { + movd mm_code, dword [actbl + c_derived_tbl.ehufco + 0] + ; code = actbl->ehufco[0]; + movzx nbits, byte [actbl + c_derived_tbl.ehufsi + 0] + ; nbits = actbl->ehufsi[0]; + sub free_bits, nbits ; if ((free_bits -= nbits) <= 0) + jg .EFN_SKIP_EMIT_CODE ; { + EMIT_QWORD size, sizeb, sizeh, , , , , , .EFN ; insert code, flush put_buffer + align 16 +.EFN_SKIP_EMIT_CODE: ; } else { + movd mm_nbits, nbits ; nbits --> MMX register + psllq mm_put_buffer, mm_nbits ; put_buffer <<= nbits; + por mm_put_buffer, mm_code ; put_buffer |= code; +.EFN: ; } } +%define frame esp + mov frame, [t + save_frame] +%define state ecx + mov state, [frame + arg_state] + movq [state + working_state.cur.put_buffer.simd], mm_put_buffer + ; state->cur.put_buffer.simd = put_buffer; + emms + mov [state + working_state.cur.free_bits], free_bits + ; state->cur.free_bits = free_bits; + POP edi + POP esi + POP ebp + POP ebx + ret + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +.EMIT_BRLOOP1: + EMIT_QWORD emit_temp, emit_tempb, emit_temph, , , , , , \ + .ERLOOP1 + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +.EMIT_ERLOOP1: + EMIT_QWORD size, sizeb, sizeh, \ + { xor size, size }, \ + { tzcnt size, index }, \ + { inc size }, \ + { test index, index }, \ + { jnz .BLOOP1 }, \ + .ELOOP1 + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +.EMIT_BRLOOP2: + EMIT_QWORD emit_temp, emit_tempb, emit_temph, , , , \ + { cmp size, 16 }, \ + { jle .ERLOOP2 }, \ + .BRLOOP2 + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +.EMIT_ERLOOP2: + EMIT_QWORD size, sizeb, sizeh, \ + { xor size, size }, \ + { tzcnt size, index }, \ + { inc size }, \ + { test index, index }, \ + { jnz .BLOOP2 }, \ + .ELOOP2 + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcphuff-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcphuff-sse2.asm new file mode 100644 index 0000000..19a183f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcphuff-sse2.asm @@ -0,0 +1,658 @@ +; +; jcphuff-sse2.asm - prepare data for progressive Huffman encoding (SSE2) +; +; Copyright (C) 2016, 2018, Matthieu Darbois +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains an SSE2 implementation of data preparation for progressive +; Huffman encoding. See jcphuff.c for more details. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +; -------------------------------------------------------------------------- +; Macros to load data for jsimd_encode_mcu_AC_first_prepare_sse2() and +; jsimd_encode_mcu_AC_refine_prepare_sse2() + +%macro LOAD16 0 + pxor N0, N0 + pxor N1, N1 + + mov T0, INT [LUT + 0*SIZEOF_INT] + mov T1, INT [LUT + 8*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 0 + pinsrw X1, word [BLOCK + T1 * 2], 0 + + mov T0, INT [LUT + 1*SIZEOF_INT] + mov T1, INT [LUT + 9*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 1 + pinsrw X1, word [BLOCK + T1 * 2], 1 + + mov T0, INT [LUT + 2*SIZEOF_INT] + mov T1, INT [LUT + 10*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 2 + pinsrw X1, word [BLOCK + T1 * 2], 2 + + mov T0, INT [LUT + 3*SIZEOF_INT] + mov T1, INT [LUT + 11*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 3 + pinsrw X1, word [BLOCK + T1 * 2], 3 + + mov T0, INT [LUT + 4*SIZEOF_INT] + mov T1, INT [LUT + 12*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 4 + pinsrw X1, word [BLOCK + T1 * 2], 4 + + mov T0, INT [LUT + 5*SIZEOF_INT] + mov T1, INT [LUT + 13*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 5 + pinsrw X1, word [BLOCK + T1 * 2], 5 + + mov T0, INT [LUT + 6*SIZEOF_INT] + mov T1, INT [LUT + 14*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 6 + pinsrw X1, word [BLOCK + T1 * 2], 6 + + mov T0, INT [LUT + 7*SIZEOF_INT] + mov T1, INT [LUT + 15*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 7 + pinsrw X1, word [BLOCK + T1 * 2], 7 +%endmacro + +%macro LOAD15 0 + pxor N0, N0 + pxor N1, N1 + pxor X1, X1 + + mov T0, INT [LUT + 0*SIZEOF_INT] + mov T1, INT [LUT + 8*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 0 + pinsrw X1, word [BLOCK + T1 * 2], 0 + + mov T0, INT [LUT + 1*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 1 + + mov T0, INT [LUT + 2*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 2 + + mov T0, INT [LUT + 3*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 3 + + mov T0, INT [LUT + 4*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 4 + + mov T0, INT [LUT + 5*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 5 + + mov T0, INT [LUT + 6*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 6 + + mov T0, INT [LUT + 7*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 7 + + cmp LENEND, 2 + jl %%.ELOAD15 + mov T1, INT [LUT + 9*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 1 + + cmp LENEND, 3 + jl %%.ELOAD15 + mov T1, INT [LUT + 10*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 2 + + cmp LENEND, 4 + jl %%.ELOAD15 + mov T1, INT [LUT + 11*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 3 + + cmp LENEND, 5 + jl %%.ELOAD15 + mov T1, INT [LUT + 12*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 4 + + cmp LENEND, 6 + jl %%.ELOAD15 + mov T1, INT [LUT + 13*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 5 + + cmp LENEND, 7 + jl %%.ELOAD15 + mov T1, INT [LUT + 14*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 6 +%%.ELOAD15: +%endmacro + +%macro LOAD8 0 + pxor N0, N0 + + mov T0, INT [LUT + 0*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 0 + + mov T0, INT [LUT + 1*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 1 + + mov T0, INT [LUT + 2*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 2 + + mov T0, INT [LUT + 3*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 3 + + mov T0, INT [LUT + 4*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 4 + + mov T0, INT [LUT + 5*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 5 + + mov T0, INT [LUT + 6*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 6 + + mov T0, INT [LUT + 7*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 7 +%endmacro + +%macro LOAD7 0 + pxor N0, N0 + pxor X0, X0 + + mov T1, INT [LUT + 0*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 0 + + cmp LENEND, 2 + jl %%.ELOAD7 + mov T1, INT [LUT + 1*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 1 + + cmp LENEND, 3 + jl %%.ELOAD7 + mov T1, INT [LUT + 2*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 2 + + cmp LENEND, 4 + jl %%.ELOAD7 + mov T1, INT [LUT + 3*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 3 + + cmp LENEND, 5 + jl %%.ELOAD7 + mov T1, INT [LUT + 4*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 4 + + cmp LENEND, 6 + jl %%.ELOAD7 + mov T1, INT [LUT + 5*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 5 + + cmp LENEND, 7 + jl %%.ELOAD7 + mov T1, INT [LUT + 6*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 6 +%%.ELOAD7: +%endmacro + +%macro REDUCE0 0 + movdqa xmm0, XMMWORD [VALUES + ( 0*2)] + movdqa xmm1, XMMWORD [VALUES + ( 8*2)] + movdqa xmm2, XMMWORD [VALUES + (16*2)] + movdqa xmm3, XMMWORD [VALUES + (24*2)] + movdqa xmm4, XMMWORD [VALUES + (32*2)] + movdqa xmm5, XMMWORD [VALUES + (40*2)] + movdqa xmm6, XMMWORD [VALUES + (48*2)] + + pcmpeqw xmm0, ZERO + pcmpeqw xmm1, ZERO + pcmpeqw xmm2, ZERO + pcmpeqw xmm3, ZERO + pcmpeqw xmm4, ZERO + pcmpeqw xmm5, ZERO + pcmpeqw xmm6, ZERO + pcmpeqw xmm7, XMMWORD [VALUES + (56*2)] + + packsswb xmm0, xmm1 + packsswb xmm2, xmm3 + packsswb xmm4, xmm5 + packsswb xmm6, xmm7 + + pmovmskb eax, xmm0 + pmovmskb ecx, xmm2 + pmovmskb edx, xmm4 + pmovmskb esi, xmm6 + + shl ecx, 16 + shl esi, 16 + + or eax, ecx + or edx, esi + + not eax + not edx + + mov edi, ZEROBITS + + mov INT [edi], eax + mov INT [edi+SIZEOF_INT], edx +%endmacro + +; +; Prepare data for jsimd_encode_mcu_AC_first(). +; +; GLOBAL(void) +; jsimd_encode_mcu_AC_first_prepare_sse2(const JCOEF *block, +; const int *jpeg_natural_order_start, +; int Sl, int Al, JCOEF *values, +; size_t *zerobits) +; +; eax + 8 = const JCOEF *block +; eax + 12 = const int *jpeg_natural_order_start +; eax + 16 = int Sl +; eax + 20 = int Al +; eax + 24 = JCOEF *values +; eax + 28 = size_t *zerobits + +%define ZERO xmm7 +%define X0 xmm0 +%define X1 xmm1 +%define N0 xmm2 +%define N1 xmm3 +%define AL xmm4 +%define K eax +%define LENEND eax +%define LUT ebx +%define T0 ecx +%define T1 edx +%define BLOCK esi +%define VALUES edi +%define LEN ebp + +%define ZEROBITS INT [esp + 5 * 4] + + align 32 + GLOBAL_FUNCTION(jsimd_encode_mcu_AC_first_prepare_sse2) + +EXTN(jsimd_encode_mcu_AC_first_prepare_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + sub esp, 4 + push ebx + push ecx +; push edx ; need not be preserved + push esi + push edi + push ebp + + mov BLOCK, INT [eax + 8] + mov LUT, INT [eax + 12] + mov VALUES, INT [eax + 24] + movd AL, INT [eax + 20] + mov T0, INT [eax + 28] + mov ZEROBITS, T0 + mov LEN, INT [eax + 16] + pxor ZERO, ZERO + mov K, LEN + and K, -16 + shr K, 4 + jz .ELOOP16 +.BLOOP16: + LOAD16 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + pxor N0, X0 + pxor N1, X1 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + movdqa XMMWORD [VALUES + (8 + DCTSIZE2) * 2], N1 + add VALUES, 16*2 + add LUT, 16*SIZEOF_INT + dec K + jnz .BLOOP16 + test LEN, 15 + je .PADDING +.ELOOP16: + mov LENEND, LEN + and LENEND, 7 + + test LEN, 8 + jz .TRY7 + test LEN, 7 + jz .TRY8 + + LOAD15 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + pxor N0, X0 + pxor N1, X1 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + movdqa XMMWORD [VALUES + (8 + DCTSIZE2) * 2], N1 + add VALUES, 16*2 + jmp .PADDING +.TRY8: + LOAD8 + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + pxor N0, X0 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + add VALUES, 8*2 + jmp .PADDING +.TRY7: + LOAD7 + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + pxor N0, X0 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + add VALUES, 8*2 +.PADDING: + mov K, LEN + add K, 7 + and K, -8 + shr K, 3 + sub K, DCTSIZE2/8 + jz .EPADDING + align 16 +.ZEROLOOP: + movdqa XMMWORD [VALUES + 0], ZERO + add VALUES, 8*2 + inc K + jnz .ZEROLOOP +.EPADDING: + sub VALUES, DCTSIZE2*2 + + REDUCE0 + + pop ebp + pop edi + pop esi +; pop edx ; need not be preserved + pop ecx + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +%undef ZERO +%undef X0 +%undef X1 +%undef N0 +%undef N1 +%undef AL +%undef K +%undef LUT +%undef T0 +%undef T1 +%undef BLOCK +%undef VALUES +%undef LEN + +; +; Prepare data for jsimd_encode_mcu_AC_refine(). +; +; GLOBAL(int) +; jsimd_encode_mcu_AC_refine_prepare_sse2(const JCOEF *block, +; const int *jpeg_natural_order_start, +; int Sl, int Al, JCOEF *absvalues, +; size_t *bits) +; +; eax + 8 = const JCOEF *block +; eax + 12 = const int *jpeg_natural_order_start +; eax + 16 = int Sl +; eax + 20 = int Al +; eax + 24 = JCOEF *values +; eax + 28 = size_t *bits + +%define ZERO xmm7 +%define ONE xmm5 +%define X0 xmm0 +%define X1 xmm1 +%define N0 xmm2 +%define N1 xmm3 +%define AL xmm4 +%define K eax +%define LENEND eax +%define LUT ebx +%define T0 ecx +%define T0w cx +%define T1 edx +%define BLOCK esi +%define VALUES edi +%define KK ebp + +%define ZEROBITS INT [esp + 5 * 4] +%define EOB INT [esp + 5 * 4 + 4] +%define LEN INT [esp + 5 * 4 + 8] + + align 32 + GLOBAL_FUNCTION(jsimd_encode_mcu_AC_refine_prepare_sse2) + +EXTN(jsimd_encode_mcu_AC_refine_prepare_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + sub esp, 16 + push ebx + push ecx +; push edx ; need not be preserved + push esi + push edi + push ebp + + pcmpeqw ONE, ONE + psrlw ONE, 15 + mov BLOCK, INT [eax + 8] + mov LUT, INT [eax + 12] + mov VALUES, INT [eax + 24] + movd AL, INT [eax + 20] + mov T0, INT [eax + 28] + mov K, INT [eax + 16] + mov INT [T0 + 2 * SIZEOF_INT], -1 + mov INT [T0 + 3 * SIZEOF_INT], -1 + mov ZEROBITS, T0 + mov LEN, K + pxor ZERO, ZERO + and K, -16 + mov EOB, 0 + xor KK, KK + shr K, 4 + jz .ELOOPR16 +.BLOOPR16: + LOAD16 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + pcmpeqw X0, ONE + pcmpeqw X1, ONE + packsswb N0, N1 + packsswb X0, X1 + pmovmskb T0, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + mov T1, ZEROBITS + not T0 + mov word [T1 + 2 * SIZEOF_INT + KK], T0w + pmovmskb T1, X0 ; idx = _mm_movemask_epi8(x1); + bsr T1, T1 ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER16 ; if (idx) { + lea T1, [T1+KK*8] + mov EOB, T1 ; EOB = k + idx; +.CONTINUER16: + add VALUES, 16*2 + add LUT, 16*SIZEOF_INT + add KK, 2 + dec K + jnz .BLOOPR16 + test LEN, 15 + je .PADDINGR +.ELOOPR16: + mov LENEND, LEN + + test LENEND, 8 + jz .TRYR7 + test LENEND, 7 + jz .TRYR8 + + and LENEND, 7 + LOAD15 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + pcmpeqw X0, ONE + pcmpeqw X1, ONE + packsswb N0, N1 + packsswb X0, X1 + pmovmskb T0, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + mov T1, ZEROBITS + not T0 + mov word [T1 + 2 * SIZEOF_INT + KK], T0w + pmovmskb T1, X0 ; idx = _mm_movemask_epi8(x1); + bsr T1, T1 ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER15 ; if (idx) { + lea T1, [T1+KK*8] + mov EOB, T1 ; EOB = k + idx; +.CONTINUER15: + add VALUES, 16*2 + jmp .PADDINGR +.TRYR8: + LOAD8 + + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + pcmpeqw X0, ONE + packsswb N0, ZERO + packsswb X0, ZERO + pmovmskb T0, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + mov T1, ZEROBITS + not T0 + mov word [T1 + 2 * SIZEOF_INT + KK], T0w + pmovmskb T1, X0 ; idx = _mm_movemask_epi8(x1); + bsr T1, T1 ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER8 ; if (idx) { + lea T1, [T1+KK*8] + mov EOB, T1 ; EOB = k + idx; +.CONTINUER8: + add VALUES, 8*2 + jmp .PADDINGR +.TRYR7: + and LENEND, 7 + LOAD7 + + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + pcmpeqw X0, ONE + packsswb N0, ZERO + packsswb X0, ZERO + pmovmskb T0, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + mov T1, ZEROBITS + not T0 + mov word [T1 + 2 * SIZEOF_INT + KK], T0w + pmovmskb T1, X0 ; idx = _mm_movemask_epi8(x1); + bsr T1, T1 ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER7 ; if (idx) { + lea T1, [T1+KK*8] + mov EOB, T1 ; EOB = k + idx; +.CONTINUER7: + add VALUES, 8*2 +.PADDINGR: + mov K, LEN + add K, 7 + and K, -8 + shr K, 3 + sub K, DCTSIZE2/8 + jz .EPADDINGR + align 16 +.ZEROLOOPR: + movdqa XMMWORD [VALUES + 0], ZERO + add VALUES, 8*2 + inc K + jnz .ZEROLOOPR +.EPADDINGR: + sub VALUES, DCTSIZE2*2 + + REDUCE0 + + mov eax, EOB + + pop ebp + pop edi + pop esi +; pop edx ; need not be preserved + pop ecx + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +%undef ZERO +%undef ONE +%undef X0 +%undef X1 +%undef N0 +%undef N1 +%undef AL +%undef K +%undef KK +%undef EOB +%undef SIGN +%undef LUT +%undef T0 +%undef T1 +%undef BLOCK +%undef VALUES +%undef LEN +%undef LENEND + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-avx2.asm new file mode 100644 index 0000000..5019829 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-avx2.asm @@ -0,0 +1,384 @@ +; +; jcsample.asm - downsampling (AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Downsample pixel values of a single component. +; This version handles the common case of 2:1 horizontal and 1:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v1_downsample_avx2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +%define img_width(b) (b) + 8 ; JDIMENSION image_width +%define max_v_samp(b) (b) + 12 ; int max_v_samp_factor +%define v_samp(b) (b) + 16 ; JDIMENSION v_samp_factor +%define width_blks(b) (b) + 20 ; JDIMENSION width_in_blocks +%define input_data(b) (b) + 24 ; JSAMPARRAY input_data +%define output_data(b) (b) + 28 ; JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_downsample_avx2) + +EXTN(jsimd_h2v1_downsample_avx2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov ecx, JDIMENSION [width_blks(ebp)] + shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols) + jz near .return + + mov edx, JDIMENSION [img_width(ebp)] + + ; -- expand_right_edge + + push ecx + shl ecx, 1 ; output_cols * 2 + sub ecx, edx + jle short .expand_end + + mov eax, INT [max_v_samp(ebp)] + test eax, eax + jle short .expand_end + + cld + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + ALIGNX 16, 7 +.expandloop: + push eax + push ecx + + mov edi, JSAMPROW [esi] + add edi, edx + mov al, JSAMPLE [edi-1] + + rep stosb + + pop ecx + pop eax + + add esi, byte SIZEOF_JSAMPROW + dec eax + jg short .expandloop + +.expand_end: + pop ecx ; output_cols + + ; -- h2v1_downsample + + mov eax, JDIMENSION [v_samp(ebp)] ; rowctr + test eax, eax + jle near .return + + mov edx, 0x00010000 ; bias pattern + vmovd xmm7, edx + vpshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1} + vperm2i128 ymm7, ymm7, ymm7, 0 ; ymm7={xmm7, xmm7} + vpcmpeqw ymm6, ymm6, ymm6 + vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..} + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, JSAMPARRAY [output_data(ebp)] ; output_data + ALIGNX 16, 7 +.rowloop: + push ecx + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + + cmp ecx, byte SIZEOF_YMMWORD + jae short .columnloop + ALIGNX 16, 7 + +.columnloop_r24: + ; ecx can possibly be 8, 16, 24 + cmp ecx, 24 + jne .columnloop_r16 + vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu xmm1, XMMWORD [esi+1*SIZEOF_YMMWORD] + mov ecx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r16: + cmp ecx, 16 + jne .columnloop_r8 + vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD] + vpxor ymm1, ymm1, ymm1 + mov ecx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r8: + vmovdqu xmm0, XMMWORD[esi+0*SIZEOF_YMMWORD] + vpxor ymm1, ymm1, ymm1 + mov ecx, SIZEOF_YMMWORD + jmp short .downsample + ALIGNX 16, 7 + +.columnloop: + vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [esi+1*SIZEOF_YMMWORD] + +.downsample: + vpsrlw ymm2, ymm0, BYTE_BIT + vpand ymm0, ymm0, ymm6 + vpsrlw ymm3, ymm1, BYTE_BIT + vpand ymm1, ymm1, ymm6 + + vpaddw ymm0, ymm0, ymm2 + vpaddw ymm1, ymm1, ymm3 + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm1, ymm1, ymm7 + vpsrlw ymm0, ymm0, 1 + vpsrlw ymm1, ymm1, 1 + + vpackuswb ymm0, ymm0, ymm1 + vpermq ymm0, ymm0, 0xd8 + + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0 + + sub ecx, byte SIZEOF_YMMWORD ; outcol + add esi, byte 2*SIZEOF_YMMWORD ; inptr + add edi, byte 1*SIZEOF_YMMWORD ; outptr + cmp ecx, byte SIZEOF_YMMWORD + jae short .columnloop + test ecx, ecx + jnz near .columnloop_r24 + + pop esi + pop edi + pop ecx + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec eax ; rowctr + jg near .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Downsample pixel values of a single component. +; This version handles the standard case of 2:1 horizontal and 2:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v2_downsample_avx2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +%define img_width(b) (b) + 8 ; JDIMENSION image_width +%define max_v_samp(b) (b) + 12 ; int max_v_samp_factor +%define v_samp(b) (b) + 16 ; JDIMENSION v_samp_factor +%define width_blks(b) (b) + 20 ; JDIMENSION width_in_blocks +%define input_data(b) (b) + 24 ; JSAMPARRAY input_data +%define output_data(b) (b) + 28 ; JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_downsample_avx2) + +EXTN(jsimd_h2v2_downsample_avx2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov ecx, JDIMENSION [width_blks(ebp)] + shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols) + jz near .return + + mov edx, JDIMENSION [img_width(ebp)] + + ; -- expand_right_edge + + push ecx + shl ecx, 1 ; output_cols * 2 + sub ecx, edx + jle short .expand_end + + mov eax, INT [max_v_samp(ebp)] + test eax, eax + jle short .expand_end + + cld + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + ALIGNX 16, 7 +.expandloop: + push eax + push ecx + + mov edi, JSAMPROW [esi] + add edi, edx + mov al, JSAMPLE [edi-1] + + rep stosb + + pop ecx + pop eax + + add esi, byte SIZEOF_JSAMPROW + dec eax + jg short .expandloop + +.expand_end: + pop ecx ; output_cols + + ; -- h2v2_downsample + + mov eax, JDIMENSION [v_samp(ebp)] ; rowctr + test eax, eax + jle near .return + + mov edx, 0x00020001 ; bias pattern + vmovd xmm7, edx + vpcmpeqw ymm6, ymm6, ymm6 + vpshufd xmm7, xmm7, 0x00 ; ymm7={1, 2, 1, 2, 1, 2, 1, 2} + vperm2i128 ymm7, ymm7, ymm7, 0 + vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..} + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, JSAMPARRAY [output_data(ebp)] ; output_data + ALIGNX 16, 7 +.rowloop: + push ecx + push edi + push esi + + mov edx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0 + mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1 + mov edi, JSAMPROW [edi] ; outptr + + cmp ecx, byte SIZEOF_YMMWORD + jae short .columnloop + ALIGNX 16, 7 + +.columnloop_r24: + cmp ecx, 24 + jne .columnloop_r16 + vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu xmm2, XMMWORD [edx+1*SIZEOF_YMMWORD] + vmovdqu xmm3, XMMWORD [esi+1*SIZEOF_YMMWORD] + mov ecx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r16: + cmp ecx, 16 + jne .columnloop_r8 + vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD] + vpxor ymm2, ymm2, ymm2 + vpxor ymm3, ymm3, ymm3 + mov ecx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r8: + vmovdqu xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD] + vmovdqu xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD] + vpxor ymm2, ymm2, ymm2 + vpxor ymm3, ymm3, ymm3 + mov ecx, SIZEOF_YMMWORD + jmp short .downsample + ALIGNX 16, 7 + +.columnloop: + vmovdqu ymm0, YMMWORD [edx+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD] + vmovdqu ymm2, YMMWORD [edx+1*SIZEOF_YMMWORD] + vmovdqu ymm3, YMMWORD [esi+1*SIZEOF_YMMWORD] + +.downsample: + vpand ymm4, ymm0, ymm6 + vpsrlw ymm0, ymm0, BYTE_BIT + vpand ymm5, ymm1, ymm6 + vpsrlw ymm1, ymm1, BYTE_BIT + vpaddw ymm0, ymm0, ymm4 + vpaddw ymm1, ymm1, ymm5 + + vpand ymm4, ymm2, ymm6 + vpsrlw ymm2, ymm2, BYTE_BIT + vpand ymm5, ymm3, ymm6 + vpsrlw ymm3, ymm3, BYTE_BIT + vpaddw ymm2, ymm2, ymm4 + vpaddw ymm3, ymm3, ymm5 + + vpaddw ymm0, ymm0, ymm1 + vpaddw ymm2, ymm2, ymm3 + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm2, ymm2, ymm7 + vpsrlw ymm0, ymm0, 2 + vpsrlw ymm2, ymm2, 2 + + vpackuswb ymm0, ymm0, ymm2 + vpermq ymm0, ymm0, 0xd8 + + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0 + + sub ecx, byte SIZEOF_YMMWORD ; outcol + add edx, byte 2*SIZEOF_YMMWORD ; inptr0 + add esi, byte 2*SIZEOF_YMMWORD ; inptr1 + add edi, byte 1*SIZEOF_YMMWORD ; outptr + cmp ecx, byte SIZEOF_YMMWORD + jae near .columnloop + test ecx, ecx + jnz near .columnloop_r24 + + pop esi + pop edi + pop ecx + + add esi, byte 2*SIZEOF_JSAMPROW ; input_data + add edi, byte 1*SIZEOF_JSAMPROW ; output_data + dec eax ; rowctr + jg near .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-mmx.asm new file mode 100644 index 0000000..94dd888 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-mmx.asm @@ -0,0 +1,320 @@ +; +; jcsample.asm - downsampling (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Downsample pixel values of a single component. +; This version handles the common case of 2:1 horizontal and 1:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v1_downsample_mmx(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +%define img_width(b) (b) + 8 ; JDIMENSION image_width +%define max_v_samp(b) (b) + 12 ; int max_v_samp_factor +%define v_samp(b) (b) + 16 ; JDIMENSION v_samp_factor +%define width_blks(b) (b) + 20 ; JDIMENSION width_in_blocks +%define input_data(b) (b) + 24 ; JSAMPARRAY input_data +%define output_data(b) (b) + 28 ; JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_downsample_mmx) + +EXTN(jsimd_h2v1_downsample_mmx): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov ecx, JDIMENSION [width_blks(ebp)] + shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols) + jz near .return + + mov edx, JDIMENSION [img_width(ebp)] + + ; -- expand_right_edge + + push ecx + shl ecx, 1 ; output_cols * 2 + sub ecx, edx + jle short .expand_end + + mov eax, INT [max_v_samp(ebp)] + test eax, eax + jle short .expand_end + + cld + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + ALIGNX 16, 7 +.expandloop: + push eax + push ecx + + mov edi, JSAMPROW [esi] + add edi, edx + mov al, JSAMPLE [edi-1] + + rep stosb + + pop ecx + pop eax + + add esi, byte SIZEOF_JSAMPROW + dec eax + jg short .expandloop + +.expand_end: + pop ecx ; output_cols + + ; -- h2v1_downsample + + mov eax, JDIMENSION [v_samp(ebp)] ; rowctr + test eax, eax + jle near .return + + mov edx, 0x00010000 ; bias pattern + movd mm7, edx + pcmpeqw mm6, mm6 + punpckldq mm7, mm7 ; mm7={0, 1, 0, 1} + psrlw mm6, BYTE_BIT ; mm6={0xFF 0x00 0xFF 0x00 ..} + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, JSAMPARRAY [output_data(ebp)] ; output_data + ALIGNX 16, 7 +.rowloop: + push ecx + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + ALIGNX 16, 7 +.columnloop: + + movq mm0, MMWORD [esi+0*SIZEOF_MMWORD] + movq mm1, MMWORD [esi+1*SIZEOF_MMWORD] + movq mm2, mm0 + movq mm3, mm1 + + pand mm0, mm6 + psrlw mm2, BYTE_BIT + pand mm1, mm6 + psrlw mm3, BYTE_BIT + + paddw mm0, mm2 + paddw mm1, mm3 + paddw mm0, mm7 + paddw mm1, mm7 + psrlw mm0, 1 + psrlw mm1, 1 + + packuswb mm0, mm1 + + movq MMWORD [edi+0*SIZEOF_MMWORD], mm0 + + add esi, byte 2*SIZEOF_MMWORD ; inptr + add edi, byte 1*SIZEOF_MMWORD ; outptr + sub ecx, byte SIZEOF_MMWORD ; outcol + jnz short .columnloop + + pop esi + pop edi + pop ecx + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec eax ; rowctr + jg short .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Downsample pixel values of a single component. +; This version handles the standard case of 2:1 horizontal and 2:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v2_downsample_mmx(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +%define img_width(b) (b) + 8 ; JDIMENSION image_width +%define max_v_samp(b) (b) + 12 ; int max_v_samp_factor +%define v_samp(b) (b) + 16 ; JDIMENSION v_samp_factor +%define width_blks(b) (b) + 20 ; JDIMENSION width_in_blocks +%define input_data(b) (b) + 24 ; JSAMPARRAY input_data +%define output_data(b) (b) + 28 ; JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_downsample_mmx) + +EXTN(jsimd_h2v2_downsample_mmx): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov ecx, JDIMENSION [width_blks(ebp)] + shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols) + jz near .return + + mov edx, JDIMENSION [img_width(ebp)] + + ; -- expand_right_edge + + push ecx + shl ecx, 1 ; output_cols * 2 + sub ecx, edx + jle short .expand_end + + mov eax, INT [max_v_samp(ebp)] + test eax, eax + jle short .expand_end + + cld + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + ALIGNX 16, 7 +.expandloop: + push eax + push ecx + + mov edi, JSAMPROW [esi] + add edi, edx + mov al, JSAMPLE [edi-1] + + rep stosb + + pop ecx + pop eax + + add esi, byte SIZEOF_JSAMPROW + dec eax + jg short .expandloop + +.expand_end: + pop ecx ; output_cols + + ; -- h2v2_downsample + + mov eax, JDIMENSION [v_samp(ebp)] ; rowctr + test eax, eax + jle near .return + + mov edx, 0x00020001 ; bias pattern + movd mm7, edx + pcmpeqw mm6, mm6 + punpckldq mm7, mm7 ; mm7={1, 2, 1, 2} + psrlw mm6, BYTE_BIT ; mm6={0xFF 0x00 0xFF 0x00 ..} + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, JSAMPARRAY [output_data(ebp)] ; output_data + ALIGNX 16, 7 +.rowloop: + push ecx + push edi + push esi + + mov edx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0 + mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1 + mov edi, JSAMPROW [edi] ; outptr + ALIGNX 16, 7 +.columnloop: + + movq mm0, MMWORD [edx+0*SIZEOF_MMWORD] + movq mm1, MMWORD [esi+0*SIZEOF_MMWORD] + movq mm2, MMWORD [edx+1*SIZEOF_MMWORD] + movq mm3, MMWORD [esi+1*SIZEOF_MMWORD] + + movq mm4, mm0 + movq mm5, mm1 + pand mm0, mm6 + psrlw mm4, BYTE_BIT + pand mm1, mm6 + psrlw mm5, BYTE_BIT + paddw mm0, mm4 + paddw mm1, mm5 + + movq mm4, mm2 + movq mm5, mm3 + pand mm2, mm6 + psrlw mm4, BYTE_BIT + pand mm3, mm6 + psrlw mm5, BYTE_BIT + paddw mm2, mm4 + paddw mm3, mm5 + + paddw mm0, mm1 + paddw mm2, mm3 + paddw mm0, mm7 + paddw mm2, mm7 + psrlw mm0, 2 + psrlw mm2, 2 + + packuswb mm0, mm2 + + movq MMWORD [edi+0*SIZEOF_MMWORD], mm0 + + add edx, byte 2*SIZEOF_MMWORD ; inptr0 + add esi, byte 2*SIZEOF_MMWORD ; inptr1 + add edi, byte 1*SIZEOF_MMWORD ; outptr + sub ecx, byte SIZEOF_MMWORD ; outcol + jnz near .columnloop + + pop esi + pop edi + pop ecx + + add esi, byte 2*SIZEOF_JSAMPROW ; input_data + add edi, byte 1*SIZEOF_JSAMPROW ; output_data + dec eax ; rowctr + jg near .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-sse2.asm new file mode 100644 index 0000000..eb8808b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jcsample-sse2.asm @@ -0,0 +1,347 @@ +; +; jcsample.asm - downsampling (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Downsample pixel values of a single component. +; This version handles the common case of 2:1 horizontal and 1:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v1_downsample_sse2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +%define img_width(b) (b) + 8 ; JDIMENSION image_width +%define max_v_samp(b) (b) + 12 ; int max_v_samp_factor +%define v_samp(b) (b) + 16 ; JDIMENSION v_samp_factor +%define width_blks(b) (b) + 20 ; JDIMENSION width_in_blocks +%define input_data(b) (b) + 24 ; JSAMPARRAY input_data +%define output_data(b) (b) + 28 ; JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_downsample_sse2) + +EXTN(jsimd_h2v1_downsample_sse2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov ecx, JDIMENSION [width_blks(ebp)] + shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols) + jz near .return + + mov edx, JDIMENSION [img_width(ebp)] + + ; -- expand_right_edge + + push ecx + shl ecx, 1 ; output_cols * 2 + sub ecx, edx + jle short .expand_end + + mov eax, INT [max_v_samp(ebp)] + test eax, eax + jle short .expand_end + + cld + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + ALIGNX 16, 7 +.expandloop: + push eax + push ecx + + mov edi, JSAMPROW [esi] + add edi, edx + mov al, JSAMPLE [edi-1] + + rep stosb + + pop ecx + pop eax + + add esi, byte SIZEOF_JSAMPROW + dec eax + jg short .expandloop + +.expand_end: + pop ecx ; output_cols + + ; -- h2v1_downsample + + mov eax, JDIMENSION [v_samp(ebp)] ; rowctr + test eax, eax + jle near .return + + mov edx, 0x00010000 ; bias pattern + movd xmm7, edx + pcmpeqw xmm6, xmm6 + pshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1} + psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..} + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, JSAMPARRAY [output_data(ebp)] ; output_data + ALIGNX 16, 7 +.rowloop: + push ecx + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + + cmp ecx, byte SIZEOF_XMMWORD + jae short .columnloop + ALIGNX 16, 7 + +.columnloop_r8: + movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD] + pxor xmm1, xmm1 + mov ecx, SIZEOF_XMMWORD + jmp short .downsample + ALIGNX 16, 7 + +.columnloop: + movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqa xmm1, XMMWORD [esi+1*SIZEOF_XMMWORD] + +.downsample: + movdqa xmm2, xmm0 + movdqa xmm3, xmm1 + + pand xmm0, xmm6 + psrlw xmm2, BYTE_BIT + pand xmm1, xmm6 + psrlw xmm3, BYTE_BIT + + paddw xmm0, xmm2 + paddw xmm1, xmm3 + paddw xmm0, xmm7 + paddw xmm1, xmm7 + psrlw xmm0, 1 + psrlw xmm1, 1 + + packuswb xmm0, xmm1 + + movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0 + + sub ecx, byte SIZEOF_XMMWORD ; outcol + add esi, byte 2*SIZEOF_XMMWORD ; inptr + add edi, byte 1*SIZEOF_XMMWORD ; outptr + cmp ecx, byte SIZEOF_XMMWORD + jae short .columnloop + test ecx, ecx + jnz short .columnloop_r8 + + pop esi + pop edi + pop ecx + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec eax ; rowctr + jg near .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Downsample pixel values of a single component. +; This version handles the standard case of 2:1 horizontal and 2:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v2_downsample_sse2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +%define img_width(b) (b) + 8 ; JDIMENSION image_width +%define max_v_samp(b) (b) + 12 ; int max_v_samp_factor +%define v_samp(b) (b) + 16 ; JDIMENSION v_samp_factor +%define width_blks(b) (b) + 20 ; JDIMENSION width_in_blocks +%define input_data(b) (b) + 24 ; JSAMPARRAY input_data +%define output_data(b) (b) + 28 ; JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_downsample_sse2) + +EXTN(jsimd_h2v2_downsample_sse2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov ecx, JDIMENSION [width_blks(ebp)] + shl ecx, 3 ; imul ecx,DCTSIZE (ecx = output_cols) + jz near .return + + mov edx, JDIMENSION [img_width(ebp)] + + ; -- expand_right_edge + + push ecx + shl ecx, 1 ; output_cols * 2 + sub ecx, edx + jle short .expand_end + + mov eax, INT [max_v_samp(ebp)] + test eax, eax + jle short .expand_end + + cld + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + ALIGNX 16, 7 +.expandloop: + push eax + push ecx + + mov edi, JSAMPROW [esi] + add edi, edx + mov al, JSAMPLE [edi-1] + + rep stosb + + pop ecx + pop eax + + add esi, byte SIZEOF_JSAMPROW + dec eax + jg short .expandloop + +.expand_end: + pop ecx ; output_cols + + ; -- h2v2_downsample + + mov eax, JDIMENSION [v_samp(ebp)] ; rowctr + test eax, eax + jle near .return + + mov edx, 0x00020001 ; bias pattern + movd xmm7, edx + pcmpeqw xmm6, xmm6 + pshufd xmm7, xmm7, 0x00 ; xmm7={1, 2, 1, 2, 1, 2, 1, 2} + psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..} + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, JSAMPARRAY [output_data(ebp)] ; output_data + ALIGNX 16, 7 +.rowloop: + push ecx + push edi + push esi + + mov edx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0 + mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1 + mov edi, JSAMPROW [edi] ; outptr + + cmp ecx, byte SIZEOF_XMMWORD + jae short .columnloop + ALIGNX 16, 7 + +.columnloop_r8: + movdqa xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD] + movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD] + pxor xmm2, xmm2 + pxor xmm3, xmm3 + mov ecx, SIZEOF_XMMWORD + jmp short .downsample + ALIGNX 16, 7 + +.columnloop: + movdqa xmm0, XMMWORD [edx+0*SIZEOF_XMMWORD] + movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqa xmm2, XMMWORD [edx+1*SIZEOF_XMMWORD] + movdqa xmm3, XMMWORD [esi+1*SIZEOF_XMMWORD] + +.downsample: + movdqa xmm4, xmm0 + movdqa xmm5, xmm1 + pand xmm0, xmm6 + psrlw xmm4, BYTE_BIT + pand xmm1, xmm6 + psrlw xmm5, BYTE_BIT + paddw xmm0, xmm4 + paddw xmm1, xmm5 + + movdqa xmm4, xmm2 + movdqa xmm5, xmm3 + pand xmm2, xmm6 + psrlw xmm4, BYTE_BIT + pand xmm3, xmm6 + psrlw xmm5, BYTE_BIT + paddw xmm2, xmm4 + paddw xmm3, xmm5 + + paddw xmm0, xmm1 + paddw xmm2, xmm3 + paddw xmm0, xmm7 + paddw xmm2, xmm7 + psrlw xmm0, 2 + psrlw xmm2, 2 + + packuswb xmm0, xmm2 + + movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0 + + sub ecx, byte SIZEOF_XMMWORD ; outcol + add edx, byte 2*SIZEOF_XMMWORD ; inptr0 + add esi, byte 2*SIZEOF_XMMWORD ; inptr1 + add edi, byte 1*SIZEOF_XMMWORD ; outptr + cmp ecx, byte SIZEOF_XMMWORD + jae near .columnloop + test ecx, ecx + jnz near .columnloop_r8 + + pop esi + pop edi + pop ecx + + add esi, byte 2*SIZEOF_JSAMPROW ; input_data + add edi, byte 1*SIZEOF_JSAMPROW ; output_data + dec eax ; rowctr + jg near .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-avx2.asm new file mode 100644 index 0000000..fd79b79 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-avx2.asm @@ -0,0 +1,511 @@ +; +; jdcolext.asm - colorspace conversion (AVX2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2012, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_ycc_rgb_convert_avx2(JDIMENSION out_width, JSAMPIMAGE input_buf, +; JDIMENSION input_row, JSAMPARRAY output_buf, +; int num_rows) +; + +%define out_width(b) (b) + 8 ; JDIMENSION out_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define input_row(b) (b) + 16 ; JDIMENSION input_row +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_YMMWORD + ; ymmword wk[WK_NUM] +%define WK_NUM 2 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_ycc_rgb_convert_avx2) + +EXTN(jsimd_ycc_rgb_convert_avx2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [out_width(eax)] ; num_cols + test ecx, ecx + jz near .return + + push ecx + + mov edi, JSAMPIMAGE [input_buf(eax)] + mov ecx, JDIMENSION [input_row(eax)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + lea esi, [esi+ecx*SIZEOF_JSAMPROW] + lea ebx, [ebx+ecx*SIZEOF_JSAMPROW] + lea edx, [edx+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov edi, JSAMPARRAY [output_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + push eax + push edi + push edx + push ebx + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr0 + mov ebx, JSAMPROW [ebx] ; inptr1 + mov edx, JSAMPROW [edx] ; inptr2 + mov edi, JSAMPROW [edi] ; outptr + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + ALIGNX 16, 7 +.columnloop: + + vmovdqu ymm5, YMMWORD [ebx] ; ymm5=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV) + vmovdqu ymm1, YMMWORD [edx] ; ymm1=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpcmpeqw ymm0, ymm0, ymm0 + vpcmpeqw ymm7, ymm7, ymm7 + vpsrlw ymm0, ymm0, BYTE_BIT ; ymm0={0xFF 0x00 0xFF 0x00 ..} + vpsllw ymm7, ymm7, 7 ; ymm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + vpand ymm4, ymm0, ymm5 ; ymm4=Cb(02468ACEGIKMOQSU)=CbE + vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Cb(13579BDFHJLNPRTV)=CbO + vpand ymm0, ymm0, ymm1 ; ymm0=Cr(02468ACEGIKMOQSU)=CrE + vpsrlw ymm1, ymm1, BYTE_BIT ; ymm1=Cr(13579BDFHJLNPRTV)=CrO + + vpaddw ymm2, ymm4, ymm7 + vpaddw ymm3, ymm5, ymm7 + vpaddw ymm6, ymm0, ymm7 + vpaddw ymm7, ymm1, ymm7 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbE + vpaddw ymm5, ymm3, ymm3 ; ymm5=2*CbO + vpaddw ymm0, ymm6, ymm6 ; ymm0=2*CrE + vpaddw ymm1, ymm7, ymm7 ; ymm1=2*CrO + + vpmulhw ymm4, ymm4, [GOTOFF(eax,PW_MF0228)] ; ymm4=(2*CbE * -FIX(0.22800)) + vpmulhw ymm5, ymm5, [GOTOFF(eax,PW_MF0228)] ; ymm5=(2*CbO * -FIX(0.22800)) + vpmulhw ymm0, ymm0, [GOTOFF(eax,PW_F0402)] ; ymm0=(2*CrE * FIX(0.40200)) + vpmulhw ymm1, ymm1, [GOTOFF(eax,PW_F0402)] ; ymm1=(2*CrO * FIX(0.40200)) + + vpaddw ymm4, ymm4, [GOTOFF(eax,PW_ONE)] + vpaddw ymm5, ymm5, [GOTOFF(eax,PW_ONE)] + vpsraw ymm4, ymm4, 1 ; ymm4=(CbE * -FIX(0.22800)) + vpsraw ymm5, ymm5, 1 ; ymm5=(CbO * -FIX(0.22800)) + vpaddw ymm0, ymm0, [GOTOFF(eax,PW_ONE)] + vpaddw ymm1, ymm1, [GOTOFF(eax,PW_ONE)] + vpsraw ymm0, ymm0, 1 ; ymm0=(CrE * FIX(0.40200)) + vpsraw ymm1, ymm1, 1 ; ymm1=(CrO * FIX(0.40200)) + + vpaddw ymm4, ymm4, ymm2 + vpaddw ymm5, ymm5, ymm3 + vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbE * FIX(1.77200))=(B-Y)E + vpaddw ymm5, ymm5, ymm3 ; ymm5=(CbO * FIX(1.77200))=(B-Y)O + vpaddw ymm0, ymm0, ymm6 ; ymm0=(CrE * FIX(1.40200))=(R-Y)E + vpaddw ymm1, ymm1, ymm7 ; ymm1=(CrO * FIX(1.40200))=(R-Y)O + + vmovdqa YMMWORD [wk(0)], ymm4 ; wk(0)=(B-Y)E + vmovdqa YMMWORD [wk(1)], ymm5 ; wk(1)=(B-Y)O + + vpunpckhwd ymm4, ymm2, ymm6 + vpunpcklwd ymm2, ymm2, ymm6 + vpmaddwd ymm2, ymm2, [GOTOFF(eax,PW_MF0344_F0285)] + vpmaddwd ymm4, ymm4, [GOTOFF(eax,PW_MF0344_F0285)] + vpunpckhwd ymm5, ymm3, ymm7 + vpunpcklwd ymm3, ymm3, ymm7 + vpmaddwd ymm3, ymm3, [GOTOFF(eax,PW_MF0344_F0285)] + vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF0344_F0285)] + + vpaddd ymm2, ymm2, [GOTOFF(eax,PD_ONEHALF)] + vpaddd ymm4, ymm4, [GOTOFF(eax,PD_ONEHALF)] + vpsrad ymm2, ymm2, SCALEBITS + vpsrad ymm4, ymm4, SCALEBITS + vpaddd ymm3, ymm3, [GOTOFF(eax,PD_ONEHALF)] + vpaddd ymm5, ymm5, [GOTOFF(eax,PD_ONEHALF)] + vpsrad ymm3, ymm3, SCALEBITS + vpsrad ymm5, ymm5, SCALEBITS + + vpackssdw ymm2, ymm2, ymm4 ; ymm2=CbE*-FIX(0.344)+CrE*FIX(0.285) + vpackssdw ymm3, ymm3, ymm5 ; ymm3=CbO*-FIX(0.344)+CrO*FIX(0.285) + vpsubw ymm2, ymm2, ymm6 ; ymm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E + vpsubw ymm3, ymm3, ymm7 ; ymm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O + + vmovdqu ymm5, YMMWORD [esi] ; ymm5=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpcmpeqw ymm4, ymm4, ymm4 + vpsrlw ymm4, ymm4, BYTE_BIT ; ymm4={0xFF 0x00 0xFF 0x00 ..} + vpand ymm4, ymm4, ymm5 ; ymm4=Y(02468ACEGIKMOQSU)=YE + vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Y(13579BDFHJLNPRTV)=YO + + vpaddw ymm0, ymm0, ymm4 ; ymm0=((R-Y)E+YE)=RE=R(02468ACEGIKMOQSU) + vpaddw ymm1, ymm1, ymm5 ; ymm1=((R-Y)O+YO)=RO=R(13579BDFHJLNPRTV) + vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********) + vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********) + + vpaddw ymm2, ymm2, ymm4 ; ymm2=((G-Y)E+YE)=GE=G(02468ACEGIKMOQSU) + vpaddw ymm3, ymm3, ymm5 ; ymm3=((G-Y)O+YO)=GO=G(13579BDFHJLNPRTV) + vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********) + vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********) + + vpaddw ymm4, ymm4, YMMWORD [wk(0)] ; ymm4=(YE+(B-Y)E)=BE=B(02468ACEGIKMOQSU) + vpaddw ymm5, ymm5, YMMWORD [wk(1)] ; ymm5=(YO+(B-Y)O)=BO=B(13579BDFHJLNPRTV) + vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********) + vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F + ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V) + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F + ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V) + + vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G + ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --) + vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F + ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07 + ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N) + + vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H + ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --) + + vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H + ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --) + vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G + ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --) + vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18 + ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O) + + vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H + ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --) + vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29 + ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P) + + vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03 + ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J) + vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14 + ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K) + vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29 + ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P) + vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07 + ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N) + + vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B + ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R) + vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C + ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S) + vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H + ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --) + vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V) + + vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + cmp ecx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test edi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF +.out0: + add edi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub ecx, byte SIZEOF_YMMWORD + jz near .nextrow + + add esi, byte SIZEOF_YMMWORD ; inptr0 + add ebx, byte SIZEOF_YMMWORD ; inptr1 + add edx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st64: + lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE + cmp ecx, byte 2*SIZEOF_YMMWORD + jb short .column_st32 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + add edi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmF + sub ecx, byte 2*SIZEOF_YMMWORD + jmp short .column_st31 +.column_st32: + cmp ecx, byte SIZEOF_YMMWORD + jb short .column_st31 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + add edi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub ecx, byte SIZEOF_YMMWORD + jmp short .column_st31 +.column_st31: + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st15 + vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + add edi, byte SIZEOF_XMMWORD ; outptr + vperm2i128 ymmA, ymmA, ymmA, 1 + sub ecx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st7 + vmovq XMM_MMWORD [edi], xmmA + add edi, byte SIZEOF_MMWORD + sub ecx, byte SIZEOF_MMWORD + vpsrldq xmmA, xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_DWORD + jb short .column_st3 + vmovd XMM_DWORD [edi], xmmA + add edi, byte SIZEOF_DWORD + sub ecx, byte SIZEOF_DWORD + vpsrldq xmmA, xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of eax to the output when it has enough + ; space. + vmovd eax, xmmA + cmp ecx, byte SIZEOF_WORD + jb short .column_st1 + mov word [edi], ax + add edi, byte SIZEOF_WORD + sub ecx, byte SIZEOF_WORD + shr eax, 16 +.column_st1: + ; Store the lower 1 byte of eax to the output when it has enough + ; space. + test ecx, ecx + jz short .nextrow + mov byte [edi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%else + vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%endif + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **) + ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E + ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U) + vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F + ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V) + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F + ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V) + + vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E + ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36 + ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M) + vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F + ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V) + vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37 + ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N) + + vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + + vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + cmp ecx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test edi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC + vmovntdq YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC + vmovdqu YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH +.out0: + add edi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub ecx, byte SIZEOF_YMMWORD + jz near .nextrow + + add esi, byte SIZEOF_YMMWORD ; inptr0 + add ebx, byte SIZEOF_YMMWORD ; inptr1 + add edx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st64: + cmp ecx, byte SIZEOF_YMMWORD/2 + jb short .column_st32 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + add edi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmC + vmovdqa ymmD, ymmH + sub ecx, byte SIZEOF_YMMWORD/2 +.column_st32: + cmp ecx, byte SIZEOF_YMMWORD/4 + jb short .column_st16 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + add edi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub ecx, byte SIZEOF_YMMWORD/4 +.column_st16: + cmp ecx, byte SIZEOF_YMMWORD/8 + jb short .column_st15 + vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + vperm2i128 ymmA, ymmA, ymmA, 1 + add edi, byte SIZEOF_XMMWORD ; outptr + sub ecx, byte SIZEOF_YMMWORD/8 +.column_st15: + ; Store two pixels (8 bytes) of ymmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_YMMWORD/16 + jb short .column_st7 + vmovq MMWORD [edi], xmmA + add edi, byte SIZEOF_YMMWORD/16*4 + sub ecx, byte SIZEOF_YMMWORD/16 + vpsrldq xmmA, SIZEOF_YMMWORD/16*4 +.column_st7: + ; Store one pixel (4 bytes) of ymmA to the output when it has enough + ; space. + test ecx, ecx + jz short .nextrow + vmovd XMM_DWORD [edi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + + ALIGNX 16, 7 + +.nextrow: + pop ecx + pop esi + pop ebx + pop edx + pop edi + pop eax + + add esi, byte SIZEOF_JSAMPROW + add ebx, byte SIZEOF_JSAMPROW + add edx, byte SIZEOF_JSAMPROW + add edi, byte SIZEOF_JSAMPROW ; output_buf + dec eax ; num_rows + jg near .rowloop + + sfence ; flush the write buffer + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-mmx.asm new file mode 100644 index 0000000..636bd6d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-mmx.asm @@ -0,0 +1,400 @@ +; +; jdcolext.asm - colorspace conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_ycc_rgb_convert_mmx(JDIMENSION out_width, JSAMPIMAGE input_buf, +; JDIMENSION input_row, JSAMPARRAY output_buf, +; int num_rows) +; + +%define out_width(b) (b) + 8 ; JDIMENSION out_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define input_row(b) (b) + 16 ; JDIMENSION input_row +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD + ; mmword wk[WK_NUM] +%define WK_NUM 2 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_ycc_rgb_convert_mmx) + +EXTN(jsimd_ycc_rgb_convert_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [out_width(eax)] ; num_cols + test ecx, ecx + jz near .return + + push ecx + + mov edi, JSAMPIMAGE [input_buf(eax)] + mov ecx, JDIMENSION [input_row(eax)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + lea esi, [esi+ecx*SIZEOF_JSAMPROW] + lea ebx, [ebx+ecx*SIZEOF_JSAMPROW] + lea edx, [edx+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov edi, JSAMPARRAY [output_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + push eax + push edi + push edx + push ebx + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr0 + mov ebx, JSAMPROW [ebx] ; inptr1 + mov edx, JSAMPROW [edx] ; inptr2 + mov edi, JSAMPROW [edi] ; outptr + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + ALIGNX 16, 7 +.columnloop: + + movq mm5, MMWORD [ebx] ; mm5=Cb(01234567) + movq mm1, MMWORD [edx] ; mm1=Cr(01234567) + + pcmpeqw mm4, mm4 + pcmpeqw mm7, mm7 + psrlw mm4, BYTE_BIT + psllw mm7, 7 ; mm7={0xFF80 0xFF80 0xFF80 0xFF80} + movq mm0, mm4 ; mm0=mm4={0xFF 0x00 0xFF 0x00 ..} + + pand mm4, mm5 ; mm4=Cb(0246)=CbE + psrlw mm5, BYTE_BIT ; mm5=Cb(1357)=CbO + pand mm0, mm1 ; mm0=Cr(0246)=CrE + psrlw mm1, BYTE_BIT ; mm1=Cr(1357)=CrO + + paddw mm4, mm7 + paddw mm5, mm7 + paddw mm0, mm7 + paddw mm1, mm7 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + movq mm2, mm4 ; mm2=CbE + movq mm3, mm5 ; mm3=CbO + paddw mm4, mm4 ; mm4=2*CbE + paddw mm5, mm5 ; mm5=2*CbO + movq mm6, mm0 ; mm6=CrE + movq mm7, mm1 ; mm7=CrO + paddw mm0, mm0 ; mm0=2*CrE + paddw mm1, mm1 ; mm1=2*CrO + + pmulhw mm4, [GOTOFF(eax,PW_MF0228)] ; mm4=(2*CbE * -FIX(0.22800)) + pmulhw mm5, [GOTOFF(eax,PW_MF0228)] ; mm5=(2*CbO * -FIX(0.22800)) + pmulhw mm0, [GOTOFF(eax,PW_F0402)] ; mm0=(2*CrE * FIX(0.40200)) + pmulhw mm1, [GOTOFF(eax,PW_F0402)] ; mm1=(2*CrO * FIX(0.40200)) + + paddw mm4, [GOTOFF(eax,PW_ONE)] + paddw mm5, [GOTOFF(eax,PW_ONE)] + psraw mm4, 1 ; mm4=(CbE * -FIX(0.22800)) + psraw mm5, 1 ; mm5=(CbO * -FIX(0.22800)) + paddw mm0, [GOTOFF(eax,PW_ONE)] + paddw mm1, [GOTOFF(eax,PW_ONE)] + psraw mm0, 1 ; mm0=(CrE * FIX(0.40200)) + psraw mm1, 1 ; mm1=(CrO * FIX(0.40200)) + + paddw mm4, mm2 + paddw mm5, mm3 + paddw mm4, mm2 ; mm4=(CbE * FIX(1.77200))=(B-Y)E + paddw mm5, mm3 ; mm5=(CbO * FIX(1.77200))=(B-Y)O + paddw mm0, mm6 ; mm0=(CrE * FIX(1.40200))=(R-Y)E + paddw mm1, mm7 ; mm1=(CrO * FIX(1.40200))=(R-Y)O + + movq MMWORD [wk(0)], mm4 ; wk(0)=(B-Y)E + movq MMWORD [wk(1)], mm5 ; wk(1)=(B-Y)O + + movq mm4, mm2 + movq mm5, mm3 + punpcklwd mm2, mm6 + punpckhwd mm4, mm6 + pmaddwd mm2, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd mm4, [GOTOFF(eax,PW_MF0344_F0285)] + punpcklwd mm3, mm7 + punpckhwd mm5, mm7 + pmaddwd mm3, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd mm5, [GOTOFF(eax,PW_MF0344_F0285)] + + paddd mm2, [GOTOFF(eax,PD_ONEHALF)] + paddd mm4, [GOTOFF(eax,PD_ONEHALF)] + psrad mm2, SCALEBITS + psrad mm4, SCALEBITS + paddd mm3, [GOTOFF(eax,PD_ONEHALF)] + paddd mm5, [GOTOFF(eax,PD_ONEHALF)] + psrad mm3, SCALEBITS + psrad mm5, SCALEBITS + + packssdw mm2, mm4 ; mm2=CbE*-FIX(0.344)+CrE*FIX(0.285) + packssdw mm3, mm5 ; mm3=CbO*-FIX(0.344)+CrO*FIX(0.285) + psubw mm2, mm6 ; mm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E + psubw mm3, mm7 ; mm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O + + movq mm5, MMWORD [esi] ; mm5=Y(01234567) + + pcmpeqw mm4, mm4 + psrlw mm4, BYTE_BIT ; mm4={0xFF 0x00 0xFF 0x00 ..} + pand mm4, mm5 ; mm4=Y(0246)=YE + psrlw mm5, BYTE_BIT ; mm5=Y(1357)=YO + + paddw mm0, mm4 ; mm0=((R-Y)E+YE)=RE=(R0 R2 R4 R6) + paddw mm1, mm5 ; mm1=((R-Y)O+YO)=RO=(R1 R3 R5 R7) + packuswb mm0, mm0 ; mm0=(R0 R2 R4 R6 ** ** ** **) + packuswb mm1, mm1 ; mm1=(R1 R3 R5 R7 ** ** ** **) + + paddw mm2, mm4 ; mm2=((G-Y)E+YE)=GE=(G0 G2 G4 G6) + paddw mm3, mm5 ; mm3=((G-Y)O+YO)=GO=(G1 G3 G5 G7) + packuswb mm2, mm2 ; mm2=(G0 G2 G4 G6 ** ** ** **) + packuswb mm3, mm3 ; mm3=(G1 G3 G5 G7 ** ** ** **) + + paddw mm4, MMWORD [wk(0)] ; mm4=(YE+(B-Y)E)=BE=(B0 B2 B4 B6) + paddw mm5, MMWORD [wk(1)] ; mm5=(YO+(B-Y)O)=BO=(B1 B3 B5 B7) + packuswb mm4, mm4 ; mm4=(B0 B2 B4 B6 ** ** ** **) + packuswb mm5, mm5 ; mm5=(B1 B3 B5 B7 ** ** ** **) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) + ; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) + ; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **) + ; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **) + + punpcklbw mmA, mmC ; mmA=(00 10 02 12 04 14 06 16) + punpcklbw mmE, mmB ; mmE=(20 01 22 03 24 05 26 07) + punpcklbw mmD, mmF ; mmD=(11 21 13 23 15 25 17 27) + + movq mmG, mmA + movq mmH, mmA + punpcklwd mmA, mmE ; mmA=(00 10 20 01 02 12 22 03) + punpckhwd mmG, mmE ; mmG=(04 14 24 05 06 16 26 07) + + psrlq mmH, 2*BYTE_BIT ; mmH=(02 12 04 14 06 16 -- --) + psrlq mmE, 2*BYTE_BIT ; mmE=(22 03 24 05 26 07 -- --) + + movq mmC, mmD + movq mmB, mmD + punpcklwd mmD, mmH ; mmD=(11 21 02 12 13 23 04 14) + punpckhwd mmC, mmH ; mmC=(15 25 06 16 17 27 -- --) + + psrlq mmB, 2*BYTE_BIT ; mmB=(13 23 15 25 17 27 -- --) + + movq mmF, mmE + punpcklwd mmE, mmB ; mmE=(22 03 13 23 24 05 15 25) + punpckhwd mmF, mmB ; mmF=(26 07 17 27 -- -- -- --) + + punpckldq mmA, mmD ; mmA=(00 10 20 01 11 21 02 12) + punpckldq mmE, mmG ; mmE=(22 03 13 23 04 14 24 05) + punpckldq mmC, mmF ; mmC=(15 25 06 16 26 07 17 27) + + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st16 + + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmE + movq MMWORD [edi+2*SIZEOF_MMWORD], mmC + + sub ecx, byte SIZEOF_MMWORD + jz short .nextrow + + add esi, byte SIZEOF_MMWORD ; inptr0 + add ebx, byte SIZEOF_MMWORD ; inptr1 + add edx, byte SIZEOF_MMWORD ; inptr2 + add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr + jmp near .columnloop + ALIGNX 16, 7 + +.column_st16: + lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE + cmp ecx, byte 2*SIZEOF_MMWORD + jb short .column_st8 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmE + movq mmA, mmC + sub ecx, byte 2*SIZEOF_MMWORD + add edi, byte 2*SIZEOF_MMWORD + jmp short .column_st4 +.column_st8: + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st4 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq mmA, mmE + sub ecx, byte SIZEOF_MMWORD + add edi, byte SIZEOF_MMWORD +.column_st4: + movd eax, mmA + cmp ecx, byte SIZEOF_DWORD + jb short .column_st2 + mov dword [edi+0*SIZEOF_DWORD], eax + psrlq mmA, DWORD_BIT + movd eax, mmA + sub ecx, byte SIZEOF_DWORD + add edi, byte SIZEOF_DWORD +.column_st2: + cmp ecx, byte SIZEOF_WORD + jb short .column_st1 + mov word [edi+0*SIZEOF_WORD], ax + shr eax, WORD_BIT + sub ecx, byte SIZEOF_WORD + add edi, byte SIZEOF_WORD +.column_st1: + cmp ecx, byte SIZEOF_BYTE + jb short .nextrow + mov byte [edi+0*SIZEOF_BYTE], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + pcmpeqb mm6, mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **) + pcmpeqb mm7, mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **) +%else + pxor mm6, mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **) + pxor mm7, mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **) +%endif + ; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) + ; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) + ; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **) + ; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **) + + punpcklbw mmA, mmC ; mmA=(00 10 02 12 04 14 06 16) + punpcklbw mmE, mmG ; mmE=(20 30 22 32 24 34 26 36) + punpcklbw mmB, mmD ; mmB=(01 11 03 13 05 15 07 17) + punpcklbw mmF, mmH ; mmF=(21 31 23 33 25 35 27 37) + + movq mmC, mmA + punpcklwd mmA, mmE ; mmA=(00 10 20 30 02 12 22 32) + punpckhwd mmC, mmE ; mmC=(04 14 24 34 06 16 26 36) + movq mmG, mmB + punpcklwd mmB, mmF ; mmB=(01 11 21 31 03 13 23 33) + punpckhwd mmG, mmF ; mmG=(05 15 25 35 07 17 27 37) + + movq mmD, mmA + punpckldq mmA, mmB ; mmA=(00 10 20 30 01 11 21 31) + punpckhdq mmD, mmB ; mmD=(02 12 22 32 03 13 23 33) + movq mmH, mmC + punpckldq mmC, mmG ; mmC=(04 14 24 34 05 15 25 35) + punpckhdq mmH, mmG ; mmH=(06 16 26 36 07 17 27 37) + + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st16 + + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmD + movq MMWORD [edi+2*SIZEOF_MMWORD], mmC + movq MMWORD [edi+3*SIZEOF_MMWORD], mmH + + sub ecx, byte SIZEOF_MMWORD + jz short .nextrow + + add esi, byte SIZEOF_MMWORD ; inptr0 + add ebx, byte SIZEOF_MMWORD ; inptr1 + add edx, byte SIZEOF_MMWORD ; inptr2 + add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr + jmp near .columnloop + ALIGNX 16, 7 + +.column_st16: + cmp ecx, byte SIZEOF_MMWORD/2 + jb short .column_st8 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmD + movq mmA, mmC + movq mmD, mmH + sub ecx, byte SIZEOF_MMWORD/2 + add edi, byte 2*SIZEOF_MMWORD +.column_st8: + cmp ecx, byte SIZEOF_MMWORD/4 + jb short .column_st4 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq mmA, mmD + sub ecx, byte SIZEOF_MMWORD/4 + add edi, byte 1*SIZEOF_MMWORD +.column_st4: + cmp ecx, byte SIZEOF_MMWORD/8 + jb short .nextrow + movd dword [edi+0*SIZEOF_DWORD], mmA + +%endif ; RGB_PIXELSIZE ; --------------- + + ALIGNX 16, 7 + +.nextrow: + pop ecx + pop esi + pop ebx + pop edx + pop edi + pop eax + + add esi, byte SIZEOF_JSAMPROW + add ebx, byte SIZEOF_JSAMPROW + add edx, byte SIZEOF_JSAMPROW + add edi, byte SIZEOF_JSAMPROW ; output_buf + dec eax ; num_rows + jg near .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-sse2.asm new file mode 100644 index 0000000..0150f2c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolext-sse2.asm @@ -0,0 +1,454 @@ +; +; jdcolext.asm - colorspace conversion (SSE2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2012, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_ycc_rgb_convert_sse2(JDIMENSION out_width, JSAMPIMAGE input_buf, +; JDIMENSION input_row, JSAMPARRAY output_buf, +; int num_rows) +; + +%define out_width(b) (b) + 8 ; JDIMENSION out_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define input_row(b) (b) + 16 ; JDIMENSION input_row +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf +%define num_rows(b) (b) + 24 ; int num_rows + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_ycc_rgb_convert_sse2) + +EXTN(jsimd_ycc_rgb_convert_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [out_width(eax)] ; num_cols + test ecx, ecx + jz near .return + + push ecx + + mov edi, JSAMPIMAGE [input_buf(eax)] + mov ecx, JDIMENSION [input_row(eax)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + lea esi, [esi+ecx*SIZEOF_JSAMPROW] + lea ebx, [ebx+ecx*SIZEOF_JSAMPROW] + lea edx, [edx+ecx*SIZEOF_JSAMPROW] + + pop ecx + + mov edi, JSAMPARRAY [output_buf(eax)] + mov eax, INT [num_rows(eax)] + test eax, eax + jle near .return + ALIGNX 16, 7 +.rowloop: + push eax + push edi + push edx + push ebx + push esi + push ecx ; col + + mov esi, JSAMPROW [esi] ; inptr0 + mov ebx, JSAMPROW [ebx] ; inptr1 + mov edx, JSAMPROW [edx] ; inptr2 + mov edi, JSAMPROW [edi] ; outptr + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + ALIGNX 16, 7 +.columnloop: + + movdqa xmm5, XMMWORD [ebx] ; xmm5=Cb(0123456789ABCDEF) + movdqa xmm1, XMMWORD [edx] ; xmm1=Cr(0123456789ABCDEF) + + pcmpeqw xmm4, xmm4 + pcmpeqw xmm7, xmm7 + psrlw xmm4, BYTE_BIT + psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + movdqa xmm0, xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..} + + pand xmm4, xmm5 ; xmm4=Cb(02468ACE)=CbE + psrlw xmm5, BYTE_BIT ; xmm5=Cb(13579BDF)=CbO + pand xmm0, xmm1 ; xmm0=Cr(02468ACE)=CrE + psrlw xmm1, BYTE_BIT ; xmm1=Cr(13579BDF)=CrO + + paddw xmm4, xmm7 + paddw xmm5, xmm7 + paddw xmm0, xmm7 + paddw xmm1, xmm7 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + movdqa xmm2, xmm4 ; xmm2=CbE + movdqa xmm3, xmm5 ; xmm3=CbO + paddw xmm4, xmm4 ; xmm4=2*CbE + paddw xmm5, xmm5 ; xmm5=2*CbO + movdqa xmm6, xmm0 ; xmm6=CrE + movdqa xmm7, xmm1 ; xmm7=CrO + paddw xmm0, xmm0 ; xmm0=2*CrE + paddw xmm1, xmm1 ; xmm1=2*CrO + + pmulhw xmm4, [GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbE * -FIX(0.22800)) + pmulhw xmm5, [GOTOFF(eax,PW_MF0228)] ; xmm5=(2*CbO * -FIX(0.22800)) + pmulhw xmm0, [GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrE * FIX(0.40200)) + pmulhw xmm1, [GOTOFF(eax,PW_F0402)] ; xmm1=(2*CrO * FIX(0.40200)) + + paddw xmm4, [GOTOFF(eax,PW_ONE)] + paddw xmm5, [GOTOFF(eax,PW_ONE)] + psraw xmm4, 1 ; xmm4=(CbE * -FIX(0.22800)) + psraw xmm5, 1 ; xmm5=(CbO * -FIX(0.22800)) + paddw xmm0, [GOTOFF(eax,PW_ONE)] + paddw xmm1, [GOTOFF(eax,PW_ONE)] + psraw xmm0, 1 ; xmm0=(CrE * FIX(0.40200)) + psraw xmm1, 1 ; xmm1=(CrO * FIX(0.40200)) + + paddw xmm4, xmm2 + paddw xmm5, xmm3 + paddw xmm4, xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E + paddw xmm5, xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O + paddw xmm0, xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E + paddw xmm1, xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E + movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O + + movdqa xmm4, xmm2 + movdqa xmm5, xmm3 + punpcklwd xmm2, xmm6 + punpckhwd xmm4, xmm6 + pmaddwd xmm2, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd xmm4, [GOTOFF(eax,PW_MF0344_F0285)] + punpcklwd xmm3, xmm7 + punpckhwd xmm5, xmm7 + pmaddwd xmm3, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd xmm5, [GOTOFF(eax,PW_MF0344_F0285)] + + paddd xmm2, [GOTOFF(eax,PD_ONEHALF)] + paddd xmm4, [GOTOFF(eax,PD_ONEHALF)] + psrad xmm2, SCALEBITS + psrad xmm4, SCALEBITS + paddd xmm3, [GOTOFF(eax,PD_ONEHALF)] + paddd xmm5, [GOTOFF(eax,PD_ONEHALF)] + psrad xmm3, SCALEBITS + psrad xmm5, SCALEBITS + + packssdw xmm2, xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285) + packssdw xmm3, xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285) + psubw xmm2, xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E + psubw xmm3, xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O + + movdqa xmm5, XMMWORD [esi] ; xmm5=Y(0123456789ABCDEF) + + pcmpeqw xmm4, xmm4 + psrlw xmm4, BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..} + pand xmm4, xmm5 ; xmm4=Y(02468ACE)=YE + psrlw xmm5, BYTE_BIT ; xmm5=Y(13579BDF)=YO + + paddw xmm0, xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE) + paddw xmm1, xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF) + packuswb xmm0, xmm0 ; xmm0=R(02468ACE********) + packuswb xmm1, xmm1 ; xmm1=R(13579BDF********) + + paddw xmm2, xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE) + paddw xmm3, xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF) + packuswb xmm2, xmm2 ; xmm2=G(02468ACE********) + packuswb xmm3, xmm3 ; xmm3=G(13579BDF********) + + paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE) + paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF) + packuswb xmm4, xmm4 ; xmm4=B(02468ACE********) + packuswb xmm5, xmm5 ; xmm5=B(13579BDF********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F) + punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F) + + movdqa xmmG, xmmA + movdqa xmmH, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07) + punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F) + + psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --) + psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --) + + movdqa xmmC, xmmD + movdqa xmmB, xmmD + punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18) + punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --) + + psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --) + + movdqa xmmF, xmmE + punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29) + punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --) + + pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03) + movdqa xmmB, xmmE + punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14) + punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07) + punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29) + + pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B) + movdqa xmmB, xmmF + punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C) + punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F) + punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --) + + punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test edi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF +.out0: + add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub ecx, byte SIZEOF_XMMWORD + jz near .nextrow + + add esi, byte SIZEOF_XMMWORD ; inptr0 + add ebx, byte SIZEOF_XMMWORD ; inptr1 + add edx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st32: + lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE + cmp ecx, byte 2*SIZEOF_XMMWORD + jb short .column_st16 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + add edi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmF + sub ecx, byte 2*SIZEOF_XMMWORD + jmp short .column_st15 +.column_st16: + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st15 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + add edi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub ecx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st7 + movq XMM_MMWORD [edi], xmmA + add edi, byte SIZEOF_MMWORD + sub ecx, byte SIZEOF_MMWORD + psrldq xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_DWORD + jb short .column_st3 + movd XMM_DWORD [edi], xmmA + add edi, byte SIZEOF_DWORD + sub ecx, byte SIZEOF_DWORD + psrldq xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of eax to the output when it has enough + ; space. + movd eax, xmmA + cmp ecx, byte SIZEOF_WORD + jb short .column_st1 + mov word [edi], ax + add edi, byte SIZEOF_WORD + sub ecx, byte SIZEOF_WORD + shr eax, 16 +.column_st1: + ; Store the lower 1 byte of eax to the output when it has enough + ; space. + test ecx, ecx + jz short .nextrow + mov byte [edi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%else + pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%endif + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E) + punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F) + punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F) + + movdqa xmmC, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36) + punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E) + movdqa xmmG, xmmB + punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37) + punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + movdqa xmmH, xmmC + punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test edi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC + movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC + movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH +.out0: + add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub ecx, byte SIZEOF_XMMWORD + jz near .nextrow + + add esi, byte SIZEOF_XMMWORD ; inptr0 + add ebx, byte SIZEOF_XMMWORD ; inptr1 + add edx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st32: + cmp ecx, byte SIZEOF_XMMWORD/2 + jb short .column_st16 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + add edi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmC + movdqa xmmD, xmmH + sub ecx, byte SIZEOF_XMMWORD/2 +.column_st16: + cmp ecx, byte SIZEOF_XMMWORD/4 + jb short .column_st15 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + add edi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub ecx, byte SIZEOF_XMMWORD/4 +.column_st15: + ; Store two pixels (8 bytes) of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_XMMWORD/8 + jb short .column_st7 + movq XMM_MMWORD [edi], xmmA + add edi, byte SIZEOF_XMMWORD/8*4 + sub ecx, byte SIZEOF_XMMWORD/8 + psrldq xmmA, SIZEOF_XMMWORD/8*4 +.column_st7: + ; Store one pixel (4 bytes) of xmmA to the output when it has enough + ; space. + test ecx, ecx + jz short .nextrow + movd XMM_DWORD [edi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + + ALIGNX 16, 7 + +.nextrow: + pop ecx + pop esi + pop ebx + pop edx + pop edi + pop eax + + add esi, byte SIZEOF_JSAMPROW + add ebx, byte SIZEOF_JSAMPROW + add edx, byte SIZEOF_JSAMPROW + add edi, byte SIZEOF_JSAMPROW ; output_buf + dec eax ; num_rows + jg near .rowloop + + sfence ; flush the write buffer + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-avx2.asm new file mode 100644 index 0000000..d3a30d6 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-avx2.asm @@ -0,0 +1,114 @@ +; +; jdcolor.asm - colorspace conversion (AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_ycc_rgb_convert_avx2) + +EXTN(jconst_ycc_rgb_convert_avx2): + +PW_F0402 times 16 dw F_0_402 +PW_MF0228 times 16 dw -F_0_228 +PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285 +PW_ONE times 16 dw 1 +PD_ONEHALF times 8 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgb_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgbx_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgr_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgrx_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxbgr_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxrgb_convert_avx2 +%include "jdcolext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-mmx.asm new file mode 100644 index 0000000..6e67e4b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-mmx.asm @@ -0,0 +1,113 @@ +; +; jdcolor.asm - colorspace conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_ycc_rgb_convert_mmx) + +EXTN(jconst_ycc_rgb_convert_mmx): + +PW_F0402 times 4 dw F_0_402 +PW_MF0228 times 4 dw -F_0_228 +PW_MF0344_F0285 times 2 dw -F_0_344, F_0_285 +PW_ONE times 4 dw 1 +PD_ONEHALF times 2 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jdcolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extrgb_convert_mmx +%include "jdcolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extrgbx_convert_mmx +%include "jdcolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extbgr_convert_mmx +%include "jdcolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extbgrx_convert_mmx +%include "jdcolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extxbgr_convert_mmx +%include "jdcolext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_mmx jsimd_ycc_extxrgb_convert_mmx +%include "jdcolext-mmx.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-sse2.asm new file mode 100644 index 0000000..79c9c68 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdcolor-sse2.asm @@ -0,0 +1,113 @@ +; +; jdcolor.asm - colorspace conversion (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_ycc_rgb_convert_sse2) + +EXTN(jconst_ycc_rgb_convert_sse2): + +PW_F0402 times 8 dw F_0_402 +PW_MF0228 times 8 dw -F_0_228 +PW_MF0344_F0285 times 4 dw -F_0_344, F_0_285 +PW_ONE times 8 dw 1 +PD_ONEHALF times 4 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgb_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgbx_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgr_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgrx_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxbgr_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxrgb_convert_sse2 +%include "jdcolext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-avx2.asm new file mode 100644 index 0000000..90493fd --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-avx2.asm @@ -0,0 +1,132 @@ +; +; jdmerge.asm - merged upsampling/color conversion (AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_merged_upsample_avx2) + +EXTN(jconst_merged_upsample_avx2): + +PW_F0402 times 16 dw F_0_402 +PW_MF0228 times 16 dw -F_0_228 +PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285 +PW_ONE times 16 dw 1 +PD_ONEHALF times 8 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extrgb_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extrgb_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extrgbx_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extrgbx_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extbgr_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extbgr_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extbgrx_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extbgrx_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extxbgr_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extxbgr_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extxrgb_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extxrgb_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-mmx.asm new file mode 100644 index 0000000..0dc204a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-mmx.asm @@ -0,0 +1,119 @@ +; +; jdmerge.asm - merged upsampling/color conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_merged_upsample_mmx) + +EXTN(jconst_merged_upsample_mmx): + +PW_F0402 times 4 dw F_0_402 +PW_MF0228 times 4 dw -F_0_228 +PW_MF0344_F0285 times 2 dw -F_0_344, F_0_285 +PW_ONE times 4 dw 1 +PD_ONEHALF times 2 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jdmrgext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extrgb_merged_upsample_mmx +%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extrgb_merged_upsample_mmx +%include "jdmrgext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extrgbx_merged_upsample_mmx +%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extrgbx_merged_upsample_mmx +%include "jdmrgext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extbgr_merged_upsample_mmx +%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extbgr_merged_upsample_mmx +%include "jdmrgext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extbgrx_merged_upsample_mmx +%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extbgrx_merged_upsample_mmx +%include "jdmrgext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extxbgr_merged_upsample_mmx +%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extxbgr_merged_upsample_mmx +%include "jdmrgext-mmx.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_mmx jsimd_h2v1_extxrgb_merged_upsample_mmx +%define jsimd_h2v2_merged_upsample_mmx jsimd_h2v2_extxrgb_merged_upsample_mmx +%include "jdmrgext-mmx.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-sse2.asm new file mode 100644 index 0000000..06f0762 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmerge-sse2.asm @@ -0,0 +1,131 @@ +; +; jdmerge.asm - merged upsampling/color conversion (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_merged_upsample_sse2) + +EXTN(jconst_merged_upsample_sse2): + +PW_F0402 times 8 dw F_0_402 +PW_MF0228 times 8 dw -F_0_228 +PW_MF0344_F0285 times 4 dw -F_0_344, F_0_285 +PW_ONE times 8 dw 1 +PD_ONEHALF times 4 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 + +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extrgb_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extrgb_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extrgbx_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extrgbx_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extbgr_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extbgr_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extbgrx_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extbgrx_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extxbgr_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extxbgr_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extxrgb_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extxrgb_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-avx2.asm new file mode 100644 index 0000000..a7aa930 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-avx2.asm @@ -0,0 +1,571 @@ +; +; jdmrgext.asm - merged upsampling/color conversion (AVX2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2012, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v1_merged_upsample_avx2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +%define output_width(b) (b) + 8 ; JDIMENSION output_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_YMMWORD + ; ymmword wk[WK_NUM] +%define WK_NUM 3 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_merged_upsample_avx2) + +EXTN(jsimd_h2v1_merged_upsample_avx2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [output_width(eax)] ; col + test ecx, ecx + jz near .return + + push ecx + + mov edi, JSAMPIMAGE [input_buf(eax)] + mov ecx, JDIMENSION [in_row_group_ctr(eax)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + mov edi, JSAMPARRAY [output_buf(eax)] + mov esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW] ; inptr0 + mov ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW] ; inptr1 + mov edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW] ; inptr2 + mov edi, JSAMPROW [edi] ; outptr + + pop ecx ; col + + ALIGNX 16, 7 +.columnloop: + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + vmovdqu ymm6, YMMWORD [ebx] ; ymm6=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV) + vmovdqu ymm7, YMMWORD [edx] ; ymm7=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's) + vpcmpeqw ymm3, ymm3, ymm3 + vpsllw ymm3, ymm3, 7 ; ymm3={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + vpermq ymm6, ymm6, 0xd8 ; ymm6=Cb(01234567GHIJKLMN89ABCDEFOPQRSTUV) + vpermq ymm7, ymm7, 0xd8 ; ymm7=Cr(01234567GHIJKLMN89ABCDEFOPQRSTUV) + vpunpcklbw ymm4, ymm6, ymm1 ; ymm4=Cb(0123456789ABCDEF)=CbL + vpunpckhbw ymm6, ymm6, ymm1 ; ymm6=Cb(GHIJKLMNOPQRSTUV)=CbH + vpunpcklbw ymm0, ymm7, ymm1 ; ymm0=Cr(0123456789ABCDEF)=CrL + vpunpckhbw ymm7, ymm7, ymm1 ; ymm7=Cr(GHIJKLMNOPQRSTUV)=CrH + + vpaddw ymm5, ymm6, ymm3 + vpaddw ymm2, ymm4, ymm3 + vpaddw ymm1, ymm7, ymm3 + vpaddw ymm3, ymm0, ymm3 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + vpaddw ymm6, ymm5, ymm5 ; ymm6=2*CbH + vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbL + vpaddw ymm7, ymm1, ymm1 ; ymm7=2*CrH + vpaddw ymm0, ymm3, ymm3 ; ymm0=2*CrL + + vpmulhw ymm6, ymm6, [GOTOFF(eax,PW_MF0228)] ; ymm6=(2*CbH * -FIX(0.22800)) + vpmulhw ymm4, ymm4, [GOTOFF(eax,PW_MF0228)] ; ymm4=(2*CbL * -FIX(0.22800)) + vpmulhw ymm7, ymm7, [GOTOFF(eax,PW_F0402)] ; ymm7=(2*CrH * FIX(0.40200)) + vpmulhw ymm0, ymm0, [GOTOFF(eax,PW_F0402)] ; ymm0=(2*CrL * FIX(0.40200)) + + vpaddw ymm6, ymm6, [GOTOFF(eax,PW_ONE)] + vpaddw ymm4, ymm4, [GOTOFF(eax,PW_ONE)] + vpsraw ymm6, ymm6, 1 ; ymm6=(CbH * -FIX(0.22800)) + vpsraw ymm4, ymm4, 1 ; ymm4=(CbL * -FIX(0.22800)) + vpaddw ymm7, ymm7, [GOTOFF(eax,PW_ONE)] + vpaddw ymm0, ymm0, [GOTOFF(eax,PW_ONE)] + vpsraw ymm7, ymm7, 1 ; ymm7=(CrH * FIX(0.40200)) + vpsraw ymm0, ymm0, 1 ; ymm0=(CrL * FIX(0.40200)) + + vpaddw ymm6, ymm6, ymm5 + vpaddw ymm4, ymm4, ymm2 + vpaddw ymm6, ymm6, ymm5 ; ymm6=(CbH * FIX(1.77200))=(B-Y)H + vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbL * FIX(1.77200))=(B-Y)L + vpaddw ymm7, ymm7, ymm1 ; ymm7=(CrH * FIX(1.40200))=(R-Y)H + vpaddw ymm0, ymm0, ymm3 ; ymm0=(CrL * FIX(1.40200))=(R-Y)L + + vmovdqa YMMWORD [wk(0)], ymm6 ; wk(0)=(B-Y)H + vmovdqa YMMWORD [wk(1)], ymm7 ; wk(1)=(R-Y)H + + vpunpckhwd ymm6, ymm5, ymm1 + vpunpcklwd ymm5, ymm5, ymm1 + vpmaddwd ymm5, ymm5, [GOTOFF(eax,PW_MF0344_F0285)] + vpmaddwd ymm6, ymm6, [GOTOFF(eax,PW_MF0344_F0285)] + vpunpckhwd ymm7, ymm2, ymm3 + vpunpcklwd ymm2, ymm2, ymm3 + vpmaddwd ymm2, ymm2, [GOTOFF(eax,PW_MF0344_F0285)] + vpmaddwd ymm7, ymm7, [GOTOFF(eax,PW_MF0344_F0285)] + + vpaddd ymm5, ymm5, [GOTOFF(eax,PD_ONEHALF)] + vpaddd ymm6, ymm6, [GOTOFF(eax,PD_ONEHALF)] + vpsrad ymm5, ymm5, SCALEBITS + vpsrad ymm6, ymm6, SCALEBITS + vpaddd ymm2, ymm2, [GOTOFF(eax,PD_ONEHALF)] + vpaddd ymm7, ymm7, [GOTOFF(eax,PD_ONEHALF)] + vpsrad ymm2, ymm2, SCALEBITS + vpsrad ymm7, ymm7, SCALEBITS + + vpackssdw ymm5, ymm5, ymm6 ; ymm5=CbH*-FIX(0.344)+CrH*FIX(0.285) + vpackssdw ymm2, ymm2, ymm7 ; ymm2=CbL*-FIX(0.344)+CrL*FIX(0.285) + vpsubw ymm5, ymm5, ymm1 ; ymm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H + vpsubw ymm2, ymm2, ymm3 ; ymm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L + + vmovdqa YMMWORD [wk(2)], ymm5 ; wk(2)=(G-Y)H + + mov al, 2 ; Yctr + jmp short .Yloop_1st + ALIGNX 16, 7 + +.Yloop_2nd: + vmovdqa ymm0, YMMWORD [wk(1)] ; ymm0=(R-Y)H + vmovdqa ymm2, YMMWORD [wk(2)] ; ymm2=(G-Y)H + vmovdqa ymm4, YMMWORD [wk(0)] ; ymm4=(B-Y)H + ALIGNX 16, 7 + +.Yloop_1st: + vmovdqu ymm7, YMMWORD [esi] ; ymm7=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpcmpeqw ymm6, ymm6, ymm6 + vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..} + vpand ymm6, ymm6, ymm7 ; ymm6=Y(02468ACEGIKMOQSU)=YE + vpsrlw ymm7, ymm7, BYTE_BIT ; ymm7=Y(13579BDFHJLNPRTV)=YO + + vmovdqa ymm1, ymm0 ; ymm1=ymm0=(R-Y)(L/H) + vmovdqa ymm3, ymm2 ; ymm3=ymm2=(G-Y)(L/H) + vmovdqa ymm5, ymm4 ; ymm5=ymm4=(B-Y)(L/H) + + vpaddw ymm0, ymm0, ymm6 ; ymm0=((R-Y)+YE)=RE=R(02468ACEGIKMOQSU) + vpaddw ymm1, ymm1, ymm7 ; ymm1=((R-Y)+YO)=RO=R(13579BDFHJLNPRTV) + vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********) + vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********) + + vpaddw ymm2, ymm2, ymm6 ; ymm2=((G-Y)+YE)=GE=G(02468ACEGIKMOQSU) + vpaddw ymm3, ymm3, ymm7 ; ymm3=((G-Y)+YO)=GO=G(13579BDFHJLNPRTV) + vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********) + vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********) + + vpaddw ymm4, ymm4, ymm6 ; ymm4=((B-Y)+YE)=BE=B(02468ACEGIKMOQSU) + vpaddw ymm5, ymm5, ymm7 ; ymm5=((B-Y)+YO)=BO=B(13579BDFHJLNPRTV) + vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********) + vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F + ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V) + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F + ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V) + + vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G + ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --) + vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F + ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07 + ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N) + + vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H + ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --) + + vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H + ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --) + vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G + ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --) + vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18 + ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O) + + vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H + ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --) + vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29 + ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P) + + vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03 + ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J) + vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14 + ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K) + vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29 + ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P) + vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07 + ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N) + + vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B + ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R) + vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C + ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S) + vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H + ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --) + vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V) + + vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + cmp ecx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test edi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmF +.out0: + add edi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub ecx, byte SIZEOF_YMMWORD + jz near .endcolumn + + add esi, byte SIZEOF_YMMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add ebx, byte SIZEOF_YMMWORD ; inptr1 + add edx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st64: + lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE + cmp ecx, byte 2*SIZEOF_YMMWORD + jb short .column_st32 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + add edi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmF + sub ecx, byte 2*SIZEOF_YMMWORD + jmp short .column_st31 +.column_st32: + cmp ecx, byte SIZEOF_YMMWORD + jb short .column_st31 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + add edi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub ecx, byte SIZEOF_YMMWORD + jmp short .column_st31 +.column_st31: + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st15 + vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + add edi, byte SIZEOF_XMMWORD ; outptr + vperm2i128 ymmA, ymmA, ymmA, 1 + sub ecx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st7 + vmovq XMM_MMWORD [edi], xmmA + add edi, byte SIZEOF_MMWORD + sub ecx, byte SIZEOF_MMWORD + vpsrldq xmmA, xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_DWORD + jb short .column_st3 + vmovd XMM_DWORD [edi], xmmA + add edi, byte SIZEOF_DWORD + sub ecx, byte SIZEOF_DWORD + vpsrldq xmmA, xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of eax to the output when it has enough + ; space. + vmovd eax, xmmA + cmp ecx, byte SIZEOF_WORD + jb short .column_st1 + mov word [edi], ax + add edi, byte SIZEOF_WORD + sub ecx, byte SIZEOF_WORD + shr eax, 16 +.column_st1: + ; Store the lower 1 byte of eax to the output when it has enough + ; space. + test ecx, ecx + jz short .endcolumn + mov byte [edi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%else + vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%endif + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **) + ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E + ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U) + vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F + ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V) + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F + ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V) + + vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E + ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36 + ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M) + vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F + ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V) + vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37 + ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N) + + vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + + vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + cmp ecx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test edi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC + vmovntdq YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymmC + vmovdqu YMMWORD [edi+3*SIZEOF_YMMWORD], ymmH +.out0: + add edi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub ecx, byte SIZEOF_YMMWORD + jz near .endcolumn + + add esi, byte SIZEOF_YMMWORD ; inptr0 + dec al + jnz near .Yloop_2nd + + add ebx, byte SIZEOF_YMMWORD ; inptr1 + add edx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st64: + cmp ecx, byte SIZEOF_YMMWORD/2 + jb short .column_st32 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymmD + add edi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmC + vmovdqa ymmD, ymmH + sub ecx, byte SIZEOF_YMMWORD/2 +.column_st32: + cmp ecx, byte SIZEOF_YMMWORD/4 + jb short .column_st16 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymmA + add edi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub ecx, byte SIZEOF_YMMWORD/4 +.column_st16: + cmp ecx, byte SIZEOF_YMMWORD/8 + jb short .column_st15 + vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + add edi, byte SIZEOF_XMMWORD ; outptr + vperm2i128 ymmA, ymmA, ymmA, 1 + sub ecx, byte SIZEOF_YMMWORD/8 +.column_st15: + ; Store two pixels (8 bytes) of ymmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_YMMWORD/16 + jb short .column_st7 + vmovq MMWORD [edi], xmmA + add edi, byte SIZEOF_YMMWORD/16*4 + sub ecx, byte SIZEOF_YMMWORD/16 + vpsrldq xmmA, SIZEOF_YMMWORD/16*4 +.column_st7: + ; Store one pixel (4 bytes) of ymmA to the output when it has enough + ; space. + test ecx, ecx + jz short .endcolumn + vmovd XMM_DWORD [edi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + +.endcolumn: + sfence ; flush the write buffer + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v2_merged_upsample_avx2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +%define output_width(b) (b) + 8 ; JDIMENSION output_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_merged_upsample_avx2) + +EXTN(jsimd_h2v2_merged_upsample_avx2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov eax, POINTER [output_width(ebp)] + + mov edi, JSAMPIMAGE [input_buf(ebp)] + mov ecx, JDIMENSION [in_row_group_ctr(ebp)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + mov edi, JSAMPARRAY [output_buf(ebp)] + lea esi, [esi+ecx*SIZEOF_JSAMPROW] + + push edx ; inptr2 + push ebx ; inptr1 + push esi ; inptr00 + mov ebx, esp + + push edi ; output_buf (outptr0) + push ecx ; in_row_group_ctr + push ebx ; input_buf + push eax ; output_width + + call near EXTN(jsimd_h2v1_merged_upsample_avx2) + + add esi, byte SIZEOF_JSAMPROW ; inptr01 + add edi, byte SIZEOF_JSAMPROW ; outptr1 + mov POINTER [ebx+0*SIZEOF_POINTER], esi + mov POINTER [ebx-1*SIZEOF_POINTER], edi + + call near EXTN(jsimd_h2v1_merged_upsample_avx2) + + add esp, byte 7*SIZEOF_DWORD + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-mmx.asm new file mode 100644 index 0000000..5627581 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-mmx.asm @@ -0,0 +1,456 @@ +; +; jdmrgext.asm - merged upsampling/color conversion (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v1_merged_upsample_mmx(JDIMENSION output_width, JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +%define output_width(b) (b) + 8 ; JDIMENSION output_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD ; mmword wk[WK_NUM] +%define WK_NUM 3 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_merged_upsample_mmx) + +EXTN(jsimd_h2v1_merged_upsample_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [output_width(eax)] ; col + test ecx, ecx + jz near .return + + push ecx + + mov edi, JSAMPIMAGE [input_buf(eax)] + mov ecx, JDIMENSION [in_row_group_ctr(eax)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + mov edi, JSAMPARRAY [output_buf(eax)] + mov esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW] ; inptr0 + mov ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW] ; inptr1 + mov edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW] ; inptr2 + mov edi, JSAMPROW [edi] ; outptr + + pop ecx ; col + + ALIGNX 16, 7 +.columnloop: + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + movq mm6, MMWORD [ebx] ; mm6=Cb(01234567) + movq mm7, MMWORD [edx] ; mm7=Cr(01234567) + + pxor mm1, mm1 ; mm1=(all 0's) + pcmpeqw mm3, mm3 + psllw mm3, 7 ; mm3={0xFF80 0xFF80 0xFF80 0xFF80} + + movq mm4, mm6 + punpckhbw mm6, mm1 ; mm6=Cb(4567)=CbH + punpcklbw mm4, mm1 ; mm4=Cb(0123)=CbL + movq mm0, mm7 + punpckhbw mm7, mm1 ; mm7=Cr(4567)=CrH + punpcklbw mm0, mm1 ; mm0=Cr(0123)=CrL + + paddw mm6, mm3 + paddw mm4, mm3 + paddw mm7, mm3 + paddw mm0, mm3 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + movq mm5, mm6 ; mm5=CbH + movq mm2, mm4 ; mm2=CbL + paddw mm6, mm6 ; mm6=2*CbH + paddw mm4, mm4 ; mm4=2*CbL + movq mm1, mm7 ; mm1=CrH + movq mm3, mm0 ; mm3=CrL + paddw mm7, mm7 ; mm7=2*CrH + paddw mm0, mm0 ; mm0=2*CrL + + pmulhw mm6, [GOTOFF(eax,PW_MF0228)] ; mm6=(2*CbH * -FIX(0.22800)) + pmulhw mm4, [GOTOFF(eax,PW_MF0228)] ; mm4=(2*CbL * -FIX(0.22800)) + pmulhw mm7, [GOTOFF(eax,PW_F0402)] ; mm7=(2*CrH * FIX(0.40200)) + pmulhw mm0, [GOTOFF(eax,PW_F0402)] ; mm0=(2*CrL * FIX(0.40200)) + + paddw mm6, [GOTOFF(eax,PW_ONE)] + paddw mm4, [GOTOFF(eax,PW_ONE)] + psraw mm6, 1 ; mm6=(CbH * -FIX(0.22800)) + psraw mm4, 1 ; mm4=(CbL * -FIX(0.22800)) + paddw mm7, [GOTOFF(eax,PW_ONE)] + paddw mm0, [GOTOFF(eax,PW_ONE)] + psraw mm7, 1 ; mm7=(CrH * FIX(0.40200)) + psraw mm0, 1 ; mm0=(CrL * FIX(0.40200)) + + paddw mm6, mm5 + paddw mm4, mm2 + paddw mm6, mm5 ; mm6=(CbH * FIX(1.77200))=(B-Y)H + paddw mm4, mm2 ; mm4=(CbL * FIX(1.77200))=(B-Y)L + paddw mm7, mm1 ; mm7=(CrH * FIX(1.40200))=(R-Y)H + paddw mm0, mm3 ; mm0=(CrL * FIX(1.40200))=(R-Y)L + + movq MMWORD [wk(0)], mm6 ; wk(0)=(B-Y)H + movq MMWORD [wk(1)], mm7 ; wk(1)=(R-Y)H + + movq mm6, mm5 + movq mm7, mm2 + punpcklwd mm5, mm1 + punpckhwd mm6, mm1 + pmaddwd mm5, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd mm6, [GOTOFF(eax,PW_MF0344_F0285)] + punpcklwd mm2, mm3 + punpckhwd mm7, mm3 + pmaddwd mm2, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd mm7, [GOTOFF(eax,PW_MF0344_F0285)] + + paddd mm5, [GOTOFF(eax,PD_ONEHALF)] + paddd mm6, [GOTOFF(eax,PD_ONEHALF)] + psrad mm5, SCALEBITS + psrad mm6, SCALEBITS + paddd mm2, [GOTOFF(eax,PD_ONEHALF)] + paddd mm7, [GOTOFF(eax,PD_ONEHALF)] + psrad mm2, SCALEBITS + psrad mm7, SCALEBITS + + packssdw mm5, mm6 ; mm5=CbH*-FIX(0.344)+CrH*FIX(0.285) + packssdw mm2, mm7 ; mm2=CbL*-FIX(0.344)+CrL*FIX(0.285) + psubw mm5, mm1 ; mm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H + psubw mm2, mm3 ; mm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L + + movq MMWORD [wk(2)], mm5 ; wk(2)=(G-Y)H + + mov al, 2 ; Yctr + jmp short .Yloop_1st + ALIGNX 16, 7 + +.Yloop_2nd: + movq mm0, MMWORD [wk(1)] ; mm0=(R-Y)H + movq mm2, MMWORD [wk(2)] ; mm2=(G-Y)H + movq mm4, MMWORD [wk(0)] ; mm4=(B-Y)H + ALIGNX 16, 7 + +.Yloop_1st: + movq mm7, MMWORD [esi] ; mm7=Y(01234567) + + pcmpeqw mm6, mm6 + psrlw mm6, BYTE_BIT ; mm6={0xFF 0x00 0xFF 0x00 ..} + pand mm6, mm7 ; mm6=Y(0246)=YE + psrlw mm7, BYTE_BIT ; mm7=Y(1357)=YO + + movq mm1, mm0 ; mm1=mm0=(R-Y)(L/H) + movq mm3, mm2 ; mm3=mm2=(G-Y)(L/H) + movq mm5, mm4 ; mm5=mm4=(B-Y)(L/H) + + paddw mm0, mm6 ; mm0=((R-Y)+YE)=RE=(R0 R2 R4 R6) + paddw mm1, mm7 ; mm1=((R-Y)+YO)=RO=(R1 R3 R5 R7) + packuswb mm0, mm0 ; mm0=(R0 R2 R4 R6 ** ** ** **) + packuswb mm1, mm1 ; mm1=(R1 R3 R5 R7 ** ** ** **) + + paddw mm2, mm6 ; mm2=((G-Y)+YE)=GE=(G0 G2 G4 G6) + paddw mm3, mm7 ; mm3=((G-Y)+YO)=GO=(G1 G3 G5 G7) + packuswb mm2, mm2 ; mm2=(G0 G2 G4 G6 ** ** ** **) + packuswb mm3, mm3 ; mm3=(G1 G3 G5 G7 ** ** ** **) + + paddw mm4, mm6 ; mm4=((B-Y)+YE)=BE=(B0 B2 B4 B6) + paddw mm5, mm7 ; mm5=((B-Y)+YO)=BO=(B1 B3 B5 B7) + packuswb mm4, mm4 ; mm4=(B0 B2 B4 B6 ** ** ** **) + packuswb mm5, mm5 ; mm5=(B1 B3 B5 B7 ** ** ** **) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) + ; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) + ; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **) + ; mmG=(** ** ** ** ** ** ** **), mmH=(** ** ** ** ** ** ** **) + + punpcklbw mmA, mmC ; mmA=(00 10 02 12 04 14 06 16) + punpcklbw mmE, mmB ; mmE=(20 01 22 03 24 05 26 07) + punpcklbw mmD, mmF ; mmD=(11 21 13 23 15 25 17 27) + + movq mmG, mmA + movq mmH, mmA + punpcklwd mmA, mmE ; mmA=(00 10 20 01 02 12 22 03) + punpckhwd mmG, mmE ; mmG=(04 14 24 05 06 16 26 07) + + psrlq mmH, 2*BYTE_BIT ; mmH=(02 12 04 14 06 16 -- --) + psrlq mmE, 2*BYTE_BIT ; mmE=(22 03 24 05 26 07 -- --) + + movq mmC, mmD + movq mmB, mmD + punpcklwd mmD, mmH ; mmD=(11 21 02 12 13 23 04 14) + punpckhwd mmC, mmH ; mmC=(15 25 06 16 17 27 -- --) + + psrlq mmB, 2*BYTE_BIT ; mmB=(13 23 15 25 17 27 -- --) + + movq mmF, mmE + punpcklwd mmE, mmB ; mmE=(22 03 13 23 24 05 15 25) + punpckhwd mmF, mmB ; mmF=(26 07 17 27 -- -- -- --) + + punpckldq mmA, mmD ; mmA=(00 10 20 01 11 21 02 12) + punpckldq mmE, mmG ; mmE=(22 03 13 23 04 14 24 05) + punpckldq mmC, mmF ; mmC=(15 25 06 16 26 07 17 27) + + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st16 + + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmE + movq MMWORD [edi+2*SIZEOF_MMWORD], mmC + + sub ecx, byte SIZEOF_MMWORD + jz near .endcolumn + + add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr + add esi, byte SIZEOF_MMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add ebx, byte SIZEOF_MMWORD ; inptr1 + add edx, byte SIZEOF_MMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st16: + lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE + cmp ecx, byte 2*SIZEOF_MMWORD + jb short .column_st8 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmE + movq mmA, mmC + sub ecx, byte 2*SIZEOF_MMWORD + add edi, byte 2*SIZEOF_MMWORD + jmp short .column_st4 +.column_st8: + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st4 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq mmA, mmE + sub ecx, byte SIZEOF_MMWORD + add edi, byte SIZEOF_MMWORD +.column_st4: + movd eax, mmA + cmp ecx, byte SIZEOF_DWORD + jb short .column_st2 + mov dword [edi+0*SIZEOF_DWORD], eax + psrlq mmA, DWORD_BIT + movd eax, mmA + sub ecx, byte SIZEOF_DWORD + add edi, byte SIZEOF_DWORD +.column_st2: + cmp ecx, byte SIZEOF_WORD + jb short .column_st1 + mov word [edi+0*SIZEOF_WORD], ax + shr eax, WORD_BIT + sub ecx, byte SIZEOF_WORD + add edi, byte SIZEOF_WORD +.column_st1: + cmp ecx, byte SIZEOF_BYTE + jb short .endcolumn + mov byte [edi+0*SIZEOF_BYTE], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + pcmpeqb mm6, mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **) + pcmpeqb mm7, mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **) +%else + pxor mm6, mm6 ; mm6=(X0 X2 X4 X6 ** ** ** **) + pxor mm7, mm7 ; mm7=(X1 X3 X5 X7 ** ** ** **) +%endif + ; mmA=(00 02 04 06 ** ** ** **), mmB=(01 03 05 07 ** ** ** **) + ; mmC=(10 12 14 16 ** ** ** **), mmD=(11 13 15 17 ** ** ** **) + ; mmE=(20 22 24 26 ** ** ** **), mmF=(21 23 25 27 ** ** ** **) + ; mmG=(30 32 34 36 ** ** ** **), mmH=(31 33 35 37 ** ** ** **) + + punpcklbw mmA, mmC ; mmA=(00 10 02 12 04 14 06 16) + punpcklbw mmE, mmG ; mmE=(20 30 22 32 24 34 26 36) + punpcklbw mmB, mmD ; mmB=(01 11 03 13 05 15 07 17) + punpcklbw mmF, mmH ; mmF=(21 31 23 33 25 35 27 37) + + movq mmC, mmA + punpcklwd mmA, mmE ; mmA=(00 10 20 30 02 12 22 32) + punpckhwd mmC, mmE ; mmC=(04 14 24 34 06 16 26 36) + movq mmG, mmB + punpcklwd mmB, mmF ; mmB=(01 11 21 31 03 13 23 33) + punpckhwd mmG, mmF ; mmG=(05 15 25 35 07 17 27 37) + + movq mmD, mmA + punpckldq mmA, mmB ; mmA=(00 10 20 30 01 11 21 31) + punpckhdq mmD, mmB ; mmD=(02 12 22 32 03 13 23 33) + movq mmH, mmC + punpckldq mmC, mmG ; mmC=(04 14 24 34 05 15 25 35) + punpckhdq mmH, mmG ; mmH=(06 16 26 36 07 17 27 37) + + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st16 + + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmD + movq MMWORD [edi+2*SIZEOF_MMWORD], mmC + movq MMWORD [edi+3*SIZEOF_MMWORD], mmH + + sub ecx, byte SIZEOF_MMWORD + jz short .endcolumn + + add edi, byte RGB_PIXELSIZE*SIZEOF_MMWORD ; outptr + add esi, byte SIZEOF_MMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add ebx, byte SIZEOF_MMWORD ; inptr1 + add edx, byte SIZEOF_MMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st16: + cmp ecx, byte SIZEOF_MMWORD/2 + jb short .column_st8 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq MMWORD [edi+1*SIZEOF_MMWORD], mmD + movq mmA, mmC + movq mmD, mmH + sub ecx, byte SIZEOF_MMWORD/2 + add edi, byte 2*SIZEOF_MMWORD +.column_st8: + cmp ecx, byte SIZEOF_MMWORD/4 + jb short .column_st4 + movq MMWORD [edi+0*SIZEOF_MMWORD], mmA + movq mmA, mmD + sub ecx, byte SIZEOF_MMWORD/4 + add edi, byte 1*SIZEOF_MMWORD +.column_st4: + cmp ecx, byte SIZEOF_MMWORD/8 + jb short .endcolumn + movd dword [edi+0*SIZEOF_DWORD], mmA + +%endif ; RGB_PIXELSIZE ; --------------- + +.endcolumn: + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v2_merged_upsample_mmx(JDIMENSION output_width, JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +%define output_width(b) (b) + 8 ; JDIMENSION output_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_merged_upsample_mmx) + +EXTN(jsimd_h2v2_merged_upsample_mmx): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov eax, JDIMENSION [output_width(ebp)] + + mov edi, JSAMPIMAGE [input_buf(ebp)] + mov ecx, JDIMENSION [in_row_group_ctr(ebp)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + mov edi, JSAMPARRAY [output_buf(ebp)] + lea esi, [esi+ecx*SIZEOF_JSAMPROW] + + push edx ; inptr2 + push ebx ; inptr1 + push esi ; inptr00 + mov ebx, esp + + push edi ; output_buf (outptr0) + push ecx ; in_row_group_ctr + push ebx ; input_buf + push eax ; output_width + + call near EXTN(jsimd_h2v1_merged_upsample_mmx) + + add esi, byte SIZEOF_JSAMPROW ; inptr01 + add edi, byte SIZEOF_JSAMPROW ; outptr1 + mov POINTER [ebx+0*SIZEOF_POINTER], esi + mov POINTER [ebx-1*SIZEOF_POINTER], edi + + call near EXTN(jsimd_h2v1_merged_upsample_mmx) + + add esp, byte 7*SIZEOF_DWORD + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-sse2.asm new file mode 100644 index 0000000..13e7d98 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdmrgext-sse2.asm @@ -0,0 +1,513 @@ +; +; jdmrgext.asm - merged upsampling/color conversion (SSE2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2012, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v1_merged_upsample_sse2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +%define output_width(b) (b) + 8 ; JDIMENSION output_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 3 +%define gotptr wk(0) - SIZEOF_POINTER ; void * gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_merged_upsample_sse2) + +EXTN(jsimd_h2v1_merged_upsample_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov ecx, JDIMENSION [output_width(eax)] ; col + test ecx, ecx + jz near .return + + push ecx + + mov edi, JSAMPIMAGE [input_buf(eax)] + mov ecx, JDIMENSION [in_row_group_ctr(eax)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + mov edi, JSAMPARRAY [output_buf(eax)] + mov esi, JSAMPROW [esi+ecx*SIZEOF_JSAMPROW] ; inptr0 + mov ebx, JSAMPROW [ebx+ecx*SIZEOF_JSAMPROW] ; inptr1 + mov edx, JSAMPROW [edx+ecx*SIZEOF_JSAMPROW] ; inptr2 + mov edi, JSAMPROW [edi] ; outptr + + pop ecx ; col + + ALIGNX 16, 7 +.columnloop: + MOVPIC eax, POINTER [gotptr] ; load GOT address (eax) + + movdqa xmm6, XMMWORD [ebx] ; xmm6=Cb(0123456789ABCDEF) + movdqa xmm7, XMMWORD [edx] ; xmm7=Cr(0123456789ABCDEF) + + pxor xmm1, xmm1 ; xmm1=(all 0's) + pcmpeqw xmm3, xmm3 + psllw xmm3, 7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + movdqa xmm4, xmm6 + punpckhbw xmm6, xmm1 ; xmm6=Cb(89ABCDEF)=CbH + punpcklbw xmm4, xmm1 ; xmm4=Cb(01234567)=CbL + movdqa xmm0, xmm7 + punpckhbw xmm7, xmm1 ; xmm7=Cr(89ABCDEF)=CrH + punpcklbw xmm0, xmm1 ; xmm0=Cr(01234567)=CrL + + paddw xmm6, xmm3 + paddw xmm4, xmm3 + paddw xmm7, xmm3 + paddw xmm0, xmm3 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + movdqa xmm5, xmm6 ; xmm5=CbH + movdqa xmm2, xmm4 ; xmm2=CbL + paddw xmm6, xmm6 ; xmm6=2*CbH + paddw xmm4, xmm4 ; xmm4=2*CbL + movdqa xmm1, xmm7 ; xmm1=CrH + movdqa xmm3, xmm0 ; xmm3=CrL + paddw xmm7, xmm7 ; xmm7=2*CrH + paddw xmm0, xmm0 ; xmm0=2*CrL + + pmulhw xmm6, [GOTOFF(eax,PW_MF0228)] ; xmm6=(2*CbH * -FIX(0.22800)) + pmulhw xmm4, [GOTOFF(eax,PW_MF0228)] ; xmm4=(2*CbL * -FIX(0.22800)) + pmulhw xmm7, [GOTOFF(eax,PW_F0402)] ; xmm7=(2*CrH * FIX(0.40200)) + pmulhw xmm0, [GOTOFF(eax,PW_F0402)] ; xmm0=(2*CrL * FIX(0.40200)) + + paddw xmm6, [GOTOFF(eax,PW_ONE)] + paddw xmm4, [GOTOFF(eax,PW_ONE)] + psraw xmm6, 1 ; xmm6=(CbH * -FIX(0.22800)) + psraw xmm4, 1 ; xmm4=(CbL * -FIX(0.22800)) + paddw xmm7, [GOTOFF(eax,PW_ONE)] + paddw xmm0, [GOTOFF(eax,PW_ONE)] + psraw xmm7, 1 ; xmm7=(CrH * FIX(0.40200)) + psraw xmm0, 1 ; xmm0=(CrL * FIX(0.40200)) + + paddw xmm6, xmm5 + paddw xmm4, xmm2 + paddw xmm6, xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H + paddw xmm4, xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L + paddw xmm7, xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H + paddw xmm0, xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L + + movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H + movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H + + movdqa xmm6, xmm5 + movdqa xmm7, xmm2 + punpcklwd xmm5, xmm1 + punpckhwd xmm6, xmm1 + pmaddwd xmm5, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd xmm6, [GOTOFF(eax,PW_MF0344_F0285)] + punpcklwd xmm2, xmm3 + punpckhwd xmm7, xmm3 + pmaddwd xmm2, [GOTOFF(eax,PW_MF0344_F0285)] + pmaddwd xmm7, [GOTOFF(eax,PW_MF0344_F0285)] + + paddd xmm5, [GOTOFF(eax,PD_ONEHALF)] + paddd xmm6, [GOTOFF(eax,PD_ONEHALF)] + psrad xmm5, SCALEBITS + psrad xmm6, SCALEBITS + paddd xmm2, [GOTOFF(eax,PD_ONEHALF)] + paddd xmm7, [GOTOFF(eax,PD_ONEHALF)] + psrad xmm2, SCALEBITS + psrad xmm7, SCALEBITS + + packssdw xmm5, xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285) + packssdw xmm2, xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285) + psubw xmm5, xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H + psubw xmm2, xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L + + movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H + + mov al, 2 ; Yctr + jmp short .Yloop_1st + ALIGNX 16, 7 + +.Yloop_2nd: + movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H + movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H + movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H + ALIGNX 16, 7 + +.Yloop_1st: + movdqa xmm7, XMMWORD [esi] ; xmm7=Y(0123456789ABCDEF) + + pcmpeqw xmm6, xmm6 + psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..} + pand xmm6, xmm7 ; xmm6=Y(02468ACE)=YE + psrlw xmm7, BYTE_BIT ; xmm7=Y(13579BDF)=YO + + movdqa xmm1, xmm0 ; xmm1=xmm0=(R-Y)(L/H) + movdqa xmm3, xmm2 ; xmm3=xmm2=(G-Y)(L/H) + movdqa xmm5, xmm4 ; xmm5=xmm4=(B-Y)(L/H) + + paddw xmm0, xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE) + paddw xmm1, xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF) + packuswb xmm0, xmm0 ; xmm0=R(02468ACE********) + packuswb xmm1, xmm1 ; xmm1=R(13579BDF********) + + paddw xmm2, xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE) + paddw xmm3, xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF) + packuswb xmm2, xmm2 ; xmm2=G(02468ACE********) + packuswb xmm3, xmm3 ; xmm3=G(13579BDF********) + + paddw xmm4, xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE) + paddw xmm5, xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF) + packuswb xmm4, xmm4 ; xmm4=B(02468ACE********) + packuswb xmm5, xmm5 ; xmm5=B(13579BDF********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F) + punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F) + + movdqa xmmG, xmmA + movdqa xmmH, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07) + punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F) + + psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --) + psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --) + + movdqa xmmC, xmmD + movdqa xmmB, xmmD + punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18) + punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --) + + psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --) + + movdqa xmmF, xmmE + punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29) + punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --) + + pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03) + movdqa xmmB, xmmE + punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14) + punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07) + punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29) + + pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B) + movdqa xmmB, xmmF + punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C) + punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F) + punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --) + + punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test edi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmF +.out0: + add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub ecx, byte SIZEOF_XMMWORD + jz near .endcolumn + + add esi, byte SIZEOF_XMMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add ebx, byte SIZEOF_XMMWORD ; inptr1 + add edx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st32: + lea ecx, [ecx+ecx*2] ; imul ecx, RGB_PIXELSIZE + cmp ecx, byte 2*SIZEOF_XMMWORD + jb short .column_st16 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + add edi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmF + sub ecx, byte 2*SIZEOF_XMMWORD + jmp short .column_st15 +.column_st16: + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st15 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + add edi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub ecx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_MMWORD + jb short .column_st7 + movq XMM_MMWORD [edi], xmmA + add edi, byte SIZEOF_MMWORD + sub ecx, byte SIZEOF_MMWORD + psrldq xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_DWORD + jb short .column_st3 + movd XMM_DWORD [edi], xmmA + add edi, byte SIZEOF_DWORD + sub ecx, byte SIZEOF_DWORD + psrldq xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of eax to the output when it has enough + ; space. + movd eax, xmmA + cmp ecx, byte SIZEOF_WORD + jb short .column_st1 + mov word [edi], ax + add edi, byte SIZEOF_WORD + sub ecx, byte SIZEOF_WORD + shr eax, 16 +.column_st1: + ; Store the lower 1 byte of eax to the output when it has enough + ; space. + test ecx, ecx + jz short .endcolumn + mov byte [edi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%else + pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%endif + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E) + punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F) + punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F) + + movdqa xmmC, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36) + punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E) + movdqa xmmG, xmmB + punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37) + punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + movdqa xmmH, xmmC + punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + cmp ecx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test edi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC + movntdq XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [edi+2*SIZEOF_XMMWORD], xmmC + movdqu XMMWORD [edi+3*SIZEOF_XMMWORD], xmmH +.out0: + add edi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub ecx, byte SIZEOF_XMMWORD + jz near .endcolumn + + add esi, byte SIZEOF_XMMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add ebx, byte SIZEOF_XMMWORD ; inptr1 + add edx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + ALIGNX 16, 7 + +.column_st32: + cmp ecx, byte SIZEOF_XMMWORD/2 + jb short .column_st16 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmmD + add edi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmC + movdqa xmmD, xmmH + sub ecx, byte SIZEOF_XMMWORD/2 +.column_st16: + cmp ecx, byte SIZEOF_XMMWORD/4 + jb short .column_st15 + movdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmmA + add edi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub ecx, byte SIZEOF_XMMWORD/4 +.column_st15: + ; Store two pixels (8 bytes) of xmmA to the output when it has enough + ; space. + cmp ecx, byte SIZEOF_XMMWORD/8 + jb short .column_st7 + movq XMM_MMWORD [edi], xmmA + add edi, byte SIZEOF_XMMWORD/8*4 + sub ecx, byte SIZEOF_XMMWORD/8 + psrldq xmmA, SIZEOF_XMMWORD/8*4 +.column_st7: + ; Store one pixel (4 bytes) of xmmA to the output when it has enough + ; space. + test ecx, ecx + jz short .endcolumn + movd XMM_DWORD [edi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + +.endcolumn: + sfence ; flush the write buffer + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v2_merged_upsample_sse2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +%define output_width(b) (b) + 8 ; JDIMENSION output_width +%define input_buf(b) (b) + 12 ; JSAMPIMAGE input_buf +%define in_row_group_ctr(b) (b) + 16 ; JDIMENSION in_row_group_ctr +%define output_buf(b) (b) + 20 ; JSAMPARRAY output_buf + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_merged_upsample_sse2) + +EXTN(jsimd_h2v2_merged_upsample_sse2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov eax, POINTER [output_width(ebp)] + + mov edi, JSAMPIMAGE [input_buf(ebp)] + mov ecx, JDIMENSION [in_row_group_ctr(ebp)] + mov esi, JSAMPARRAY [edi+0*SIZEOF_JSAMPARRAY] + mov ebx, JSAMPARRAY [edi+1*SIZEOF_JSAMPARRAY] + mov edx, JSAMPARRAY [edi+2*SIZEOF_JSAMPARRAY] + mov edi, JSAMPARRAY [output_buf(ebp)] + lea esi, [esi+ecx*SIZEOF_JSAMPROW] + + push edx ; inptr2 + push ebx ; inptr1 + push esi ; inptr00 + mov ebx, esp + + push edi ; output_buf (outptr0) + push ecx ; in_row_group_ctr + push ebx ; input_buf + push eax ; output_width + + call near EXTN(jsimd_h2v1_merged_upsample_sse2) + + add esi, byte SIZEOF_JSAMPROW ; inptr01 + add edi, byte SIZEOF_JSAMPROW ; outptr1 + mov POINTER [ebx+0*SIZEOF_POINTER], esi + mov POINTER [ebx-1*SIZEOF_POINTER], edi + + call near EXTN(jsimd_h2v1_merged_upsample_sse2) + + add esp, byte 7*SIZEOF_DWORD + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-avx2.asm new file mode 100644 index 0000000..eba53ef --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-avx2.asm @@ -0,0 +1,756 @@ +; +; jdsample.asm - upsampling (AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fancy_upsample_avx2) + +EXTN(jconst_fancy_upsample_avx2): + +PW_ONE times 16 dw 1 +PW_TWO times 16 dw 2 +PW_THREE times 16 dw 3 +PW_SEVEN times 16 dw 7 +PW_EIGHT times 16 dw 8 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical. +; +; The upsampling algorithm is linear interpolation between pixel centers, +; also known as a "triangle filter". This is a good compromise between +; speed and visual quality. The centers of the output pixels are 1/4 and 3/4 +; of the way between input pixel centers. +; +; GLOBAL(void) +; jsimd_h2v1_fancy_upsample_avx2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define downsamp_width(b) (b) + 12 ; JDIMENSION downsampled_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_fancy_upsample_avx2) + +EXTN(jsimd_h2v1_fancy_upsample_avx2): + push ebp + mov ebp, esp + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + mov eax, JDIMENSION [downsamp_width(ebp)] ; colctr + test eax, eax + jz near .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push eax ; colctr + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + + test eax, SIZEOF_YMMWORD-1 + jz short .skip + mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample +.skip: + vpxor ymm0, ymm0, ymm0 ; ymm0=(all 0's) + vpcmpeqb xmm7, xmm7, xmm7 + vpsrldq xmm7, xmm7, (SIZEOF_XMMWORD-1) ; (ff -- -- -- ... -- --) LSB is ff + vpand ymm7, ymm7, YMMWORD [esi+0*SIZEOF_YMMWORD] + + add eax, byte SIZEOF_YMMWORD-1 + and eax, byte -SIZEOF_YMMWORD + cmp eax, byte SIZEOF_YMMWORD + ja short .columnloop + ALIGNX 16, 7 + +.columnloop_last: + vpcmpeqb xmm6, xmm6, xmm6 + vpslldq xmm6, xmm6, (SIZEOF_XMMWORD-1) + vperm2i128 ymm6, ymm6, ymm6, 1 ; (---- ---- ... ---- ---- ff) MSB is ff + vpand ymm6, ymm6, YMMWORD [esi+0*SIZEOF_YMMWORD] + jmp short .upsample + ALIGNX 16, 7 + +.columnloop: + vmovdqu ymm6, YMMWORD [esi+1*SIZEOF_YMMWORD] + vperm2i128 ymm6, ymm0, ymm6, 0x20 + vpslldq ymm6, ymm6, 15 + +.upsample: + vmovdqu ymm1, YMMWORD [esi+0*SIZEOF_YMMWORD] ; ymm1=( 0 1 2 ... 29 30 31) + + vperm2i128 ymm2, ymm0, ymm1, 0x20 + vpalignr ymm2, ymm1, ymm2, 15 ; ymm2=(-- 0 1 ... 28 29 30) + vperm2i128 ymm4, ymm0, ymm1, 0x03 + vpalignr ymm3, ymm4, ymm1, 1 ; ymm3=( 1 2 3 ... 30 31 --) + + vpor ymm2, ymm2, ymm7 ; ymm2=(-1 0 1 ... 28 29 30) + vpor ymm3, ymm3, ymm6 ; ymm3=( 1 2 3 ... 30 31 32) + + vpsrldq ymm7, ymm4, (SIZEOF_XMMWORD-1) ; ymm7=(31 -- -- ... -- -- --) + + vpunpckhbw ymm4, ymm1, ymm0 ; ymm4=( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm5, ymm1, ymm0 ; ymm5=( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm1, ymm5, ymm4, 0x20 ; ymm1=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm5, ymm2, ymm0 ; ymm5=( 7 8 9 10 11 12 13 14 23 24 25 26 27 28 29 30) + vpunpcklbw ymm6, ymm2, ymm0 ; ymm6=(-1 0 1 2 3 4 5 6 15 16 17 18 19 20 21 22) + vperm2i128 ymm2, ymm6, ymm5, 0x20 ; ymm2=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vpunpckhbw ymm6, ymm3, ymm0 ; ymm6=( 1 2 3 4 5 6 7 8 17 18 19 20 21 22 23 24) + vpunpcklbw ymm0, ymm3, ymm0 ; ymm0=( 9 10 11 12 13 14 15 16 25 26 27 28 29 30 31 32) + vperm2i128 ymm3, ymm0, ymm6, 0x20 ; ymm3=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16) + vperm2i128 ymm6, ymm0, ymm6, 0x31 ; ymm6=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32) + + vpxor ymm0, ymm0, ymm0 ; ymm0=(all 0's) + + vpmullw ymm1, ymm1, [GOTOFF(ebx,PW_THREE)] + vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)] + vpaddw ymm2, ymm2, [GOTOFF(ebx,PW_ONE)] + vpaddw ymm5, ymm5, [GOTOFF(ebx,PW_ONE)] + vpaddw ymm3, ymm3, [GOTOFF(ebx,PW_TWO)] + vpaddw ymm6, ymm6, [GOTOFF(ebx,PW_TWO)] + + vpaddw ymm2, ymm2, ymm1 + vpaddw ymm5, ymm5, ymm4 + vpsrlw ymm2, ymm2, 2 ; ymm2=OutLE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30) + vpsrlw ymm5, ymm5, 2 ; ymm5=OutHE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62) + vpaddw ymm3, ymm3, ymm1 + vpaddw ymm6, ymm6, ymm4 + vpsrlw ymm3, ymm3, 2 ; ymm3=OutLO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31) + vpsrlw ymm6, ymm6, 2 ; ymm6=OutHO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63) + + vpsllw ymm3, ymm3, BYTE_BIT + vpsllw ymm6, ymm6, BYTE_BIT + vpor ymm2, ymm2, ymm3 ; ymm2=OutL=( 0 1 2 ... 29 30 31) + vpor ymm5, ymm5, ymm6 ; ymm5=OutH=(32 33 34 ... 61 62 63) + + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm2 + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm5 + + sub eax, byte SIZEOF_YMMWORD + add esi, byte 1*SIZEOF_YMMWORD ; inptr + add edi, byte 2*SIZEOF_YMMWORD ; outptr + cmp eax, byte SIZEOF_YMMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop esi + pop edi + pop eax + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec ecx ; rowctr + jg near .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical. +; Again a triangle filter; see comments for h2v1 case, above. +; +; GLOBAL(void) +; jsimd_h2v2_fancy_upsample_avx2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define downsamp_width(b) (b) + 12 ; JDIMENSION downsampled_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_YMMWORD + ; ymmword wk[WK_NUM] +%define WK_NUM 4 +%define gotptr wk(0) - SIZEOF_POINTER ; void *gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_fancy_upsample_avx2) + +EXTN(jsimd_h2v2_fancy_upsample_avx2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov edx, eax ; edx = original ebp + mov eax, JDIMENSION [downsamp_width(edx)] ; colctr + test eax, eax + jz near .return + + mov ecx, INT [max_v_samp(edx)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(edx)] ; input_data + mov edi, POINTER [output_data_ptr(edx)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push eax ; colctr + push ecx + push edi + push esi + + mov ecx, JSAMPROW [esi-1*SIZEOF_JSAMPROW] ; inptr1(above) + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0 + mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1(below) + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0 + mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1 + + test eax, SIZEOF_YMMWORD-1 + jz short .skip + push edx + mov dl, JSAMPLE [ecx+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [ecx+eax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [ebx+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [ebx+eax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample + pop edx +.skip: + ; -- process the first column block + + vmovdqu ymm0, YMMWORD [ebx+0*SIZEOF_YMMWORD] ; ymm0=row[ 0][0] + vmovdqu ymm1, YMMWORD [ecx+0*SIZEOF_YMMWORD] ; ymm1=row[-1][0] + vmovdqu ymm2, YMMWORD [esi+0*SIZEOF_YMMWORD] ; ymm2=row[+1][0] + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + vpxor ymm3, ymm3, ymm3 ; ymm3=(all 0's) + + vpunpckhbw ymm4, ymm0, ymm3 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm5, ymm0, ymm3 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm5, ymm1, ymm3 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm6, ymm1, ymm3 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm6, ymm2, ymm3 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm3, ymm2, ymm3 ; ymm3=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm2, ymm3, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm6, ymm3, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpmullw ymm0, ymm0, [GOTOFF(ebx,PW_THREE)] + vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)] + + vpcmpeqb xmm7, xmm7, xmm7 + vpsrldq xmm7, xmm7, (SIZEOF_XMMWORD-2) ; (ffff ---- ---- ... ---- ----) LSB is ffff + + vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vmovdqu YMMWORD [edx+0*SIZEOF_YMMWORD], ymm1 ; temporarily save + vmovdqu YMMWORD [edx+1*SIZEOF_YMMWORD], ymm5 ; the intermediate data + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm2 + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm6 + + vpand ymm1, ymm1, ymm7 ; ymm1=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vpand ymm2, ymm2, ymm7 ; ymm2=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + + vmovdqa YMMWORD [wk(0)], ymm1 + vmovdqa YMMWORD [wk(1)], ymm2 + + POPPIC ebx + + add eax, byte SIZEOF_YMMWORD-1 + and eax, byte -SIZEOF_YMMWORD + cmp eax, byte SIZEOF_YMMWORD + ja short .columnloop + ALIGNX 16, 7 + +.columnloop_last: + ; -- process the last column block + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + vpcmpeqb xmm1, xmm1, xmm1 + vpslldq xmm1, xmm1, (SIZEOF_XMMWORD-2) + vperm2i128 ymm1, ymm1, ymm1, 1 ; (---- ---- ... ---- ---- ffff) MSB is ffff + + vpand ymm2, ymm1, YMMWORD [edi+1*SIZEOF_YMMWORD] + vpand ymm1, ymm1, YMMWORD [edx+1*SIZEOF_YMMWORD] + + vmovdqa YMMWORD [wk(2)], ymm1 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31) + vmovdqa YMMWORD [wk(3)], ymm2 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31) + + jmp near .upsample + ALIGNX 16, 7 + +.columnloop: + ; -- process the next column block + + vmovdqu ymm0, YMMWORD [ebx+1*SIZEOF_YMMWORD] ; ymm0=row[ 0][1] + vmovdqu ymm1, YMMWORD [ecx+1*SIZEOF_YMMWORD] ; ymm1=row[-1][1] + vmovdqu ymm2, YMMWORD [esi+1*SIZEOF_YMMWORD] ; ymm2=row[+1][1] + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + vpxor ymm3, ymm3, ymm3 ; ymm3=(all 0's) + + vpunpckhbw ymm4, ymm0, ymm3 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm5, ymm0, ymm3 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm5, ymm1, ymm3 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm6, ymm1, ymm3 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm6, ymm2, ymm3 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm7, ymm2, ymm3 ; ymm7=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm2, ymm7, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm6, ymm7, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpmullw ymm0, ymm0, [GOTOFF(ebx,PW_THREE)] + vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)] + + vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vmovdqu YMMWORD [edx+2*SIZEOF_YMMWORD], ymm1 ; temporarily save + vmovdqu YMMWORD [edx+3*SIZEOF_YMMWORD], ymm5 ; the intermediate data + vmovdqu YMMWORD [edi+2*SIZEOF_YMMWORD], ymm2 + vmovdqu YMMWORD [edi+3*SIZEOF_YMMWORD], ymm6 + + vperm2i128 ymm1, ymm3, ymm1, 0x20 + vpslldq ymm1, ymm1, 14 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0) + vperm2i128 ymm2, ymm3, ymm2, 0x20 + vpslldq ymm2, ymm2, 14 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0) + + vmovdqa YMMWORD [wk(2)], ymm1 + vmovdqa YMMWORD [wk(3)], ymm2 + +.upsample: + ; -- process the upper row + + vmovdqu ymm7, YMMWORD [edx+0*SIZEOF_YMMWORD] ; ymm7=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vmovdqu ymm3, YMMWORD [edx+1*SIZEOF_YMMWORD] ; ymm3=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's) + + vperm2i128 ymm0, ymm1, ymm7, 0x03 + vpalignr ymm0, ymm0, ymm7, 2 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --) + vperm2i128 ymm4, ymm1, ymm3, 0x20 + vpslldq ymm4, ymm4, 14 ; ymm4=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16) + + vperm2i128 ymm5, ymm1, ymm7, 0x03 + vpsrldq ymm5, ymm5, 14 ; ymm5=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm6, ymm1, ymm3, 0x20 + vpalignr ymm6, ymm3, ymm6, 14 ; ymm6=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vpor ymm0, ymm0, ymm4 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16) + vpor ymm5, ymm5, ymm6 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vperm2i128 ymm2, ymm1, ymm3, 0x03 + vpalignr ymm2, ymm2, ymm3, 2 ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --) + vperm2i128 ymm4, ymm1, ymm3, 0x03 + vpsrldq ymm4, ymm4, 14 ; ymm4=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm1, ymm1, ymm7, 0x20 + vpalignr ymm1, ymm7, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + + vpor ymm1, ymm1, YMMWORD [wk(0)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + vpor ymm2, ymm2, YMMWORD [wk(2)] ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32) + + vmovdqa YMMWORD [wk(0)], ymm4 + + vpmullw ymm7, ymm7, [GOTOFF(ebx,PW_THREE)] + vpmullw ymm3, ymm3, [GOTOFF(ebx,PW_THREE)] + vpaddw ymm1, ymm1, [GOTOFF(ebx,PW_EIGHT)] + vpaddw ymm5, ymm5, [GOTOFF(ebx,PW_EIGHT)] + vpaddw ymm0, ymm0, [GOTOFF(ebx,PW_SEVEN)] + vpaddw ymm2, [GOTOFF(ebx,PW_SEVEN)] + + vpaddw ymm1, ymm1, ymm7 + vpaddw ymm5, ymm5, ymm3 + vpsrlw ymm1, ymm1, 4 ; ymm1=Out0LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30) + vpsrlw ymm5, ymm5, 4 ; ymm5=Out0HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62) + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm2, ymm2, ymm3 + vpsrlw ymm0, ymm0, 4 ; ymm0=Out0LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31) + vpsrlw ymm2, ymm2, 4 ; ymm2=Out0HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63) + + vpsllw ymm0, ymm0, BYTE_BIT + vpsllw ymm2, ymm2, BYTE_BIT + vpor ymm1, ymm1, ymm0 ; ymm1=Out0L=( 0 1 2 ... 29 30 31) + vpor ymm5, ymm5, ymm2 ; ymm5=Out0H=(32 33 34 ... 61 62 63) + + vmovdqu YMMWORD [edx+0*SIZEOF_YMMWORD], ymm1 + vmovdqu YMMWORD [edx+1*SIZEOF_YMMWORD], ymm5 + + ; -- process the lower row + + vmovdqu ymm6, YMMWORD [edi+0*SIZEOF_YMMWORD] ; ymm6=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vmovdqu ymm4, YMMWORD [edi+1*SIZEOF_YMMWORD] ; ymm4=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's) + + vperm2i128 ymm7, ymm1, ymm6, 0x03 + vpalignr ymm7, ymm7, ymm6, 2 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --) + vperm2i128 ymm3, ymm1, ymm4, 0x20 + vpslldq ymm3, ymm3, 14 ; ymm3=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16) + + vperm2i128 ymm0, ymm1, ymm6, 0x03 + vpsrldq ymm0, ymm0, 14 ; ymm0=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm2, ymm1, ymm4, 0x20 + vpalignr ymm2, ymm4, ymm2, 14 ; ymm2=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vpor ymm7, ymm7, ymm3 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16) + vpor ymm0, ymm0, ymm2 ; ymm0=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vperm2i128 ymm5, ymm1, ymm4, 0x03 + vpalignr ymm5, ymm5, ymm4, 2 ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --) + vperm2i128 ymm3, ymm1, ymm4, 0x03 + vpsrldq ymm3, ymm3, 14 ; ymm3=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm1, ymm1, ymm6, 0x20 + vpalignr ymm1, ymm6, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + + vpor ymm1, ymm1, YMMWORD [wk(1)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + vpor ymm5, ymm5, YMMWORD [wk(3)] ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32) + + vmovdqa YMMWORD [wk(1)], ymm3 + + vpmullw ymm6, ymm6, [GOTOFF(ebx,PW_THREE)] + vpmullw ymm4, ymm4, [GOTOFF(ebx,PW_THREE)] + vpaddw ymm1, ymm1, [GOTOFF(ebx,PW_EIGHT)] + vpaddw ymm0, ymm0, [GOTOFF(ebx,PW_EIGHT)] + vpaddw ymm7, ymm7, [GOTOFF(ebx,PW_SEVEN)] + vpaddw ymm5, ymm5, [GOTOFF(ebx,PW_SEVEN)] + + vpaddw ymm1, ymm1, ymm6 + vpaddw ymm0, ymm0, ymm4 + vpsrlw ymm1, ymm1, 4 ; ymm1=Out1LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30) + vpsrlw ymm0, ymm0, 4 ; ymm0=Out1HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62) + vpaddw ymm7, ymm7, ymm6 + vpaddw ymm5, ymm5, ymm4 + vpsrlw ymm7, ymm7, 4 ; ymm7=Out1LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31) + vpsrlw ymm5, ymm5, 4 ; ymm5=Out1HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63) + + vpsllw ymm7, ymm7, BYTE_BIT + vpsllw ymm5, ymm5, BYTE_BIT + vpor ymm1, ymm1, ymm7 ; ymm1=Out1L=( 0 1 2 ... 29 30 31) + vpor ymm0, ymm0, ymm5 ; ymm0=Out1H=(32 33 34 ... 61 62 63) + + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm1 + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm0 + + POPPIC ebx + + sub eax, byte SIZEOF_YMMWORD + add ecx, byte 1*SIZEOF_YMMWORD ; inptr1(above) + add ebx, byte 1*SIZEOF_YMMWORD ; inptr0 + add esi, byte 1*SIZEOF_YMMWORD ; inptr1(below) + add edx, byte 2*SIZEOF_YMMWORD ; outptr0 + add edi, byte 2*SIZEOF_YMMWORD ; outptr1 + cmp eax, byte SIZEOF_YMMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop esi + pop edi + pop ecx + pop eax + + add esi, byte 1*SIZEOF_JSAMPROW ; input_data + add edi, byte 2*SIZEOF_JSAMPROW ; output_data + sub ecx, byte 2 ; rowctr + jg near .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 1:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v1_upsample_avx2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define output_width(b) (b) + 12 ; JDIMENSION output_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_upsample_avx2) + +EXTN(jsimd_h2v1_upsample_avx2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov edx, JDIMENSION [output_width(ebp)] + add edx, byte (SIZEOF_YMMWORD-1) + and edx, -SIZEOF_YMMWORD + jz short .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz short .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + mov eax, edx ; colctr + ALIGNX 16, 7 +.columnloop: + + cmp eax, byte SIZEOF_YMMWORD + ja near .above_16 + + vmovdqu xmm0, XMMWORD [esi+0*SIZEOF_YMMWORD] + vpunpckhbw xmm1, xmm0, xmm0 + vpunpcklbw xmm0, xmm0, xmm0 + + vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0 + vmovdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1 + + jmp short .nextrow + +.above_16: + vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD] + + vpermq ymm0, ymm0, 0xd8 + vpunpckhbw ymm1, ymm0, ymm0 + vpunpcklbw ymm0, ymm0, ymm0 + + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0 + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm1 + + sub eax, byte 2*SIZEOF_YMMWORD + jz short .nextrow + + add esi, byte SIZEOF_YMMWORD ; inptr + add edi, byte 2*SIZEOF_YMMWORD ; outptr + jmp short .columnloop + ALIGNX 16, 7 + +.nextrow: + pop esi + pop edi + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec ecx ; rowctr + jg short .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 2:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v2_upsample_avx2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define output_width(b) (b) + 12 ; JDIMENSION output_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_upsample_avx2) + +EXTN(jsimd_h2v2_upsample_avx2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov edx, JDIMENSION [output_width(ebp)] + add edx, byte (SIZEOF_YMMWORD-1) + and edx, -SIZEOF_YMMWORD + jz near .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov ebx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0 + mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1 + mov eax, edx ; colctr + ALIGNX 16, 7 +.columnloop: + + cmp eax, byte SIZEOF_YMMWORD + ja short .above_16 + + vmovdqu xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD] + vpunpckhbw xmm1, xmm0, xmm0 + vpunpcklbw xmm0, xmm0, xmm0 + + vmovdqu XMMWORD [ebx+0*SIZEOF_XMMWORD], xmm0 + vmovdqu XMMWORD [ebx+1*SIZEOF_XMMWORD], xmm1 + vmovdqu XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0 + vmovdqu XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1 + + jmp near .nextrow + +.above_16: + vmovdqu ymm0, YMMWORD [esi+0*SIZEOF_YMMWORD] + + vpermq ymm0, ymm0, 0xd8 + vpunpckhbw ymm1, ymm0, ymm0 + vpunpcklbw ymm0, ymm0, ymm0 + + vmovdqu YMMWORD [ebx+0*SIZEOF_YMMWORD], ymm0 + vmovdqu YMMWORD [ebx+1*SIZEOF_YMMWORD], ymm1 + vmovdqu YMMWORD [edi+0*SIZEOF_YMMWORD], ymm0 + vmovdqu YMMWORD [edi+1*SIZEOF_YMMWORD], ymm1 + + sub eax, byte 2*SIZEOF_YMMWORD + jz short .nextrow + + add esi, byte SIZEOF_YMMWORD ; inptr + add ebx, 2*SIZEOF_YMMWORD ; outptr0 + add edi, 2*SIZEOF_YMMWORD ; outptr1 + jmp short .columnloop + ALIGNX 16, 7 + +.nextrow: + pop esi + pop edi + + add esi, byte 1*SIZEOF_JSAMPROW ; input_data + add edi, byte 2*SIZEOF_JSAMPROW ; output_data + sub ecx, byte 2 ; rowctr + jg near .rowloop + +.return: + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-mmx.asm new file mode 100644 index 0000000..01d09e6 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-mmx.asm @@ -0,0 +1,727 @@ +; +; jdsample.asm - upsampling (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fancy_upsample_mmx) + +EXTN(jconst_fancy_upsample_mmx): + +PW_ONE times 4 dw 1 +PW_TWO times 4 dw 2 +PW_THREE times 4 dw 3 +PW_SEVEN times 4 dw 7 +PW_EIGHT times 4 dw 8 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical. +; +; The upsampling algorithm is linear interpolation between pixel centers, +; also known as a "triangle filter". This is a good compromise between +; speed and visual quality. The centers of the output pixels are 1/4 and 3/4 +; of the way between input pixel centers. +; +; GLOBAL(void) +; jsimd_h2v1_fancy_upsample_mmx(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define downsamp_width(b) (b) + 12 ; JDIMENSION downsampled_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_fancy_upsample_mmx) + +EXTN(jsimd_h2v1_fancy_upsample_mmx): + push ebp + mov ebp, esp + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + mov eax, JDIMENSION [downsamp_width(ebp)] ; colctr + test eax, eax + jz near .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push eax ; colctr + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + + test eax, SIZEOF_MMWORD-1 + jz short .skip + mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample +.skip: + pxor mm0, mm0 ; mm0=(all 0's) + pcmpeqb mm7, mm7 + psrlq mm7, (SIZEOF_MMWORD-1)*BYTE_BIT + pand mm7, MMWORD [esi+0*SIZEOF_MMWORD] + + add eax, byte SIZEOF_MMWORD-1 + and eax, byte -SIZEOF_MMWORD + cmp eax, byte SIZEOF_MMWORD + ja short .columnloop + ALIGNX 16, 7 + +.columnloop_last: + pcmpeqb mm6, mm6 + psllq mm6, (SIZEOF_MMWORD-1)*BYTE_BIT + pand mm6, MMWORD [esi+0*SIZEOF_MMWORD] + jmp short .upsample + ALIGNX 16, 7 + +.columnloop: + movq mm6, MMWORD [esi+1*SIZEOF_MMWORD] + psllq mm6, (SIZEOF_MMWORD-1)*BYTE_BIT + +.upsample: + movq mm1, MMWORD [esi+0*SIZEOF_MMWORD] + movq mm2, mm1 + movq mm3, mm1 ; mm1=( 0 1 2 3 4 5 6 7) + psllq mm2, BYTE_BIT ; mm2=( - 0 1 2 3 4 5 6) + psrlq mm3, BYTE_BIT ; mm3=( 1 2 3 4 5 6 7 -) + + por mm2, mm7 ; mm2=(-1 0 1 2 3 4 5 6) + por mm3, mm6 ; mm3=( 1 2 3 4 5 6 7 8) + + movq mm7, mm1 + psrlq mm7, (SIZEOF_MMWORD-1)*BYTE_BIT ; mm7=( 7 - - - - - - -) + + movq mm4, mm1 + punpcklbw mm1, mm0 ; mm1=( 0 1 2 3) + punpckhbw mm4, mm0 ; mm4=( 4 5 6 7) + movq mm5, mm2 + punpcklbw mm2, mm0 ; mm2=(-1 0 1 2) + punpckhbw mm5, mm0 ; mm5=( 3 4 5 6) + movq mm6, mm3 + punpcklbw mm3, mm0 ; mm3=( 1 2 3 4) + punpckhbw mm6, mm0 ; mm6=( 5 6 7 8) + + pmullw mm1, [GOTOFF(ebx,PW_THREE)] + pmullw mm4, [GOTOFF(ebx,PW_THREE)] + paddw mm2, [GOTOFF(ebx,PW_ONE)] + paddw mm5, [GOTOFF(ebx,PW_ONE)] + paddw mm3, [GOTOFF(ebx,PW_TWO)] + paddw mm6, [GOTOFF(ebx,PW_TWO)] + + paddw mm2, mm1 + paddw mm5, mm4 + psrlw mm2, 2 ; mm2=OutLE=( 0 2 4 6) + psrlw mm5, 2 ; mm5=OutHE=( 8 10 12 14) + paddw mm3, mm1 + paddw mm6, mm4 + psrlw mm3, 2 ; mm3=OutLO=( 1 3 5 7) + psrlw mm6, 2 ; mm6=OutHO=( 9 11 13 15) + + psllw mm3, BYTE_BIT + psllw mm6, BYTE_BIT + por mm2, mm3 ; mm2=OutL=( 0 1 2 3 4 5 6 7) + por mm5, mm6 ; mm5=OutH=( 8 9 10 11 12 13 14 15) + + movq MMWORD [edi+0*SIZEOF_MMWORD], mm2 + movq MMWORD [edi+1*SIZEOF_MMWORD], mm5 + + sub eax, byte SIZEOF_MMWORD + add esi, byte 1*SIZEOF_MMWORD ; inptr + add edi, byte 2*SIZEOF_MMWORD ; outptr + cmp eax, byte SIZEOF_MMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop esi + pop edi + pop eax + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec ecx ; rowctr + jg near .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical. +; Again a triangle filter; see comments for h2v1 case, above. +; +; GLOBAL(void) +; jsimd_h2v2_fancy_upsample_mmx(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define downsamp_width(b) (b) + 12 ; JDIMENSION downsampled_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD ; mmword wk[WK_NUM] +%define WK_NUM 4 +%define gotptr wk(0) - SIZEOF_POINTER ; void *gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_fancy_upsample_mmx) + +EXTN(jsimd_h2v2_fancy_upsample_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov edx, eax ; edx = original ebp + mov eax, JDIMENSION [downsamp_width(edx)] ; colctr + test eax, eax + jz near .return + + mov ecx, INT [max_v_samp(edx)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(edx)] ; input_data + mov edi, POINTER [output_data_ptr(edx)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push eax ; colctr + push ecx + push edi + push esi + + mov ecx, JSAMPROW [esi-1*SIZEOF_JSAMPROW] ; inptr1(above) + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0 + mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1(below) + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0 + mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1 + + test eax, SIZEOF_MMWORD-1 + jz short .skip + push edx + mov dl, JSAMPLE [ecx+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [ecx+eax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [ebx+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [ebx+eax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample + pop edx +.skip: + ; -- process the first column block + + movq mm0, MMWORD [ebx+0*SIZEOF_MMWORD] ; mm0=row[ 0][0] + movq mm1, MMWORD [ecx+0*SIZEOF_MMWORD] ; mm1=row[-1][0] + movq mm2, MMWORD [esi+0*SIZEOF_MMWORD] ; mm2=row[+1][0] + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + pxor mm3, mm3 ; mm3=(all 0's) + movq mm4, mm0 + punpcklbw mm0, mm3 ; mm0=row[ 0][0]( 0 1 2 3) + punpckhbw mm4, mm3 ; mm4=row[ 0][0]( 4 5 6 7) + movq mm5, mm1 + punpcklbw mm1, mm3 ; mm1=row[-1][0]( 0 1 2 3) + punpckhbw mm5, mm3 ; mm5=row[-1][0]( 4 5 6 7) + movq mm6, mm2 + punpcklbw mm2, mm3 ; mm2=row[+1][0]( 0 1 2 3) + punpckhbw mm6, mm3 ; mm6=row[+1][0]( 4 5 6 7) + + pmullw mm0, [GOTOFF(ebx,PW_THREE)] + pmullw mm4, [GOTOFF(ebx,PW_THREE)] + + pcmpeqb mm7, mm7 + psrlq mm7, (SIZEOF_MMWORD-2)*BYTE_BIT + + paddw mm1, mm0 ; mm1=Int0L=( 0 1 2 3) + paddw mm5, mm4 ; mm5=Int0H=( 4 5 6 7) + paddw mm2, mm0 ; mm2=Int1L=( 0 1 2 3) + paddw mm6, mm4 ; mm6=Int1H=( 4 5 6 7) + + movq MMWORD [edx+0*SIZEOF_MMWORD], mm1 ; temporarily save + movq MMWORD [edx+1*SIZEOF_MMWORD], mm5 ; the intermediate data + movq MMWORD [edi+0*SIZEOF_MMWORD], mm2 + movq MMWORD [edi+1*SIZEOF_MMWORD], mm6 + + pand mm1, mm7 ; mm1=( 0 - - -) + pand mm2, mm7 ; mm2=( 0 - - -) + + movq MMWORD [wk(0)], mm1 + movq MMWORD [wk(1)], mm2 + + POPPIC ebx + + add eax, byte SIZEOF_MMWORD-1 + and eax, byte -SIZEOF_MMWORD + cmp eax, byte SIZEOF_MMWORD + ja short .columnloop + ALIGNX 16, 7 + +.columnloop_last: + ; -- process the last column block + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + pcmpeqb mm1, mm1 + psllq mm1, (SIZEOF_MMWORD-2)*BYTE_BIT + movq mm2, mm1 + + pand mm1, MMWORD [edx+1*SIZEOF_MMWORD] ; mm1=( - - - 7) + pand mm2, MMWORD [edi+1*SIZEOF_MMWORD] ; mm2=( - - - 7) + + movq MMWORD [wk(2)], mm1 + movq MMWORD [wk(3)], mm2 + + jmp short .upsample + ALIGNX 16, 7 + +.columnloop: + ; -- process the next column block + + movq mm0, MMWORD [ebx+1*SIZEOF_MMWORD] ; mm0=row[ 0][1] + movq mm1, MMWORD [ecx+1*SIZEOF_MMWORD] ; mm1=row[-1][1] + movq mm2, MMWORD [esi+1*SIZEOF_MMWORD] ; mm2=row[+1][1] + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + pxor mm3, mm3 ; mm3=(all 0's) + movq mm4, mm0 + punpcklbw mm0, mm3 ; mm0=row[ 0][1]( 0 1 2 3) + punpckhbw mm4, mm3 ; mm4=row[ 0][1]( 4 5 6 7) + movq mm5, mm1 + punpcklbw mm1, mm3 ; mm1=row[-1][1]( 0 1 2 3) + punpckhbw mm5, mm3 ; mm5=row[-1][1]( 4 5 6 7) + movq mm6, mm2 + punpcklbw mm2, mm3 ; mm2=row[+1][1]( 0 1 2 3) + punpckhbw mm6, mm3 ; mm6=row[+1][1]( 4 5 6 7) + + pmullw mm0, [GOTOFF(ebx,PW_THREE)] + pmullw mm4, [GOTOFF(ebx,PW_THREE)] + + paddw mm1, mm0 ; mm1=Int0L=( 0 1 2 3) + paddw mm5, mm4 ; mm5=Int0H=( 4 5 6 7) + paddw mm2, mm0 ; mm2=Int1L=( 0 1 2 3) + paddw mm6, mm4 ; mm6=Int1H=( 4 5 6 7) + + movq MMWORD [edx+2*SIZEOF_MMWORD], mm1 ; temporarily save + movq MMWORD [edx+3*SIZEOF_MMWORD], mm5 ; the intermediate data + movq MMWORD [edi+2*SIZEOF_MMWORD], mm2 + movq MMWORD [edi+3*SIZEOF_MMWORD], mm6 + + psllq mm1, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm1=( - - - 0) + psllq mm2, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm2=( - - - 0) + + movq MMWORD [wk(2)], mm1 + movq MMWORD [wk(3)], mm2 + +.upsample: + ; -- process the upper row + + movq mm7, MMWORD [edx+0*SIZEOF_MMWORD] ; mm7=Int0L=( 0 1 2 3) + movq mm3, MMWORD [edx+1*SIZEOF_MMWORD] ; mm3=Int0H=( 4 5 6 7) + + movq mm0, mm7 + movq mm4, mm3 + psrlq mm0, 2*BYTE_BIT ; mm0=( 1 2 3 -) + psllq mm4, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm4=( - - - 4) + movq mm5, mm7 + movq mm6, mm3 + psrlq mm5, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm5=( 3 - - -) + psllq mm6, 2*BYTE_BIT ; mm6=( - 4 5 6) + + por mm0, mm4 ; mm0=( 1 2 3 4) + por mm5, mm6 ; mm5=( 3 4 5 6) + + movq mm1, mm7 + movq mm2, mm3 + psllq mm1, 2*BYTE_BIT ; mm1=( - 0 1 2) + psrlq mm2, 2*BYTE_BIT ; mm2=( 5 6 7 -) + movq mm4, mm3 + psrlq mm4, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm4=( 7 - - -) + + por mm1, MMWORD [wk(0)] ; mm1=(-1 0 1 2) + por mm2, MMWORD [wk(2)] ; mm2=( 5 6 7 8) + + movq MMWORD [wk(0)], mm4 + + pmullw mm7, [GOTOFF(ebx,PW_THREE)] + pmullw mm3, [GOTOFF(ebx,PW_THREE)] + paddw mm1, [GOTOFF(ebx,PW_EIGHT)] + paddw mm5, [GOTOFF(ebx,PW_EIGHT)] + paddw mm0, [GOTOFF(ebx,PW_SEVEN)] + paddw mm2, [GOTOFF(ebx,PW_SEVEN)] + + paddw mm1, mm7 + paddw mm5, mm3 + psrlw mm1, 4 ; mm1=Out0LE=( 0 2 4 6) + psrlw mm5, 4 ; mm5=Out0HE=( 8 10 12 14) + paddw mm0, mm7 + paddw mm2, mm3 + psrlw mm0, 4 ; mm0=Out0LO=( 1 3 5 7) + psrlw mm2, 4 ; mm2=Out0HO=( 9 11 13 15) + + psllw mm0, BYTE_BIT + psllw mm2, BYTE_BIT + por mm1, mm0 ; mm1=Out0L=( 0 1 2 3 4 5 6 7) + por mm5, mm2 ; mm5=Out0H=( 8 9 10 11 12 13 14 15) + + movq MMWORD [edx+0*SIZEOF_MMWORD], mm1 + movq MMWORD [edx+1*SIZEOF_MMWORD], mm5 + + ; -- process the lower row + + movq mm6, MMWORD [edi+0*SIZEOF_MMWORD] ; mm6=Int1L=( 0 1 2 3) + movq mm4, MMWORD [edi+1*SIZEOF_MMWORD] ; mm4=Int1H=( 4 5 6 7) + + movq mm7, mm6 + movq mm3, mm4 + psrlq mm7, 2*BYTE_BIT ; mm7=( 1 2 3 -) + psllq mm3, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm3=( - - - 4) + movq mm0, mm6 + movq mm2, mm4 + psrlq mm0, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm0=( 3 - - -) + psllq mm2, 2*BYTE_BIT ; mm2=( - 4 5 6) + + por mm7, mm3 ; mm7=( 1 2 3 4) + por mm0, mm2 ; mm0=( 3 4 5 6) + + movq mm1, mm6 + movq mm5, mm4 + psllq mm1, 2*BYTE_BIT ; mm1=( - 0 1 2) + psrlq mm5, 2*BYTE_BIT ; mm5=( 5 6 7 -) + movq mm3, mm4 + psrlq mm3, (SIZEOF_MMWORD-2)*BYTE_BIT ; mm3=( 7 - - -) + + por mm1, MMWORD [wk(1)] ; mm1=(-1 0 1 2) + por mm5, MMWORD [wk(3)] ; mm5=( 5 6 7 8) + + movq MMWORD [wk(1)], mm3 + + pmullw mm6, [GOTOFF(ebx,PW_THREE)] + pmullw mm4, [GOTOFF(ebx,PW_THREE)] + paddw mm1, [GOTOFF(ebx,PW_EIGHT)] + paddw mm0, [GOTOFF(ebx,PW_EIGHT)] + paddw mm7, [GOTOFF(ebx,PW_SEVEN)] + paddw mm5, [GOTOFF(ebx,PW_SEVEN)] + + paddw mm1, mm6 + paddw mm0, mm4 + psrlw mm1, 4 ; mm1=Out1LE=( 0 2 4 6) + psrlw mm0, 4 ; mm0=Out1HE=( 8 10 12 14) + paddw mm7, mm6 + paddw mm5, mm4 + psrlw mm7, 4 ; mm7=Out1LO=( 1 3 5 7) + psrlw mm5, 4 ; mm5=Out1HO=( 9 11 13 15) + + psllw mm7, BYTE_BIT + psllw mm5, BYTE_BIT + por mm1, mm7 ; mm1=Out1L=( 0 1 2 3 4 5 6 7) + por mm0, mm5 ; mm0=Out1H=( 8 9 10 11 12 13 14 15) + + movq MMWORD [edi+0*SIZEOF_MMWORD], mm1 + movq MMWORD [edi+1*SIZEOF_MMWORD], mm0 + + POPPIC ebx + + sub eax, byte SIZEOF_MMWORD + add ecx, byte 1*SIZEOF_MMWORD ; inptr1(above) + add ebx, byte 1*SIZEOF_MMWORD ; inptr0 + add esi, byte 1*SIZEOF_MMWORD ; inptr1(below) + add edx, byte 2*SIZEOF_MMWORD ; outptr0 + add edi, byte 2*SIZEOF_MMWORD ; outptr1 + cmp eax, byte SIZEOF_MMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop esi + pop edi + pop ecx + pop eax + + add esi, byte 1*SIZEOF_JSAMPROW ; input_data + add edi, byte 2*SIZEOF_JSAMPROW ; output_data + sub ecx, byte 2 ; rowctr + jg near .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 1:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v1_upsample_mmx(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define output_width(b) (b) + 12 ; JDIMENSION output_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_upsample_mmx) + +EXTN(jsimd_h2v1_upsample_mmx): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov edx, JDIMENSION [output_width(ebp)] + add edx, byte (2*SIZEOF_MMWORD)-1 + and edx, byte -(2*SIZEOF_MMWORD) + jz short .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz short .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + mov eax, edx ; colctr + ALIGNX 16, 7 +.columnloop: + + movq mm0, MMWORD [esi+0*SIZEOF_MMWORD] + + movq mm1, mm0 + punpcklbw mm0, mm0 + punpckhbw mm1, mm1 + + movq MMWORD [edi+0*SIZEOF_MMWORD], mm0 + movq MMWORD [edi+1*SIZEOF_MMWORD], mm1 + + sub eax, byte 2*SIZEOF_MMWORD + jz short .nextrow + + movq mm2, MMWORD [esi+1*SIZEOF_MMWORD] + + movq mm3, mm2 + punpcklbw mm2, mm2 + punpckhbw mm3, mm3 + + movq MMWORD [edi+2*SIZEOF_MMWORD], mm2 + movq MMWORD [edi+3*SIZEOF_MMWORD], mm3 + + sub eax, byte 2*SIZEOF_MMWORD + jz short .nextrow + + add esi, byte 2*SIZEOF_MMWORD ; inptr + add edi, byte 4*SIZEOF_MMWORD ; outptr + jmp short .columnloop + ALIGNX 16, 7 + +.nextrow: + pop esi + pop edi + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec ecx ; rowctr + jg short .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 2:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v2_upsample_mmx(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define output_width(b) (b) + 12 ; JDIMENSION output_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_upsample_mmx) + +EXTN(jsimd_h2v2_upsample_mmx): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov edx, JDIMENSION [output_width(ebp)] + add edx, byte (2*SIZEOF_MMWORD)-1 + and edx, byte -(2*SIZEOF_MMWORD) + jz near .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz short .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov ebx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0 + mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1 + mov eax, edx ; colctr + ALIGNX 16, 7 +.columnloop: + + movq mm0, MMWORD [esi+0*SIZEOF_MMWORD] + + movq mm1, mm0 + punpcklbw mm0, mm0 + punpckhbw mm1, mm1 + + movq MMWORD [ebx+0*SIZEOF_MMWORD], mm0 + movq MMWORD [ebx+1*SIZEOF_MMWORD], mm1 + movq MMWORD [edi+0*SIZEOF_MMWORD], mm0 + movq MMWORD [edi+1*SIZEOF_MMWORD], mm1 + + sub eax, byte 2*SIZEOF_MMWORD + jz short .nextrow + + movq mm2, MMWORD [esi+1*SIZEOF_MMWORD] + + movq mm3, mm2 + punpcklbw mm2, mm2 + punpckhbw mm3, mm3 + + movq MMWORD [ebx+2*SIZEOF_MMWORD], mm2 + movq MMWORD [ebx+3*SIZEOF_MMWORD], mm3 + movq MMWORD [edi+2*SIZEOF_MMWORD], mm2 + movq MMWORD [edi+3*SIZEOF_MMWORD], mm3 + + sub eax, byte 2*SIZEOF_MMWORD + jz short .nextrow + + add esi, byte 2*SIZEOF_MMWORD ; inptr + add ebx, byte 4*SIZEOF_MMWORD ; outptr0 + add edi, byte 4*SIZEOF_MMWORD ; outptr1 + jmp short .columnloop + ALIGNX 16, 7 + +.nextrow: + pop esi + pop edi + + add esi, byte 1*SIZEOF_JSAMPROW ; input_data + add edi, byte 2*SIZEOF_JSAMPROW ; output_data + sub ecx, byte 2 ; rowctr + jg short .rowloop + + emms ; empty MMX state + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-sse2.asm new file mode 100644 index 0000000..b10d922 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jdsample-sse2.asm @@ -0,0 +1,720 @@ +; +; jdsample.asm - upsampling (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fancy_upsample_sse2) + +EXTN(jconst_fancy_upsample_sse2): + +PW_ONE times 8 dw 1 +PW_TWO times 8 dw 2 +PW_THREE times 8 dw 3 +PW_SEVEN times 8 dw 7 +PW_EIGHT times 8 dw 8 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical. +; +; The upsampling algorithm is linear interpolation between pixel centers, +; also known as a "triangle filter". This is a good compromise between +; speed and visual quality. The centers of the output pixels are 1/4 and 3/4 +; of the way between input pixel centers. +; +; GLOBAL(void) +; jsimd_h2v1_fancy_upsample_sse2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define downsamp_width(b) (b) + 12 ; JDIMENSION downsampled_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_fancy_upsample_sse2) + +EXTN(jsimd_h2v1_fancy_upsample_sse2): + push ebp + mov ebp, esp + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + mov eax, JDIMENSION [downsamp_width(ebp)] ; colctr + test eax, eax + jz near .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push eax ; colctr + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + + test eax, SIZEOF_XMMWORD-1 + jz short .skip + mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample +.skip: + pxor xmm0, xmm0 ; xmm0=(all 0's) + pcmpeqb xmm7, xmm7 + psrldq xmm7, (SIZEOF_XMMWORD-1) + pand xmm7, XMMWORD [esi+0*SIZEOF_XMMWORD] + + add eax, byte SIZEOF_XMMWORD-1 + and eax, byte -SIZEOF_XMMWORD + cmp eax, byte SIZEOF_XMMWORD + ja short .columnloop + ALIGNX 16, 7 + +.columnloop_last: + pcmpeqb xmm6, xmm6 + pslldq xmm6, (SIZEOF_XMMWORD-1) + pand xmm6, XMMWORD [esi+0*SIZEOF_XMMWORD] + jmp short .upsample + ALIGNX 16, 7 + +.columnloop: + movdqa xmm6, XMMWORD [esi+1*SIZEOF_XMMWORD] + pslldq xmm6, (SIZEOF_XMMWORD-1) + +.upsample: + movdqa xmm1, XMMWORD [esi+0*SIZEOF_XMMWORD] + movdqa xmm2, xmm1 + movdqa xmm3, xmm1 ; xmm1=( 0 1 2 ... 13 14 15) + pslldq xmm2, 1 ; xmm2=(-- 0 1 ... 12 13 14) + psrldq xmm3, 1 ; xmm3=( 1 2 3 ... 14 15 --) + + por xmm2, xmm7 ; xmm2=(-1 0 1 ... 12 13 14) + por xmm3, xmm6 ; xmm3=( 1 2 3 ... 14 15 16) + + movdqa xmm7, xmm1 + psrldq xmm7, (SIZEOF_XMMWORD-1) ; xmm7=(15 -- -- ... -- -- --) + + movdqa xmm4, xmm1 + punpcklbw xmm1, xmm0 ; xmm1=( 0 1 2 3 4 5 6 7) + punpckhbw xmm4, xmm0 ; xmm4=( 8 9 10 11 12 13 14 15) + movdqa xmm5, xmm2 + punpcklbw xmm2, xmm0 ; xmm2=(-1 0 1 2 3 4 5 6) + punpckhbw xmm5, xmm0 ; xmm5=( 7 8 9 10 11 12 13 14) + movdqa xmm6, xmm3 + punpcklbw xmm3, xmm0 ; xmm3=( 1 2 3 4 5 6 7 8) + punpckhbw xmm6, xmm0 ; xmm6=( 9 10 11 12 13 14 15 16) + + pmullw xmm1, [GOTOFF(ebx,PW_THREE)] + pmullw xmm4, [GOTOFF(ebx,PW_THREE)] + paddw xmm2, [GOTOFF(ebx,PW_ONE)] + paddw xmm5, [GOTOFF(ebx,PW_ONE)] + paddw xmm3, [GOTOFF(ebx,PW_TWO)] + paddw xmm6, [GOTOFF(ebx,PW_TWO)] + + paddw xmm2, xmm1 + paddw xmm5, xmm4 + psrlw xmm2, 2 ; xmm2=OutLE=( 0 2 4 6 8 10 12 14) + psrlw xmm5, 2 ; xmm5=OutHE=(16 18 20 22 24 26 28 30) + paddw xmm3, xmm1 + paddw xmm6, xmm4 + psrlw xmm3, 2 ; xmm3=OutLO=( 1 3 5 7 9 11 13 15) + psrlw xmm6, 2 ; xmm6=OutHO=(17 19 21 23 25 27 29 31) + + psllw xmm3, BYTE_BIT + psllw xmm6, BYTE_BIT + por xmm2, xmm3 ; xmm2=OutL=( 0 1 2 ... 13 14 15) + por xmm5, xmm6 ; xmm5=OutH=(16 17 18 ... 29 30 31) + + movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm5 + + sub eax, byte SIZEOF_XMMWORD + add esi, byte 1*SIZEOF_XMMWORD ; inptr + add edi, byte 2*SIZEOF_XMMWORD ; outptr + cmp eax, byte SIZEOF_XMMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop esi + pop edi + pop eax + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec ecx ; rowctr + jg near .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical. +; Again a triangle filter; see comments for h2v1 case, above. +; +; GLOBAL(void) +; jsimd_h2v2_fancy_upsample_sse2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define downsamp_width(b) (b) + 12 ; JDIMENSION downsampled_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 4 +%define gotptr wk(0) - SIZEOF_POINTER ; void *gotptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_fancy_upsample_sse2) + +EXTN(jsimd_h2v2_fancy_upsample_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC eax ; make a room for GOT address + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + MOVPIC POINTER [gotptr], ebx ; save GOT address + + mov edx, eax ; edx = original ebp + mov eax, JDIMENSION [downsamp_width(edx)] ; colctr + test eax, eax + jz near .return + + mov ecx, INT [max_v_samp(edx)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(edx)] ; input_data + mov edi, POINTER [output_data_ptr(edx)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push eax ; colctr + push ecx + push edi + push esi + + mov ecx, JSAMPROW [esi-1*SIZEOF_JSAMPROW] ; inptr1(above) + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; inptr0 + mov esi, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; inptr1(below) + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0 + mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1 + + test eax, SIZEOF_XMMWORD-1 + jz short .skip + push edx + mov dl, JSAMPLE [ecx+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [ecx+eax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [ebx+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [ebx+eax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [esi+(eax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [esi+eax*SIZEOF_JSAMPLE], dl ; insert a dummy sample + pop edx +.skip: + ; -- process the first column block + + movdqa xmm0, XMMWORD [ebx+0*SIZEOF_XMMWORD] ; xmm0=row[ 0][0] + movdqa xmm1, XMMWORD [ecx+0*SIZEOF_XMMWORD] ; xmm1=row[-1][0] + movdqa xmm2, XMMWORD [esi+0*SIZEOF_XMMWORD] ; xmm2=row[+1][0] + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + pxor xmm3, xmm3 ; xmm3=(all 0's) + movdqa xmm4, xmm0 + punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7) + punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15) + movdqa xmm5, xmm1 + punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15) + movdqa xmm6, xmm2 + punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15) + + pmullw xmm0, [GOTOFF(ebx,PW_THREE)] + pmullw xmm4, [GOTOFF(ebx,PW_THREE)] + + pcmpeqb xmm7, xmm7 + psrldq xmm7, (SIZEOF_XMMWORD-2) + + paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7) + paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15) + paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7) + paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15) + + movdqa XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1 ; temporarily save + movdqa XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5 ; the intermediate data + movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm6 + + pand xmm1, xmm7 ; xmm1=( 0 -- -- -- -- -- -- --) + pand xmm2, xmm7 ; xmm2=( 0 -- -- -- -- -- -- --) + + movdqa XMMWORD [wk(0)], xmm1 + movdqa XMMWORD [wk(1)], xmm2 + + POPPIC ebx + + add eax, byte SIZEOF_XMMWORD-1 + and eax, byte -SIZEOF_XMMWORD + cmp eax, byte SIZEOF_XMMWORD + ja short .columnloop + ALIGNX 16, 7 + +.columnloop_last: + ; -- process the last column block + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + pcmpeqb xmm1, xmm1 + pslldq xmm1, (SIZEOF_XMMWORD-2) + movdqa xmm2, xmm1 + + pand xmm1, XMMWORD [edx+1*SIZEOF_XMMWORD] + pand xmm2, XMMWORD [edi+1*SIZEOF_XMMWORD] + + movdqa XMMWORD [wk(2)], xmm1 ; xmm1=(-- -- -- -- -- -- -- 15) + movdqa XMMWORD [wk(3)], xmm2 ; xmm2=(-- -- -- -- -- -- -- 15) + + jmp near .upsample + ALIGNX 16, 7 + +.columnloop: + ; -- process the next column block + + movdqa xmm0, XMMWORD [ebx+1*SIZEOF_XMMWORD] ; xmm0=row[ 0][1] + movdqa xmm1, XMMWORD [ecx+1*SIZEOF_XMMWORD] ; xmm1=row[-1][1] + movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD] ; xmm2=row[+1][1] + + PUSHPIC ebx + MOVPIC ebx, POINTER [gotptr] ; load GOT address + + pxor xmm3, xmm3 ; xmm3=(all 0's) + movdqa xmm4, xmm0 + punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7) + punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15) + movdqa xmm5, xmm1 + punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15) + movdqa xmm6, xmm2 + punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15) + + pmullw xmm0, [GOTOFF(ebx,PW_THREE)] + pmullw xmm4, [GOTOFF(ebx,PW_THREE)] + + paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7) + paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15) + paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7) + paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15) + + movdqa XMMWORD [edx+2*SIZEOF_XMMWORD], xmm1 ; temporarily save + movdqa XMMWORD [edx+3*SIZEOF_XMMWORD], xmm5 ; the intermediate data + movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm6 + + pslldq xmm1, (SIZEOF_XMMWORD-2) ; xmm1=(-- -- -- -- -- -- -- 0) + pslldq xmm2, (SIZEOF_XMMWORD-2) ; xmm2=(-- -- -- -- -- -- -- 0) + + movdqa XMMWORD [wk(2)], xmm1 + movdqa XMMWORD [wk(3)], xmm2 + +.upsample: + ; -- process the upper row + + movdqa xmm7, XMMWORD [edx+0*SIZEOF_XMMWORD] + movdqa xmm3, XMMWORD [edx+1*SIZEOF_XMMWORD] + + movdqa xmm0, xmm7 ; xmm7=Int0L=( 0 1 2 3 4 5 6 7) + movdqa xmm4, xmm3 ; xmm3=Int0H=( 8 9 10 11 12 13 14 15) + psrldq xmm0, 2 ; xmm0=( 1 2 3 4 5 6 7 --) + pslldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(-- -- -- -- -- -- -- 8) + movdqa xmm5, xmm7 + movdqa xmm6, xmm3 + psrldq xmm5, (SIZEOF_XMMWORD-2) ; xmm5=( 7 -- -- -- -- -- -- --) + pslldq xmm6, 2 ; xmm6=(-- 8 9 10 11 12 13 14) + + por xmm0, xmm4 ; xmm0=( 1 2 3 4 5 6 7 8) + por xmm5, xmm6 ; xmm5=( 7 8 9 10 11 12 13 14) + + movdqa xmm1, xmm7 + movdqa xmm2, xmm3 + pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6) + psrldq xmm2, 2 ; xmm2=( 9 10 11 12 13 14 15 --) + movdqa xmm4, xmm3 + psrldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(15 -- -- -- -- -- -- --) + + por xmm1, XMMWORD [wk(0)] ; xmm1=(-1 0 1 2 3 4 5 6) + por xmm2, XMMWORD [wk(2)] ; xmm2=( 9 10 11 12 13 14 15 16) + + movdqa XMMWORD [wk(0)], xmm4 + + pmullw xmm7, [GOTOFF(ebx,PW_THREE)] + pmullw xmm3, [GOTOFF(ebx,PW_THREE)] + paddw xmm1, [GOTOFF(ebx,PW_EIGHT)] + paddw xmm5, [GOTOFF(ebx,PW_EIGHT)] + paddw xmm0, [GOTOFF(ebx,PW_SEVEN)] + paddw xmm2, [GOTOFF(ebx,PW_SEVEN)] + + paddw xmm1, xmm7 + paddw xmm5, xmm3 + psrlw xmm1, 4 ; xmm1=Out0LE=( 0 2 4 6 8 10 12 14) + psrlw xmm5, 4 ; xmm5=Out0HE=(16 18 20 22 24 26 28 30) + paddw xmm0, xmm7 + paddw xmm2, xmm3 + psrlw xmm0, 4 ; xmm0=Out0LO=( 1 3 5 7 9 11 13 15) + psrlw xmm2, 4 ; xmm2=Out0HO=(17 19 21 23 25 27 29 31) + + psllw xmm0, BYTE_BIT + psllw xmm2, BYTE_BIT + por xmm1, xmm0 ; xmm1=Out0L=( 0 1 2 ... 13 14 15) + por xmm5, xmm2 ; xmm5=Out0H=(16 17 18 ... 29 30 31) + + movdqa XMMWORD [edx+0*SIZEOF_XMMWORD], xmm1 + movdqa XMMWORD [edx+1*SIZEOF_XMMWORD], xmm5 + + ; -- process the lower row + + movdqa xmm6, XMMWORD [edi+0*SIZEOF_XMMWORD] + movdqa xmm4, XMMWORD [edi+1*SIZEOF_XMMWORD] + + movdqa xmm7, xmm6 ; xmm6=Int1L=( 0 1 2 3 4 5 6 7) + movdqa xmm3, xmm4 ; xmm4=Int1H=( 8 9 10 11 12 13 14 15) + psrldq xmm7, 2 ; xmm7=( 1 2 3 4 5 6 7 --) + pslldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(-- -- -- -- -- -- -- 8) + movdqa xmm0, xmm6 + movdqa xmm2, xmm4 + psrldq xmm0, (SIZEOF_XMMWORD-2) ; xmm0=( 7 -- -- -- -- -- -- --) + pslldq xmm2, 2 ; xmm2=(-- 8 9 10 11 12 13 14) + + por xmm7, xmm3 ; xmm7=( 1 2 3 4 5 6 7 8) + por xmm0, xmm2 ; xmm0=( 7 8 9 10 11 12 13 14) + + movdqa xmm1, xmm6 + movdqa xmm5, xmm4 + pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6) + psrldq xmm5, 2 ; xmm5=( 9 10 11 12 13 14 15 --) + movdqa xmm3, xmm4 + psrldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(15 -- -- -- -- -- -- --) + + por xmm1, XMMWORD [wk(1)] ; xmm1=(-1 0 1 2 3 4 5 6) + por xmm5, XMMWORD [wk(3)] ; xmm5=( 9 10 11 12 13 14 15 16) + + movdqa XMMWORD [wk(1)], xmm3 + + pmullw xmm6, [GOTOFF(ebx,PW_THREE)] + pmullw xmm4, [GOTOFF(ebx,PW_THREE)] + paddw xmm1, [GOTOFF(ebx,PW_EIGHT)] + paddw xmm0, [GOTOFF(ebx,PW_EIGHT)] + paddw xmm7, [GOTOFF(ebx,PW_SEVEN)] + paddw xmm5, [GOTOFF(ebx,PW_SEVEN)] + + paddw xmm1, xmm6 + paddw xmm0, xmm4 + psrlw xmm1, 4 ; xmm1=Out1LE=( 0 2 4 6 8 10 12 14) + psrlw xmm0, 4 ; xmm0=Out1HE=(16 18 20 22 24 26 28 30) + paddw xmm7, xmm6 + paddw xmm5, xmm4 + psrlw xmm7, 4 ; xmm7=Out1LO=( 1 3 5 7 9 11 13 15) + psrlw xmm5, 4 ; xmm5=Out1HO=(17 19 21 23 25 27 29 31) + + psllw xmm7, BYTE_BIT + psllw xmm5, BYTE_BIT + por xmm1, xmm7 ; xmm1=Out1L=( 0 1 2 ... 13 14 15) + por xmm0, xmm5 ; xmm0=Out1H=(16 17 18 ... 29 30 31) + + movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm1 + movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm0 + + POPPIC ebx + + sub eax, byte SIZEOF_XMMWORD + add ecx, byte 1*SIZEOF_XMMWORD ; inptr1(above) + add ebx, byte 1*SIZEOF_XMMWORD ; inptr0 + add esi, byte 1*SIZEOF_XMMWORD ; inptr1(below) + add edx, byte 2*SIZEOF_XMMWORD ; outptr0 + add edi, byte 2*SIZEOF_XMMWORD ; outptr1 + cmp eax, byte SIZEOF_XMMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop esi + pop edi + pop ecx + pop eax + + add esi, byte 1*SIZEOF_JSAMPROW ; input_data + add edi, byte 2*SIZEOF_JSAMPROW ; output_data + sub ecx, byte 2 ; rowctr + jg near .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 1:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v1_upsample_sse2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define output_width(b) (b) + 12 ; JDIMENSION output_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_upsample_sse2) + +EXTN(jsimd_h2v1_upsample_sse2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov edx, JDIMENSION [output_width(ebp)] + add edx, byte (2*SIZEOF_XMMWORD)-1 + and edx, byte -(2*SIZEOF_XMMWORD) + jz short .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz short .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov edi, JSAMPROW [edi] ; outptr + mov eax, edx ; colctr + ALIGNX 16, 7 +.columnloop: + + movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD] + + movdqa xmm1, xmm0 + punpcklbw xmm0, xmm0 + punpckhbw xmm1, xmm1 + + movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0 + movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1 + + sub eax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD] + + movdqa xmm3, xmm2 + punpcklbw xmm2, xmm2 + punpckhbw xmm3, xmm3 + + movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3 + + sub eax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + add esi, byte 2*SIZEOF_XMMWORD ; inptr + add edi, byte 4*SIZEOF_XMMWORD ; outptr + jmp short .columnloop + ALIGNX 16, 7 + +.nextrow: + pop esi + pop edi + + add esi, byte SIZEOF_JSAMPROW ; input_data + add edi, byte SIZEOF_JSAMPROW ; output_data + dec ecx ; rowctr + jg short .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved +; pop ebx ; unused + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 2:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v2_upsample_sse2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +%define max_v_samp(b) (b) + 8 ; int max_v_samp_factor +%define output_width(b) (b) + 12 ; JDIMENSION output_width +%define input_data(b) (b) + 16 ; JSAMPARRAY input_data +%define output_data_ptr(b) (b) + 20 ; JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_upsample_sse2) + +EXTN(jsimd_h2v2_upsample_sse2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov edx, JDIMENSION [output_width(ebp)] + add edx, byte (2*SIZEOF_XMMWORD)-1 + and edx, byte -(2*SIZEOF_XMMWORD) + jz near .return + + mov ecx, INT [max_v_samp(ebp)] ; rowctr + test ecx, ecx + jz near .return + + mov esi, JSAMPARRAY [input_data(ebp)] ; input_data + mov edi, POINTER [output_data_ptr(ebp)] + mov edi, JSAMPARRAY [edi] ; output_data + ALIGNX 16, 7 +.rowloop: + push edi + push esi + + mov esi, JSAMPROW [esi] ; inptr + mov ebx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; outptr0 + mov edi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; outptr1 + mov eax, edx ; colctr + ALIGNX 16, 7 +.columnloop: + + movdqa xmm0, XMMWORD [esi+0*SIZEOF_XMMWORD] + + movdqa xmm1, xmm0 + punpcklbw xmm0, xmm0 + punpckhbw xmm1, xmm1 + + movdqa XMMWORD [ebx+0*SIZEOF_XMMWORD], xmm0 + movdqa XMMWORD [ebx+1*SIZEOF_XMMWORD], xmm1 + movdqa XMMWORD [edi+0*SIZEOF_XMMWORD], xmm0 + movdqa XMMWORD [edi+1*SIZEOF_XMMWORD], xmm1 + + sub eax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + movdqa xmm2, XMMWORD [esi+1*SIZEOF_XMMWORD] + + movdqa xmm3, xmm2 + punpcklbw xmm2, xmm2 + punpckhbw xmm3, xmm3 + + movdqa XMMWORD [ebx+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [ebx+3*SIZEOF_XMMWORD], xmm3 + movdqa XMMWORD [edi+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [edi+3*SIZEOF_XMMWORD], xmm3 + + sub eax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + add esi, byte 2*SIZEOF_XMMWORD ; inptr + add ebx, byte 4*SIZEOF_XMMWORD ; outptr0 + add edi, byte 4*SIZEOF_XMMWORD ; outptr1 + jmp short .columnloop + ALIGNX 16, 7 + +.nextrow: + pop esi + pop edi + + add esi, byte 1*SIZEOF_JSAMPROW ; input_data + add edi, byte 2*SIZEOF_JSAMPROW ; output_data + sub ecx, byte 2 ; rowctr + jg short .rowloop + +.return: + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-3dn.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-3dn.asm new file mode 100644 index 0000000..0cedc6c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-3dn.asm @@ -0,0 +1,314 @@ +; +; jfdctflt.asm - floating-point FDCT (3DNow!) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a floating-point implementation of the forward DCT +; (Discrete Cosine Transform). The following code is based directly on +; the IJG's original jfdctflt.c; see the jfdctflt.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_float_3dnow) + +EXTN(jconst_fdct_float_3dnow): + +PD_0_382 times 2 dd 0.382683432365089771728460 +PD_0_707 times 2 dd 0.707106781186547524400844 +PD_0_541 times 2 dd 0.541196100146196984399723 +PD_1_306 times 2 dd 1.306562964876376527856643 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_float_3dnow(FAST_FLOAT *data) +; + +%define data(b) (b) + 8 ; FAST_FLOAT *data + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD ; mmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_float_3dnow) + +EXTN(jsimd_fdct_float_3dnow): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved +; push esi ; unused +; push edi ; unused + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process rows. + + mov edx, POINTER [data(eax)] ; (FAST_FLOAT *) + mov ecx, DCTSIZE/2 + ALIGNX 16, 7 +.rowloop: + + movq mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] + movq mm1, MMWORD [MMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] + movq mm2, MMWORD [MMBLOCK(0,3,edx,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(1,3,edx,SIZEOF_FAST_FLOAT)] + + ; mm0=(00 01), mm1=(10 11), mm2=(06 07), mm3=(16 17) + + movq mm4, mm0 ; transpose coefficients + punpckldq mm0, mm1 ; mm0=(00 10)=data0 + punpckhdq mm4, mm1 ; mm4=(01 11)=data1 + movq mm5, mm2 ; transpose coefficients + punpckldq mm2, mm3 ; mm2=(06 16)=data6 + punpckhdq mm5, mm3 ; mm5=(07 17)=data7 + + movq mm6, mm4 + movq mm7, mm0 + pfsub mm4, mm2 ; mm4=data1-data6=tmp6 + pfsub mm0, mm5 ; mm0=data0-data7=tmp7 + pfadd mm6, mm2 ; mm6=data1+data6=tmp1 + pfadd mm7, mm5 ; mm7=data0+data7=tmp0 + + movq mm1, MMWORD [MMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)] + movq mm2, MMWORD [MMBLOCK(0,2,edx,SIZEOF_FAST_FLOAT)] + movq mm5, MMWORD [MMBLOCK(1,2,edx,SIZEOF_FAST_FLOAT)] + + ; mm1=(02 03), mm3=(12 13), mm2=(04 05), mm5=(14 15) + + movq MMWORD [wk(0)], mm4 ; wk(0)=tmp6 + movq MMWORD [wk(1)], mm0 ; wk(1)=tmp7 + + movq mm4, mm1 ; transpose coefficients + punpckldq mm1, mm3 ; mm1=(02 12)=data2 + punpckhdq mm4, mm3 ; mm4=(03 13)=data3 + movq mm0, mm2 ; transpose coefficients + punpckldq mm2, mm5 ; mm2=(04 14)=data4 + punpckhdq mm0, mm5 ; mm0=(05 15)=data5 + + movq mm3, mm4 + movq mm5, mm1 + pfadd mm4, mm2 ; mm4=data3+data4=tmp3 + pfadd mm1, mm0 ; mm1=data2+data5=tmp2 + pfsub mm3, mm2 ; mm3=data3-data4=tmp4 + pfsub mm5, mm0 ; mm5=data2-data5=tmp5 + + ; -- Even part + + movq mm2, mm7 + movq mm0, mm6 + pfsub mm7, mm4 ; mm7=tmp13 + pfsub mm6, mm1 ; mm6=tmp12 + pfadd mm2, mm4 ; mm2=tmp10 + pfadd mm0, mm1 ; mm0=tmp11 + + pfadd mm6, mm7 + pfmul mm6, [GOTOFF(ebx,PD_0_707)] ; mm6=z1 + + movq mm4, mm2 + movq mm1, mm7 + pfsub mm2, mm0 ; mm2=data4 + pfsub mm7, mm6 ; mm7=data6 + pfadd mm4, mm0 ; mm4=data0 + pfadd mm1, mm6 ; mm1=data2 + + movq MMWORD [MMBLOCK(0,2,edx,SIZEOF_FAST_FLOAT)], mm2 + movq MMWORD [MMBLOCK(0,3,edx,SIZEOF_FAST_FLOAT)], mm7 + movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], mm4 + movq MMWORD [MMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)], mm1 + + ; -- Odd part + + movq mm0, MMWORD [wk(0)] ; mm0=tmp6 + movq mm6, MMWORD [wk(1)] ; mm6=tmp7 + + pfadd mm3, mm5 ; mm3=tmp10 + pfadd mm5, mm0 ; mm5=tmp11 + pfadd mm0, mm6 ; mm0=tmp12, mm6=tmp7 + + pfmul mm5, [GOTOFF(ebx,PD_0_707)] ; mm5=z3 + + movq mm2, mm3 ; mm2=tmp10 + pfsub mm3, mm0 + pfmul mm3, [GOTOFF(ebx,PD_0_382)] ; mm3=z5 + pfmul mm2, [GOTOFF(ebx,PD_0_541)] ; mm2=MULTIPLY(tmp10,FIX_0_54119610) + pfmul mm0, [GOTOFF(ebx,PD_1_306)] ; mm0=MULTIPLY(tmp12,FIX_1_30656296) + pfadd mm2, mm3 ; mm2=z2 + pfadd mm0, mm3 ; mm0=z4 + + movq mm7, mm6 + pfsub mm6, mm5 ; mm6=z13 + pfadd mm7, mm5 ; mm7=z11 + + movq mm4, mm6 + movq mm1, mm7 + pfsub mm6, mm2 ; mm6=data3 + pfsub mm7, mm0 ; mm7=data7 + pfadd mm4, mm2 ; mm4=data5 + pfadd mm1, mm0 ; mm1=data1 + + movq MMWORD [MMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)], mm6 + movq MMWORD [MMBLOCK(1,3,edx,SIZEOF_FAST_FLOAT)], mm7 + movq MMWORD [MMBLOCK(1,2,edx,SIZEOF_FAST_FLOAT)], mm4 + movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], mm1 + + add edx, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT + dec ecx + jnz near .rowloop + + ; ---- Pass 2: process columns. + + mov edx, POINTER [data(eax)] ; (FAST_FLOAT *) + mov ecx, DCTSIZE/2 + ALIGNX 16, 7 +.columnloop: + + movq mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] + movq mm1, MMWORD [MMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] + movq mm2, MMWORD [MMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)] + + ; mm0=(00 10), mm1=(01 11), mm2=(60 70), mm3=(61 71) + + movq mm4, mm0 ; transpose coefficients + punpckldq mm0, mm1 ; mm0=(00 01)=data0 + punpckhdq mm4, mm1 ; mm4=(10 11)=data1 + movq mm5, mm2 ; transpose coefficients + punpckldq mm2, mm3 ; mm2=(60 61)=data6 + punpckhdq mm5, mm3 ; mm5=(70 71)=data7 + + movq mm6, mm4 + movq mm7, mm0 + pfsub mm4, mm2 ; mm4=data1-data6=tmp6 + pfsub mm0, mm5 ; mm0=data0-data7=tmp7 + pfadd mm6, mm2 ; mm6=data1+data6=tmp1 + pfadd mm7, mm5 ; mm7=data0+data7=tmp0 + + movq mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)] + movq mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)] + movq mm5, MMWORD [MMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)] + + ; mm1=(20 30), mm3=(21 31), mm2=(40 50), mm5=(41 51) + + movq MMWORD [wk(0)], mm4 ; wk(0)=tmp6 + movq MMWORD [wk(1)], mm0 ; wk(1)=tmp7 + + movq mm4, mm1 ; transpose coefficients + punpckldq mm1, mm3 ; mm1=(20 21)=data2 + punpckhdq mm4, mm3 ; mm4=(30 31)=data3 + movq mm0, mm2 ; transpose coefficients + punpckldq mm2, mm5 ; mm2=(40 41)=data4 + punpckhdq mm0, mm5 ; mm0=(50 51)=data5 + + movq mm3, mm4 + movq mm5, mm1 + pfadd mm4, mm2 ; mm4=data3+data4=tmp3 + pfadd mm1, mm0 ; mm1=data2+data5=tmp2 + pfsub mm3, mm2 ; mm3=data3-data4=tmp4 + pfsub mm5, mm0 ; mm5=data2-data5=tmp5 + + ; -- Even part + + movq mm2, mm7 + movq mm0, mm6 + pfsub mm7, mm4 ; mm7=tmp13 + pfsub mm6, mm1 ; mm6=tmp12 + pfadd mm2, mm4 ; mm2=tmp10 + pfadd mm0, mm1 ; mm0=tmp11 + + pfadd mm6, mm7 + pfmul mm6, [GOTOFF(ebx,PD_0_707)] ; mm6=z1 + + movq mm4, mm2 + movq mm1, mm7 + pfsub mm2, mm0 ; mm2=data4 + pfsub mm7, mm6 ; mm7=data6 + pfadd mm4, mm0 ; mm4=data0 + pfadd mm1, mm6 ; mm1=data2 + + movq MMWORD [MMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)], mm2 + movq MMWORD [MMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)], mm7 + movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], mm4 + movq MMWORD [MMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], mm1 + + ; -- Odd part + + movq mm0, MMWORD [wk(0)] ; mm0=tmp6 + movq mm6, MMWORD [wk(1)] ; mm6=tmp7 + + pfadd mm3, mm5 ; mm3=tmp10 + pfadd mm5, mm0 ; mm5=tmp11 + pfadd mm0, mm6 ; mm0=tmp12, mm6=tmp7 + + pfmul mm5, [GOTOFF(ebx,PD_0_707)] ; mm5=z3 + + movq mm2, mm3 ; mm2=tmp10 + pfsub mm3, mm0 + pfmul mm3, [GOTOFF(ebx,PD_0_382)] ; mm3=z5 + pfmul mm2, [GOTOFF(ebx,PD_0_541)] ; mm2=MULTIPLY(tmp10,FIX_0_54119610) + pfmul mm0, [GOTOFF(ebx,PD_1_306)] ; mm0=MULTIPLY(tmp12,FIX_1_30656296) + pfadd mm2, mm3 ; mm2=z2 + pfadd mm0, mm3 ; mm0=z4 + + movq mm7, mm6 + pfsub mm6, mm5 ; mm6=z13 + pfadd mm7, mm5 ; mm7=z11 + + movq mm4, mm6 + movq mm1, mm7 + pfsub mm6, mm2 ; mm6=data3 + pfsub mm7, mm0 ; mm7=data7 + pfadd mm4, mm2 ; mm4=data5 + pfadd mm1, mm0 ; mm1=data1 + + movq MMWORD [MMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], mm6 + movq MMWORD [MMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)], mm7 + movq MMWORD [MMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)], mm4 + movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], mm1 + + add edx, byte 2*SIZEOF_FAST_FLOAT + dec ecx + jnz near .columnloop + + femms ; empty MMX/3DNow! state + +; pop edi ; unused +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-sse.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-sse.asm new file mode 100644 index 0000000..2cb9533 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctflt-sse.asm @@ -0,0 +1,365 @@ +; +; jfdctflt.asm - floating-point FDCT (SSE) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a floating-point implementation of the forward DCT +; (Discrete Cosine Transform). The following code is based directly on +; the IJG's original jfdctflt.c; see the jfdctflt.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5) + shufps %1, %2, 0x44 +%endmacro + +%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7) + shufps %1, %2, 0xEE +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_float_sse) + +EXTN(jconst_fdct_float_sse): + +PD_0_382 times 4 dd 0.382683432365089771728460 +PD_0_707 times 4 dd 0.707106781186547524400844 +PD_0_541 times 4 dd 0.541196100146196984399723 +PD_1_306 times 4 dd 1.306562964876376527856643 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_float_sse(FAST_FLOAT *data) +; + +%define data(b) (b) + 8 ; FAST_FLOAT *data + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_float_sse) + +EXTN(jsimd_fdct_float_sse): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved +; push esi ; unused +; push edi ; unused + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process rows. + + mov edx, POINTER [data(eax)] ; (FAST_FLOAT *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.rowloop: + + movaps xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)] + + ; xmm0=(20 21 22 23), xmm2=(24 25 26 27) + ; xmm1=(30 31 32 33), xmm3=(34 35 36 37) + + movaps xmm4, xmm0 ; transpose coefficients(phase 1) + unpcklps xmm0, xmm1 ; xmm0=(20 30 21 31) + unpckhps xmm4, xmm1 ; xmm4=(22 32 23 33) + movaps xmm5, xmm2 ; transpose coefficients(phase 1) + unpcklps xmm2, xmm3 ; xmm2=(24 34 25 35) + unpckhps xmm5, xmm3 ; xmm5=(26 36 27 37) + + movaps xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)] + + ; xmm6=(00 01 02 03), xmm1=(04 05 06 07) + ; xmm7=(10 11 12 13), xmm3=(14 15 16 17) + + movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 32 23 33) + movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(24 34 25 35) + + movaps xmm4, xmm6 ; transpose coefficients(phase 1) + unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11) + unpckhps xmm4, xmm7 ; xmm4=(02 12 03 13) + movaps xmm2, xmm1 ; transpose coefficients(phase 1) + unpcklps xmm1, xmm3 ; xmm1=(04 14 05 15) + unpckhps xmm2, xmm3 ; xmm2=(06 16 07 17) + + movaps xmm7, xmm6 ; transpose coefficients(phase 2) + unpcklps2 xmm6, xmm0 ; xmm6=(00 10 20 30)=data0 + unpckhps2 xmm7, xmm0 ; xmm7=(01 11 21 31)=data1 + movaps xmm3, xmm2 ; transpose coefficients(phase 2) + unpcklps2 xmm2, xmm5 ; xmm2=(06 16 26 36)=data6 + unpckhps2 xmm3, xmm5 ; xmm3=(07 17 27 37)=data7 + + movaps xmm0, xmm7 + movaps xmm5, xmm6 + subps xmm7, xmm2 ; xmm7=data1-data6=tmp6 + subps xmm6, xmm3 ; xmm6=data0-data7=tmp7 + addps xmm0, xmm2 ; xmm0=data1+data6=tmp1 + addps xmm5, xmm3 ; xmm5=data0+data7=tmp0 + + movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 32 23 33) + movaps xmm3, XMMWORD [wk(1)] ; xmm3=(24 34 25 35) + movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6 + movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 + + movaps xmm7, xmm4 ; transpose coefficients(phase 2) + unpcklps2 xmm4, xmm2 ; xmm4=(02 12 22 32)=data2 + unpckhps2 xmm7, xmm2 ; xmm7=(03 13 23 33)=data3 + movaps xmm6, xmm1 ; transpose coefficients(phase 2) + unpcklps2 xmm1, xmm3 ; xmm1=(04 14 24 34)=data4 + unpckhps2 xmm6, xmm3 ; xmm6=(05 15 25 35)=data5 + + movaps xmm2, xmm7 + movaps xmm3, xmm4 + addps xmm7, xmm1 ; xmm7=data3+data4=tmp3 + addps xmm4, xmm6 ; xmm4=data2+data5=tmp2 + subps xmm2, xmm1 ; xmm2=data3-data4=tmp4 + subps xmm3, xmm6 ; xmm3=data2-data5=tmp5 + + ; -- Even part + + movaps xmm1, xmm5 + movaps xmm6, xmm0 + subps xmm5, xmm7 ; xmm5=tmp13 + subps xmm0, xmm4 ; xmm0=tmp12 + addps xmm1, xmm7 ; xmm1=tmp10 + addps xmm6, xmm4 ; xmm6=tmp11 + + addps xmm0, xmm5 + mulps xmm0, [GOTOFF(ebx,PD_0_707)] ; xmm0=z1 + + movaps xmm7, xmm1 + movaps xmm4, xmm5 + subps xmm1, xmm6 ; xmm1=data4 + subps xmm5, xmm0 ; xmm5=data6 + addps xmm7, xmm6 ; xmm7=data0 + addps xmm4, xmm0 ; xmm4=data2 + + movaps XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(2,1,edx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4 + + ; -- Odd part + + movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6 + movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7 + + addps xmm2, xmm3 ; xmm2=tmp10 + addps xmm3, xmm6 ; xmm3=tmp11 + addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7 + + mulps xmm3, [GOTOFF(ebx,PD_0_707)] ; xmm3=z3 + + movaps xmm1, xmm2 ; xmm1=tmp10 + subps xmm2, xmm6 + mulps xmm2, [GOTOFF(ebx,PD_0_382)] ; xmm2=z5 + mulps xmm1, [GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196) + mulps xmm6, [GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562) + addps xmm1, xmm2 ; xmm1=z2 + addps xmm6, xmm2 ; xmm6=z4 + + movaps xmm5, xmm0 + subps xmm0, xmm3 ; xmm0=z13 + addps xmm5, xmm3 ; xmm5=z11 + + movaps xmm7, xmm0 + movaps xmm4, xmm5 + subps xmm0, xmm1 ; xmm0=data3 + subps xmm5, xmm6 ; xmm5=data7 + addps xmm7, xmm1 ; xmm7=data5 + addps xmm4, xmm6 ; xmm4=data1 + + movaps XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(3,1,edx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4 + + add edx, 4*DCTSIZE*SIZEOF_FAST_FLOAT + dec ecx + jnz near .rowloop + + ; ---- Pass 2: process columns. + + mov edx, POINTER [data(eax)] ; (FAST_FLOAT *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.columnloop: + + movaps xmm0, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)] + + ; xmm0=(02 12 22 32), xmm2=(42 52 62 72) + ; xmm1=(03 13 23 33), xmm3=(43 53 63 73) + + movaps xmm4, xmm0 ; transpose coefficients(phase 1) + unpcklps xmm0, xmm1 ; xmm0=(02 03 12 13) + unpckhps xmm4, xmm1 ; xmm4=(22 23 32 33) + movaps xmm5, xmm2 ; transpose coefficients(phase 1) + unpcklps xmm2, xmm3 ; xmm2=(42 43 52 53) + unpckhps xmm5, xmm3 ; xmm5=(62 63 72 73) + + movaps xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm7, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)] + + ; xmm6=(00 10 20 30), xmm1=(40 50 60 70) + ; xmm7=(01 11 21 31), xmm3=(41 51 61 71) + + movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 23 32 33) + movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(42 43 52 53) + + movaps xmm4, xmm6 ; transpose coefficients(phase 1) + unpcklps xmm6, xmm7 ; xmm6=(00 01 10 11) + unpckhps xmm4, xmm7 ; xmm4=(20 21 30 31) + movaps xmm2, xmm1 ; transpose coefficients(phase 1) + unpcklps xmm1, xmm3 ; xmm1=(40 41 50 51) + unpckhps xmm2, xmm3 ; xmm2=(60 61 70 71) + + movaps xmm7, xmm6 ; transpose coefficients(phase 2) + unpcklps2 xmm6, xmm0 ; xmm6=(00 01 02 03)=data0 + unpckhps2 xmm7, xmm0 ; xmm7=(10 11 12 13)=data1 + movaps xmm3, xmm2 ; transpose coefficients(phase 2) + unpcklps2 xmm2, xmm5 ; xmm2=(60 61 62 63)=data6 + unpckhps2 xmm3, xmm5 ; xmm3=(70 71 72 73)=data7 + + movaps xmm0, xmm7 + movaps xmm5, xmm6 + subps xmm7, xmm2 ; xmm7=data1-data6=tmp6 + subps xmm6, xmm3 ; xmm6=data0-data7=tmp7 + addps xmm0, xmm2 ; xmm0=data1+data6=tmp1 + addps xmm5, xmm3 ; xmm5=data0+data7=tmp0 + + movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 23 32 33) + movaps xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53) + movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6 + movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 + + movaps xmm7, xmm4 ; transpose coefficients(phase 2) + unpcklps2 xmm4, xmm2 ; xmm4=(20 21 22 23)=data2 + unpckhps2 xmm7, xmm2 ; xmm7=(30 31 32 33)=data3 + movaps xmm6, xmm1 ; transpose coefficients(phase 2) + unpcklps2 xmm1, xmm3 ; xmm1=(40 41 42 43)=data4 + unpckhps2 xmm6, xmm3 ; xmm6=(50 51 52 53)=data5 + + movaps xmm2, xmm7 + movaps xmm3, xmm4 + addps xmm7, xmm1 ; xmm7=data3+data4=tmp3 + addps xmm4, xmm6 ; xmm4=data2+data5=tmp2 + subps xmm2, xmm1 ; xmm2=data3-data4=tmp4 + subps xmm3, xmm6 ; xmm3=data2-data5=tmp5 + + ; -- Even part + + movaps xmm1, xmm5 + movaps xmm6, xmm0 + subps xmm5, xmm7 ; xmm5=tmp13 + subps xmm0, xmm4 ; xmm0=tmp12 + addps xmm1, xmm7 ; xmm1=tmp10 + addps xmm6, xmm4 ; xmm6=tmp11 + + addps xmm0, xmm5 + mulps xmm0, [GOTOFF(ebx,PD_0_707)] ; xmm0=z1 + + movaps xmm7, xmm1 + movaps xmm4, xmm5 + subps xmm1, xmm6 ; xmm1=data4 + subps xmm5, xmm0 ; xmm5=data6 + addps xmm7, xmm6 ; xmm7=data0 + addps xmm4, xmm0 ; xmm4=data2 + + movaps XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FAST_FLOAT)], xmm4 + + ; -- Odd part + + movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6 + movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7 + + addps xmm2, xmm3 ; xmm2=tmp10 + addps xmm3, xmm6 ; xmm3=tmp11 + addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7 + + mulps xmm3, [GOTOFF(ebx,PD_0_707)] ; xmm3=z3 + + movaps xmm1, xmm2 ; xmm1=tmp10 + subps xmm2, xmm6 + mulps xmm2, [GOTOFF(ebx,PD_0_382)] ; xmm2=z5 + mulps xmm1, [GOTOFF(ebx,PD_0_541)] ; xmm1=MULTIPLY(tmp10,FIX_0_541196) + mulps xmm6, [GOTOFF(ebx,PD_1_306)] ; xmm6=MULTIPLY(tmp12,FIX_1_306562) + addps xmm1, xmm2 ; xmm1=z2 + addps xmm6, xmm2 ; xmm6=z4 + + movaps xmm5, xmm0 + subps xmm0, xmm3 ; xmm0=z13 + addps xmm5, xmm3 ; xmm5=z11 + + movaps xmm7, xmm0 + movaps xmm4, xmm5 + subps xmm0, xmm1 ; xmm0=data3 + subps xmm5, xmm6 ; xmm5=data7 + addps xmm7, xmm1 ; xmm7=data5 + addps xmm4, xmm6 ; xmm4=data1 + + movaps XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)], xmm4 + + add edx, byte 4*SIZEOF_FAST_FLOAT + dec ecx + jnz near .columnloop + +; pop edi ; unused +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-mmx.asm new file mode 100644 index 0000000..fe16e83 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-mmx.asm @@ -0,0 +1,391 @@ +; +; jfdctfst.asm - fast integer FDCT (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a fast, not so accurate integer implementation of +; the forward DCT (Discrete Cosine Transform). The following code is +; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c +; for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 8 ; 14 is also OK. + +%if CONST_BITS == 8 +F_0_382 equ 98 ; FIX(0.382683433) +F_0_541 equ 139 ; FIX(0.541196100) +F_0_707 equ 181 ; FIX(0.707106781) +F_1_306 equ 334 ; FIX(1.306562965) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_382 equ DESCALE( 410903207, 30 - CONST_BITS) ; FIX(0.382683433) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_707 equ DESCALE( 759250124, 30 - CONST_BITS) ; FIX(0.707106781) +F_1_306 equ DESCALE(1402911301, 30 - CONST_BITS) ; FIX(1.306562965) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + +; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow) +; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw) + +%define PRE_MULTIPLY_SCALE_BITS 2 +%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS) + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_ifast_mmx) + +EXTN(jconst_fdct_ifast_mmx): + +PW_F0707 times 4 dw F_0_707 << CONST_SHIFT +PW_F0382 times 4 dw F_0_382 << CONST_SHIFT +PW_F0541 times 4 dw F_0_541 << CONST_SHIFT +PW_F1306 times 4 dw F_1_306 << CONST_SHIFT + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_ifast_mmx(DCTELEM *data) +; + +%define data(b) (b) + 8 ; DCTELEM *data + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD ; mmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_ifast_mmx) + +EXTN(jsimd_fdct_ifast_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved +; push esi ; unused +; push edi ; unused + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process rows. + + mov edx, POINTER [data(eax)] ; (DCTELEM *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.rowloop: + + movq mm0, MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)] + movq mm2, MMWORD [MMBLOCK(2,1,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(3,1,edx,SIZEOF_DCTELEM)] + + ; mm0=(20 21 22 23), mm2=(24 25 26 27) + ; mm1=(30 31 32 33), mm3=(34 35 36 37) + + movq mm4, mm0 ; transpose coefficients(phase 1) + punpcklwd mm0, mm1 ; mm0=(20 30 21 31) + punpckhwd mm4, mm1 ; mm4=(22 32 23 33) + movq mm5, mm2 ; transpose coefficients(phase 1) + punpcklwd mm2, mm3 ; mm2=(24 34 25 35) + punpckhwd mm5, mm3 ; mm5=(26 36 27 37) + + movq mm6, MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)] + movq mm7, MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(0,1,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(1,1,edx,SIZEOF_DCTELEM)] + + ; mm6=(00 01 02 03), mm1=(04 05 06 07) + ; mm7=(10 11 12 13), mm3=(14 15 16 17) + + movq MMWORD [wk(0)], mm4 ; wk(0)=(22 32 23 33) + movq MMWORD [wk(1)], mm2 ; wk(1)=(24 34 25 35) + + movq mm4, mm6 ; transpose coefficients(phase 1) + punpcklwd mm6, mm7 ; mm6=(00 10 01 11) + punpckhwd mm4, mm7 ; mm4=(02 12 03 13) + movq mm2, mm1 ; transpose coefficients(phase 1) + punpcklwd mm1, mm3 ; mm1=(04 14 05 15) + punpckhwd mm2, mm3 ; mm2=(06 16 07 17) + + movq mm7, mm6 ; transpose coefficients(phase 2) + punpckldq mm6, mm0 ; mm6=(00 10 20 30)=data0 + punpckhdq mm7, mm0 ; mm7=(01 11 21 31)=data1 + movq mm3, mm2 ; transpose coefficients(phase 2) + punpckldq mm2, mm5 ; mm2=(06 16 26 36)=data6 + punpckhdq mm3, mm5 ; mm3=(07 17 27 37)=data7 + + movq mm0, mm7 + movq mm5, mm6 + psubw mm7, mm2 ; mm7=data1-data6=tmp6 + psubw mm6, mm3 ; mm6=data0-data7=tmp7 + paddw mm0, mm2 ; mm0=data1+data6=tmp1 + paddw mm5, mm3 ; mm5=data0+data7=tmp0 + + movq mm2, MMWORD [wk(0)] ; mm2=(22 32 23 33) + movq mm3, MMWORD [wk(1)] ; mm3=(24 34 25 35) + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp6 + movq MMWORD [wk(1)], mm6 ; wk(1)=tmp7 + + movq mm7, mm4 ; transpose coefficients(phase 2) + punpckldq mm4, mm2 ; mm4=(02 12 22 32)=data2 + punpckhdq mm7, mm2 ; mm7=(03 13 23 33)=data3 + movq mm6, mm1 ; transpose coefficients(phase 2) + punpckldq mm1, mm3 ; mm1=(04 14 24 34)=data4 + punpckhdq mm6, mm3 ; mm6=(05 15 25 35)=data5 + + movq mm2, mm7 + movq mm3, mm4 + paddw mm7, mm1 ; mm7=data3+data4=tmp3 + paddw mm4, mm6 ; mm4=data2+data5=tmp2 + psubw mm2, mm1 ; mm2=data3-data4=tmp4 + psubw mm3, mm6 ; mm3=data2-data5=tmp5 + + ; -- Even part + + movq mm1, mm5 + movq mm6, mm0 + psubw mm5, mm7 ; mm5=tmp13 + psubw mm0, mm4 ; mm0=tmp12 + paddw mm1, mm7 ; mm1=tmp10 + paddw mm6, mm4 ; mm6=tmp11 + + paddw mm0, mm5 + psllw mm0, PRE_MULTIPLY_SCALE_BITS + pmulhw mm0, [GOTOFF(ebx,PW_F0707)] ; mm0=z1 + + movq mm7, mm1 + movq mm4, mm5 + psubw mm1, mm6 ; mm1=data4 + psubw mm5, mm0 ; mm5=data6 + paddw mm7, mm6 ; mm7=data0 + paddw mm4, mm0 ; mm4=data2 + + movq MMWORD [MMBLOCK(0,1,edx,SIZEOF_DCTELEM)], mm1 + movq MMWORD [MMBLOCK(2,1,edx,SIZEOF_DCTELEM)], mm5 + movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)], mm7 + movq MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)], mm4 + + ; -- Odd part + + movq mm6, MMWORD [wk(0)] ; mm6=tmp6 + movq mm0, MMWORD [wk(1)] ; mm0=tmp7 + + paddw mm2, mm3 ; mm2=tmp10 + paddw mm3, mm6 ; mm3=tmp11 + paddw mm6, mm0 ; mm6=tmp12, mm0=tmp7 + + psllw mm2, PRE_MULTIPLY_SCALE_BITS + psllw mm6, PRE_MULTIPLY_SCALE_BITS + + psllw mm3, PRE_MULTIPLY_SCALE_BITS + pmulhw mm3, [GOTOFF(ebx,PW_F0707)] ; mm3=z3 + + movq mm1, mm2 ; mm1=tmp10 + psubw mm2, mm6 + pmulhw mm2, [GOTOFF(ebx,PW_F0382)] ; mm2=z5 + pmulhw mm1, [GOTOFF(ebx,PW_F0541)] ; mm1=MULTIPLY(tmp10,FIX_0_54119610) + pmulhw mm6, [GOTOFF(ebx,PW_F1306)] ; mm6=MULTIPLY(tmp12,FIX_1_30656296) + paddw mm1, mm2 ; mm1=z2 + paddw mm6, mm2 ; mm6=z4 + + movq mm5, mm0 + psubw mm0, mm3 ; mm0=z13 + paddw mm5, mm3 ; mm5=z11 + + movq mm7, mm0 + movq mm4, mm5 + psubw mm0, mm1 ; mm0=data3 + psubw mm5, mm6 ; mm5=data7 + paddw mm7, mm1 ; mm7=data5 + paddw mm4, mm6 ; mm4=data1 + + movq MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)], mm0 + movq MMWORD [MMBLOCK(3,1,edx,SIZEOF_DCTELEM)], mm5 + movq MMWORD [MMBLOCK(1,1,edx,SIZEOF_DCTELEM)], mm7 + movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)], mm4 + + add edx, byte 4*DCTSIZE*SIZEOF_DCTELEM + dec ecx + jnz near .rowloop + + ; ---- Pass 2: process columns. + + mov edx, POINTER [data(eax)] ; (DCTELEM *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.columnloop: + + movq mm0, MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)] + movq mm2, MMWORD [MMBLOCK(6,0,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_DCTELEM)] + + ; mm0=(02 12 22 32), mm2=(42 52 62 72) + ; mm1=(03 13 23 33), mm3=(43 53 63 73) + + movq mm4, mm0 ; transpose coefficients(phase 1) + punpcklwd mm0, mm1 ; mm0=(02 03 12 13) + punpckhwd mm4, mm1 ; mm4=(22 23 32 33) + movq mm5, mm2 ; transpose coefficients(phase 1) + punpcklwd mm2, mm3 ; mm2=(42 43 52 53) + punpckhwd mm5, mm3 ; mm5=(62 63 72 73) + + movq mm6, MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)] + movq mm7, MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(4,0,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(5,0,edx,SIZEOF_DCTELEM)] + + ; mm6=(00 10 20 30), mm1=(40 50 60 70) + ; mm7=(01 11 21 31), mm3=(41 51 61 71) + + movq MMWORD [wk(0)], mm4 ; wk(0)=(22 23 32 33) + movq MMWORD [wk(1)], mm2 ; wk(1)=(42 43 52 53) + + movq mm4, mm6 ; transpose coefficients(phase 1) + punpcklwd mm6, mm7 ; mm6=(00 01 10 11) + punpckhwd mm4, mm7 ; mm4=(20 21 30 31) + movq mm2, mm1 ; transpose coefficients(phase 1) + punpcklwd mm1, mm3 ; mm1=(40 41 50 51) + punpckhwd mm2, mm3 ; mm2=(60 61 70 71) + + movq mm7, mm6 ; transpose coefficients(phase 2) + punpckldq mm6, mm0 ; mm6=(00 01 02 03)=data0 + punpckhdq mm7, mm0 ; mm7=(10 11 12 13)=data1 + movq mm3, mm2 ; transpose coefficients(phase 2) + punpckldq mm2, mm5 ; mm2=(60 61 62 63)=data6 + punpckhdq mm3, mm5 ; mm3=(70 71 72 73)=data7 + + movq mm0, mm7 + movq mm5, mm6 + psubw mm7, mm2 ; mm7=data1-data6=tmp6 + psubw mm6, mm3 ; mm6=data0-data7=tmp7 + paddw mm0, mm2 ; mm0=data1+data6=tmp1 + paddw mm5, mm3 ; mm5=data0+data7=tmp0 + + movq mm2, MMWORD [wk(0)] ; mm2=(22 23 32 33) + movq mm3, MMWORD [wk(1)] ; mm3=(42 43 52 53) + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp6 + movq MMWORD [wk(1)], mm6 ; wk(1)=tmp7 + + movq mm7, mm4 ; transpose coefficients(phase 2) + punpckldq mm4, mm2 ; mm4=(20 21 22 23)=data2 + punpckhdq mm7, mm2 ; mm7=(30 31 32 33)=data3 + movq mm6, mm1 ; transpose coefficients(phase 2) + punpckldq mm1, mm3 ; mm1=(40 41 42 43)=data4 + punpckhdq mm6, mm3 ; mm6=(50 51 52 53)=data5 + + movq mm2, mm7 + movq mm3, mm4 + paddw mm7, mm1 ; mm7=data3+data4=tmp3 + paddw mm4, mm6 ; mm4=data2+data5=tmp2 + psubw mm2, mm1 ; mm2=data3-data4=tmp4 + psubw mm3, mm6 ; mm3=data2-data5=tmp5 + + ; -- Even part + + movq mm1, mm5 + movq mm6, mm0 + psubw mm5, mm7 ; mm5=tmp13 + psubw mm0, mm4 ; mm0=tmp12 + paddw mm1, mm7 ; mm1=tmp10 + paddw mm6, mm4 ; mm6=tmp11 + + paddw mm0, mm5 + psllw mm0, PRE_MULTIPLY_SCALE_BITS + pmulhw mm0, [GOTOFF(ebx,PW_F0707)] ; mm0=z1 + + movq mm7, mm1 + movq mm4, mm5 + psubw mm1, mm6 ; mm1=data4 + psubw mm5, mm0 ; mm5=data6 + paddw mm7, mm6 ; mm7=data0 + paddw mm4, mm0 ; mm4=data2 + + movq MMWORD [MMBLOCK(4,0,edx,SIZEOF_DCTELEM)], mm1 + movq MMWORD [MMBLOCK(6,0,edx,SIZEOF_DCTELEM)], mm5 + movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)], mm7 + movq MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)], mm4 + + ; -- Odd part + + movq mm6, MMWORD [wk(0)] ; mm6=tmp6 + movq mm0, MMWORD [wk(1)] ; mm0=tmp7 + + paddw mm2, mm3 ; mm2=tmp10 + paddw mm3, mm6 ; mm3=tmp11 + paddw mm6, mm0 ; mm6=tmp12, mm0=tmp7 + + psllw mm2, PRE_MULTIPLY_SCALE_BITS + psllw mm6, PRE_MULTIPLY_SCALE_BITS + + psllw mm3, PRE_MULTIPLY_SCALE_BITS + pmulhw mm3, [GOTOFF(ebx,PW_F0707)] ; mm3=z3 + + movq mm1, mm2 ; mm1=tmp10 + psubw mm2, mm6 + pmulhw mm2, [GOTOFF(ebx,PW_F0382)] ; mm2=z5 + pmulhw mm1, [GOTOFF(ebx,PW_F0541)] ; mm1=MULTIPLY(tmp10,FIX_0_54119610) + pmulhw mm6, [GOTOFF(ebx,PW_F1306)] ; mm6=MULTIPLY(tmp12,FIX_1_30656296) + paddw mm1, mm2 ; mm1=z2 + paddw mm6, mm2 ; mm6=z4 + + movq mm5, mm0 + psubw mm0, mm3 ; mm0=z13 + paddw mm5, mm3 ; mm5=z11 + + movq mm7, mm0 + movq mm4, mm5 + psubw mm0, mm1 ; mm0=data3 + psubw mm5, mm6 ; mm5=data7 + paddw mm7, mm1 ; mm7=data5 + paddw mm4, mm6 ; mm4=data1 + + movq MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)], mm0 + movq MMWORD [MMBLOCK(7,0,edx,SIZEOF_DCTELEM)], mm5 + movq MMWORD [MMBLOCK(5,0,edx,SIZEOF_DCTELEM)], mm7 + movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)], mm4 + + add edx, byte 4*SIZEOF_DCTELEM + dec ecx + jnz near .columnloop + + emms ; empty MMX state + +; pop edi ; unused +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-sse2.asm new file mode 100644 index 0000000..890482e --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctfst-sse2.asm @@ -0,0 +1,399 @@ +; +; jfdctfst.asm - fast integer FDCT (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a fast, not so accurate integer implementation of +; the forward DCT (Discrete Cosine Transform). The following code is +; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c +; for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 8 ; 14 is also OK. + +%if CONST_BITS == 8 +F_0_382 equ 98 ; FIX(0.382683433) +F_0_541 equ 139 ; FIX(0.541196100) +F_0_707 equ 181 ; FIX(0.707106781) +F_1_306 equ 334 ; FIX(1.306562965) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_382 equ DESCALE( 410903207, 30 - CONST_BITS) ; FIX(0.382683433) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_707 equ DESCALE( 759250124, 30 - CONST_BITS) ; FIX(0.707106781) +F_1_306 equ DESCALE(1402911301, 30 - CONST_BITS) ; FIX(1.306562965) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + +; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow) +; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw) + +%define PRE_MULTIPLY_SCALE_BITS 2 +%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS) + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_ifast_sse2) + +EXTN(jconst_fdct_ifast_sse2): + +PW_F0707 times 8 dw F_0_707 << CONST_SHIFT +PW_F0382 times 8 dw F_0_382 << CONST_SHIFT +PW_F0541 times 8 dw F_0_541 << CONST_SHIFT +PW_F1306 times 8 dw F_1_306 << CONST_SHIFT + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_ifast_sse2(DCTELEM *data) +; + +%define data(b) (b) + 8 ; DCTELEM *data + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_ifast_sse2) + +EXTN(jsimd_fdct_ifast_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; unused +; push edx ; need not be preserved +; push esi ; unused +; push edi ; unused + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process rows. + + mov edx, POINTER [data(eax)] ; (DCTELEM *) + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)] + movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)] + + ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27) + ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37) + + movdqa xmm4, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13) + punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17) + movdqa xmm5, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33) + punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37) + + movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)] + movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)] + + ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62) + ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63) + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33) + movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37) + + movdqa xmm2, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53) + punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57) + movdqa xmm5, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73) + punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77) + + movdqa xmm7, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71) + punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73) + movdqa xmm3, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75) + punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77) + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33) + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37) + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(42 52 62 72 43 53 63 73) + movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=(44 54 64 74 45 55 65 75) + + movdqa xmm7, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31) + punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33) + movdqa xmm2, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35) + punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37) + + movdqa xmm1, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0 + punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1 + movdqa xmm5, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6 + punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7 + + movdqa xmm6, xmm1 + movdqa xmm3, xmm0 + psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6 + psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7 + paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1 + paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0 + + movdqa xmm2, XMMWORD [wk(0)] ; xmm2=(42 52 62 72 43 53 63 73) + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(44 54 64 74 45 55 65 75) + movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7 + + movdqa xmm1, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2 + punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3 + movdqa xmm0, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4 + punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5 + + movdqa xmm2, xmm1 + movdqa xmm5, xmm7 + paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3 + paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2 + psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4 + psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm4, xmm3 + movdqa xmm0, xmm6 + psubw xmm3, xmm1 ; xmm3=tmp13 + psubw xmm6, xmm7 ; xmm6=tmp12 + paddw xmm4, xmm1 ; xmm4=tmp10 + paddw xmm0, xmm7 ; xmm0=tmp11 + + paddw xmm6, xmm3 + psllw xmm6, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm6, [GOTOFF(ebx,PW_F0707)] ; xmm6=z1 + + movdqa xmm1, xmm4 + movdqa xmm7, xmm3 + psubw xmm4, xmm0 ; xmm4=data4 + psubw xmm3, xmm6 ; xmm3=data6 + paddw xmm1, xmm0 ; xmm1=data0 + paddw xmm7, xmm6 ; xmm7=data2 + + movdqa xmm0, XMMWORD [wk(0)] ; xmm0=tmp6 + movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp7 + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=data4 + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=data6 + + ; -- Odd part + + paddw xmm2, xmm5 ; xmm2=tmp10 + paddw xmm5, xmm0 ; xmm5=tmp11 + paddw xmm0, xmm6 ; xmm0=tmp12, xmm6=tmp7 + + psllw xmm2, PRE_MULTIPLY_SCALE_BITS + psllw xmm0, PRE_MULTIPLY_SCALE_BITS + + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm5, [GOTOFF(ebx,PW_F0707)] ; xmm5=z3 + + movdqa xmm4, xmm2 ; xmm4=tmp10 + psubw xmm2, xmm0 + pmulhw xmm2, [GOTOFF(ebx,PW_F0382)] ; xmm2=z5 + pmulhw xmm4, [GOTOFF(ebx,PW_F0541)] ; xmm4=MULTIPLY(tmp10,FIX_0_541196) + pmulhw xmm0, [GOTOFF(ebx,PW_F1306)] ; xmm0=MULTIPLY(tmp12,FIX_1_306562) + paddw xmm4, xmm2 ; xmm4=z2 + paddw xmm0, xmm2 ; xmm0=z4 + + movdqa xmm3, xmm6 + psubw xmm6, xmm5 ; xmm6=z13 + paddw xmm3, xmm5 ; xmm3=z11 + + movdqa xmm2, xmm6 + movdqa xmm5, xmm3 + psubw xmm6, xmm4 ; xmm6=data3 + psubw xmm3, xmm0 ; xmm3=data7 + paddw xmm2, xmm4 ; xmm2=data5 + paddw xmm5, xmm0 ; xmm5=data1 + + ; ---- Pass 2: process columns. + +; mov edx, POINTER [data(eax)] ; (DCTELEM *) + + ; xmm1=(00 10 20 30 40 50 60 70), xmm7=(02 12 22 32 42 52 62 72) + ; xmm5=(01 11 21 31 41 51 61 71), xmm6=(03 13 23 33 43 53 63 73) + + movdqa xmm4, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm5 ; xmm1=(00 01 10 11 20 21 30 31) + punpckhwd xmm4, xmm5 ; xmm4=(40 41 50 51 60 61 70 71) + movdqa xmm0, xmm7 ; transpose coefficients(phase 1) + punpcklwd xmm7, xmm6 ; xmm7=(02 03 12 13 22 23 32 33) + punpckhwd xmm0, xmm6 ; xmm0=(42 43 52 53 62 63 72 73) + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=col4 + movdqa xmm6, XMMWORD [wk(1)] ; xmm6=col6 + + ; xmm5=(04 14 24 34 44 54 64 74), xmm6=(06 16 26 36 46 56 66 76) + ; xmm2=(05 15 25 35 45 55 65 75), xmm3=(07 17 27 37 47 57 67 77) + + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(02 03 12 13 22 23 32 33) + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(42 43 52 53 62 63 72 73) + + movdqa xmm7, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm2 ; xmm5=(04 05 14 15 24 25 34 35) + punpckhwd xmm7, xmm2 ; xmm7=(44 45 54 55 64 65 74 75) + movdqa xmm0, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm3 ; xmm6=(06 07 16 17 26 27 36 37) + punpckhwd xmm0, xmm3 ; xmm0=(46 47 56 57 66 67 76 77) + + movdqa xmm2, xmm5 ; transpose coefficients(phase 2) + punpckldq xmm5, xmm6 ; xmm5=(04 05 06 07 14 15 16 17) + punpckhdq xmm2, xmm6 ; xmm2=(24 25 26 27 34 35 36 37) + movdqa xmm3, xmm7 ; transpose coefficients(phase 2) + punpckldq xmm7, xmm0 ; xmm7=(44 45 46 47 54 55 56 57) + punpckhdq xmm3, xmm0 ; xmm3=(64 65 66 67 74 75 76 77) + + movdqa xmm6, XMMWORD [wk(0)] ; xmm6=(02 03 12 13 22 23 32 33) + movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(42 43 52 53 62 63 72 73) + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(24 25 26 27 34 35 36 37) + movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(44 45 46 47 54 55 56 57) + + movdqa xmm2, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 10 11 12 13) + punpckhdq xmm2, xmm6 ; xmm2=(20 21 22 23 30 31 32 33) + movdqa xmm7, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm0 ; xmm4=(40 41 42 43 50 51 52 53) + punpckhdq xmm7, xmm0 ; xmm7=(60 61 62 63 70 71 72 73) + + movdqa xmm6, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm5 ; xmm1=(00 01 02 03 04 05 06 07)=data0 + punpckhqdq xmm6, xmm5 ; xmm6=(10 11 12 13 14 15 16 17)=data1 + movdqa xmm0, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm3 ; xmm7=(60 61 62 63 64 65 66 67)=data6 + punpckhqdq xmm0, xmm3 ; xmm0=(70 71 72 73 74 75 76 77)=data7 + + movdqa xmm5, xmm6 + movdqa xmm3, xmm1 + psubw xmm6, xmm7 ; xmm6=data1-data6=tmp6 + psubw xmm1, xmm0 ; xmm1=data0-data7=tmp7 + paddw xmm5, xmm7 ; xmm5=data1+data6=tmp1 + paddw xmm3, xmm0 ; xmm3=data0+data7=tmp0 + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(24 25 26 27 34 35 36 37) + movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(44 45 46 47 54 55 56 57) + movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=tmp7 + + movdqa xmm6, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm7 ; xmm2=(20 21 22 23 24 25 26 27)=data2 + punpckhqdq xmm6, xmm7 ; xmm6=(30 31 32 33 34 35 36 37)=data3 + movdqa xmm1, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm0 ; xmm4=(40 41 42 43 44 45 46 47)=data4 + punpckhqdq xmm1, xmm0 ; xmm1=(50 51 52 53 54 55 56 57)=data5 + + movdqa xmm7, xmm6 + movdqa xmm0, xmm2 + paddw xmm6, xmm4 ; xmm6=data3+data4=tmp3 + paddw xmm2, xmm1 ; xmm2=data2+data5=tmp2 + psubw xmm7, xmm4 ; xmm7=data3-data4=tmp4 + psubw xmm0, xmm1 ; xmm0=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm4, xmm3 + movdqa xmm1, xmm5 + psubw xmm3, xmm6 ; xmm3=tmp13 + psubw xmm5, xmm2 ; xmm5=tmp12 + paddw xmm4, xmm6 ; xmm4=tmp10 + paddw xmm1, xmm2 ; xmm1=tmp11 + + paddw xmm5, xmm3 + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm5, [GOTOFF(ebx,PW_F0707)] ; xmm5=z1 + + movdqa xmm6, xmm4 + movdqa xmm2, xmm3 + psubw xmm4, xmm1 ; xmm4=data4 + psubw xmm3, xmm5 ; xmm3=data6 + paddw xmm6, xmm1 ; xmm6=data0 + paddw xmm2, xmm5 ; xmm2=data2 + + movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm4 + movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm3 + movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm6 + movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm2 + + ; -- Odd part + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=tmp6 + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7 + + paddw xmm7, xmm0 ; xmm7=tmp10 + paddw xmm0, xmm1 ; xmm0=tmp11 + paddw xmm1, xmm5 ; xmm1=tmp12, xmm5=tmp7 + + psllw xmm7, PRE_MULTIPLY_SCALE_BITS + psllw xmm1, PRE_MULTIPLY_SCALE_BITS + + psllw xmm0, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm0, [GOTOFF(ebx,PW_F0707)] ; xmm0=z3 + + movdqa xmm4, xmm7 ; xmm4=tmp10 + psubw xmm7, xmm1 + pmulhw xmm7, [GOTOFF(ebx,PW_F0382)] ; xmm7=z5 + pmulhw xmm4, [GOTOFF(ebx,PW_F0541)] ; xmm4=MULTIPLY(tmp10,FIX_0_541196) + pmulhw xmm1, [GOTOFF(ebx,PW_F1306)] ; xmm1=MULTIPLY(tmp12,FIX_1_306562) + paddw xmm4, xmm7 ; xmm4=z2 + paddw xmm1, xmm7 ; xmm1=z4 + + movdqa xmm3, xmm5 + psubw xmm5, xmm0 ; xmm5=z13 + paddw xmm3, xmm0 ; xmm3=z11 + + movdqa xmm6, xmm5 + movdqa xmm2, xmm3 + psubw xmm5, xmm4 ; xmm5=data3 + psubw xmm3, xmm1 ; xmm3=data7 + paddw xmm6, xmm4 ; xmm6=data5 + paddw xmm2, xmm1 ; xmm2=data1 + + movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm5 + movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm3 + movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm6 + movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm2 + +; pop edi ; unused +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; unused + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-avx2.asm new file mode 100644 index 0000000..05ea865 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-avx2.asm @@ -0,0 +1,327 @@ +; +; jfdctint.asm - accurate integer FDCT (AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2018, 2020, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; forward DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jfdctint.c; see the jfdctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit matrix transpose using AVX2 instructions +; %1-%4: Input/output registers +; %5-%8: Temp registers + +%macro DOTRANSPOSE 8 + ; %1=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47) + ; %2=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57) + ; %3=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67) + ; %4=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77) + + vpunpcklwd %5, %1, %2 + vpunpckhwd %6, %1, %2 + vpunpcklwd %7, %3, %4 + vpunpckhwd %8, %3, %4 + ; transpose coefficients(phase 1) + ; %5=(00 10 01 11 02 12 03 13 40 50 41 51 42 52 43 53) + ; %6=(04 14 05 15 06 16 07 17 44 54 45 55 46 56 47 57) + ; %7=(20 30 21 31 22 32 23 33 60 70 61 71 62 72 63 73) + ; %8=(24 34 25 35 26 36 27 37 64 74 65 75 66 76 67 77) + + vpunpckldq %1, %5, %7 + vpunpckhdq %2, %5, %7 + vpunpckldq %3, %6, %8 + vpunpckhdq %4, %6, %8 + ; transpose coefficients(phase 2) + ; %1=(00 10 20 30 01 11 21 31 40 50 60 70 41 51 61 71) + ; %2=(02 12 22 32 03 13 23 33 42 52 62 72 43 53 63 73) + ; %3=(04 14 24 34 05 15 25 35 44 54 64 74 45 55 65 75) + ; %4=(06 16 26 36 07 17 27 37 46 56 66 76 47 57 67 77) + + vpermq %1, %1, 0x8D + vpermq %2, %2, 0x8D + vpermq %3, %3, 0xD8 + vpermq %4, %4, 0xD8 + ; transpose coefficients(phase 3) + ; %1=(01 11 21 31 41 51 61 71 00 10 20 30 40 50 60 70) + ; %2=(03 13 23 33 43 53 63 73 02 12 22 32 42 52 62 72) + ; %3=(04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75) + ; %4=(06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77) +%endmacro + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit accurate integer forward DCT using AVX2 instructions +; %1-%4: Input/output registers +; %5-%8: Temp registers +; %9: Pass (1 or 2) + +%macro DODCT 9 + vpsubw %5, %1, %4 ; %5=data1_0-data6_7=tmp6_7 + vpaddw %6, %1, %4 ; %6=data1_0+data6_7=tmp1_0 + vpaddw %7, %2, %3 ; %7=data3_2+data4_5=tmp3_2 + vpsubw %8, %2, %3 ; %8=data3_2-data4_5=tmp4_5 + + ; -- Even part + + vperm2i128 %6, %6, %6, 0x01 ; %6=tmp0_1 + vpaddw %1, %6, %7 ; %1=tmp0_1+tmp3_2=tmp10_11 + vpsubw %6, %6, %7 ; %6=tmp0_1-tmp3_2=tmp13_12 + + vperm2i128 %7, %1, %1, 0x01 ; %7=tmp11_10 + vpsignw %1, %1, [GOTOFF(ebx, PW_1_NEG1)] ; %1=tmp10_neg11 + vpaddw %7, %7, %1 ; %7=(tmp10+tmp11)_(tmp10-tmp11) +%if %9 == 1 + vpsllw %1, %7, PASS1_BITS ; %1=data0_4 +%else + vpaddw %7, %7, [GOTOFF(ebx, PW_DESCALE_P2X)] + vpsraw %1, %7, PASS1_BITS ; %1=data0_4 +%endif + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + vperm2i128 %7, %6, %6, 0x01 ; %7=tmp12_13 + vpunpcklwd %2, %6, %7 + vpunpckhwd %6, %6, %7 + vpmaddwd %2, %2, [GOTOFF(ebx, PW_F130_F054_MF130_F054)] ; %2=data2_6L + vpmaddwd %6, %6, [GOTOFF(ebx, PW_F130_F054_MF130_F054)] ; %6=data2_6H + + vpaddd %2, %2, [GOTOFF(ebx, PD_DESCALE_P %+ %9)] + vpaddd %6, %6, [GOTOFF(ebx, PD_DESCALE_P %+ %9)] + vpsrad %2, %2, DESCALE_P %+ %9 + vpsrad %6, %6, DESCALE_P %+ %9 + + vpackssdw %3, %2, %6 ; %6=data2_6 + + ; -- Odd part + + vpaddw %7, %8, %5 ; %7=tmp4_5+tmp6_7=z3_4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + vperm2i128 %2, %7, %7, 0x01 ; %2=z4_3 + vpunpcklwd %6, %7, %2 + vpunpckhwd %7, %7, %2 + vpmaddwd %6, %6, [GOTOFF(ebx, PW_MF078_F117_F078_F117)] ; %6=z3_4L + vpmaddwd %7, %7, [GOTOFF(ebx, PW_MF078_F117_F078_F117)] ; %7=z3_4H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + vperm2i128 %4, %5, %5, 0x01 ; %4=tmp7_6 + vpunpcklwd %2, %8, %4 + vpunpckhwd %4, %8, %4 + vpmaddwd %2, %2, [GOTOFF(ebx, PW_MF060_MF089_MF050_MF256)] ; %2=tmp4_5L + vpmaddwd %4, %4, [GOTOFF(ebx, PW_MF060_MF089_MF050_MF256)] ; %4=tmp4_5H + + vpaddd %2, %2, %6 ; %2=data7_5L + vpaddd %4, %4, %7 ; %4=data7_5H + + vpaddd %2, %2, [GOTOFF(ebx, PD_DESCALE_P %+ %9)] + vpaddd %4, %4, [GOTOFF(ebx, PD_DESCALE_P %+ %9)] + vpsrad %2, %2, DESCALE_P %+ %9 + vpsrad %4, %4, DESCALE_P %+ %9 + + vpackssdw %4, %2, %4 ; %4=data7_5 + + vperm2i128 %2, %8, %8, 0x01 ; %2=tmp5_4 + vpunpcklwd %8, %5, %2 + vpunpckhwd %5, %5, %2 + vpmaddwd %8, %8, [GOTOFF(ebx, PW_F050_MF256_F060_MF089)] ; %8=tmp6_7L + vpmaddwd %5, %5, [GOTOFF(ebx, PW_F050_MF256_F060_MF089)] ; %5=tmp6_7H + + vpaddd %8, %8, %6 ; %8=data3_1L + vpaddd %5, %5, %7 ; %5=data3_1H + + vpaddd %8, %8, [GOTOFF(ebx, PD_DESCALE_P %+ %9)] + vpaddd %5, %5, [GOTOFF(ebx, PD_DESCALE_P %+ %9)] + vpsrad %8, %8, DESCALE_P %+ %9 + vpsrad %5, %5, DESCALE_P %+ %9 + + vpackssdw %2, %8, %5 ; %2=data3_1 +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_islow_avx2) + +EXTN(jconst_fdct_islow_avx2): + +PW_F130_F054_MF130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 + times 4 dw (F_0_541 - F_1_847), F_0_541 +PW_MF078_F117_F078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 + times 4 dw (F_1_175 - F_0_390), F_1_175 +PW_MF060_MF089_MF050_MF256 times 4 dw (F_0_298 - F_0_899), -F_0_899 + times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_F050_MF256_F060_MF089 times 4 dw (F_3_072 - F_2_562), -F_2_562 + times 4 dw (F_1_501 - F_0_899), -F_0_899 +PD_DESCALE_P1 times 8 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 8 dd 1 << (DESCALE_P2 - 1) +PW_DESCALE_P2X times 16 dw 1 << (PASS1_BITS - 1) +PW_1_NEG1 times 8 dw 1 + times 8 dw -1 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_islow_avx2(DCTELEM *data) +; + +%define data(b) (b) + 8 ; DCTELEM *data + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_islow_avx2) + +EXTN(jsimd_fdct_islow_avx2): + push ebp + mov ebp, esp + PUSHPIC ebx +; push ecx ; unused +; push edx ; need not be preserved +; push esi ; unused +; push edi ; unused + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process rows. + + mov edx, POINTER [data(ebp)] ; (DCTELEM *) + + vmovdqu ymm4, YMMWORD [YMMBLOCK(0,0,edx,SIZEOF_DCTELEM)] + vmovdqu ymm5, YMMWORD [YMMBLOCK(2,0,edx,SIZEOF_DCTELEM)] + vmovdqu ymm6, YMMWORD [YMMBLOCK(4,0,edx,SIZEOF_DCTELEM)] + vmovdqu ymm7, YMMWORD [YMMBLOCK(6,0,edx,SIZEOF_DCTELEM)] + ; ymm4=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + ; ymm5=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + ; ymm6=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) + ; ymm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) + + vperm2i128 ymm0, ymm4, ymm6, 0x20 + vperm2i128 ymm1, ymm4, ymm6, 0x31 + vperm2i128 ymm2, ymm5, ymm7, 0x20 + vperm2i128 ymm3, ymm5, ymm7, 0x31 + ; ymm0=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47) + ; ymm1=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57) + ; ymm2=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67) + ; ymm3=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77) + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7 + + DODCT ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7, 1 + ; ymm0=data0_4, ymm1=data3_1, ymm2=data2_6, ymm3=data7_5 + + ; ---- Pass 2: process columns. + + vperm2i128 ymm4, ymm1, ymm3, 0x20 ; ymm4=data3_7 + vperm2i128 ymm1, ymm1, ymm3, 0x31 ; ymm1=data1_5 + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7 + + DODCT ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7, 2 + ; ymm0=data0_4, ymm1=data3_1, ymm2=data2_6, ymm4=data7_5 + + vperm2i128 ymm3, ymm0, ymm1, 0x30 ; ymm3=data0_1 + vperm2i128 ymm5, ymm2, ymm1, 0x20 ; ymm5=data2_3 + vperm2i128 ymm6, ymm0, ymm4, 0x31 ; ymm6=data4_5 + vperm2i128 ymm7, ymm2, ymm4, 0x21 ; ymm7=data6_7 + + vmovdqu YMMWORD [YMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], ymm3 + vmovdqu YMMWORD [YMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], ymm5 + vmovdqu YMMWORD [YMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], ymm6 + vmovdqu YMMWORD [YMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], ymm7 + + vzeroupper +; pop edi ; unused +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; unused + POPPIC ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-mmx.asm new file mode 100644 index 0000000..7d4c61c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-mmx.asm @@ -0,0 +1,616 @@ +; +; jfdctint.asm - accurate integer FDCT (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2020, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; forward DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jfdctint.c; see the jfdctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_islow_mmx) + +EXTN(jconst_fdct_islow_mmx): + +PW_F130_F054 times 2 dw (F_0_541 + F_0_765), F_0_541 +PW_F054_MF130 times 2 dw F_0_541, (F_0_541 - F_1_847) +PW_MF078_F117 times 2 dw (F_1_175 - F_1_961), F_1_175 +PW_F117_F078 times 2 dw F_1_175, (F_1_175 - F_0_390) +PW_MF060_MF089 times 2 dw (F_0_298 - F_0_899), -F_0_899 +PW_MF089_F060 times 2 dw -F_0_899, (F_1_501 - F_0_899) +PW_MF050_MF256 times 2 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF256_F050 times 2 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 2 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2 - 1) +PW_DESCALE_P2X times 4 dw 1 << (PASS1_BITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_islow_mmx(DCTELEM *data) +; + +%define data(b) (b) + 8 ; DCTELEM *data + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD ; mmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_islow_mmx) + +EXTN(jsimd_fdct_islow_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved +; push esi ; unused +; push edi ; unused + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process rows. + + mov edx, POINTER [data(eax)] ; (DCTELEM *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.rowloop: + + movq mm0, MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)] + movq mm2, MMWORD [MMBLOCK(2,1,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(3,1,edx,SIZEOF_DCTELEM)] + + ; mm0=(20 21 22 23), mm2=(24 25 26 27) + ; mm1=(30 31 32 33), mm3=(34 35 36 37) + + movq mm4, mm0 ; transpose coefficients(phase 1) + punpcklwd mm0, mm1 ; mm0=(20 30 21 31) + punpckhwd mm4, mm1 ; mm4=(22 32 23 33) + movq mm5, mm2 ; transpose coefficients(phase 1) + punpcklwd mm2, mm3 ; mm2=(24 34 25 35) + punpckhwd mm5, mm3 ; mm5=(26 36 27 37) + + movq mm6, MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)] + movq mm7, MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(0,1,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(1,1,edx,SIZEOF_DCTELEM)] + + ; mm6=(00 01 02 03), mm1=(04 05 06 07) + ; mm7=(10 11 12 13), mm3=(14 15 16 17) + + movq MMWORD [wk(0)], mm4 ; wk(0)=(22 32 23 33) + movq MMWORD [wk(1)], mm2 ; wk(1)=(24 34 25 35) + + movq mm4, mm6 ; transpose coefficients(phase 1) + punpcklwd mm6, mm7 ; mm6=(00 10 01 11) + punpckhwd mm4, mm7 ; mm4=(02 12 03 13) + movq mm2, mm1 ; transpose coefficients(phase 1) + punpcklwd mm1, mm3 ; mm1=(04 14 05 15) + punpckhwd mm2, mm3 ; mm2=(06 16 07 17) + + movq mm7, mm6 ; transpose coefficients(phase 2) + punpckldq mm6, mm0 ; mm6=(00 10 20 30)=data0 + punpckhdq mm7, mm0 ; mm7=(01 11 21 31)=data1 + movq mm3, mm2 ; transpose coefficients(phase 2) + punpckldq mm2, mm5 ; mm2=(06 16 26 36)=data6 + punpckhdq mm3, mm5 ; mm3=(07 17 27 37)=data7 + + movq mm0, mm7 + movq mm5, mm6 + psubw mm7, mm2 ; mm7=data1-data6=tmp6 + psubw mm6, mm3 ; mm6=data0-data7=tmp7 + paddw mm0, mm2 ; mm0=data1+data6=tmp1 + paddw mm5, mm3 ; mm5=data0+data7=tmp0 + + movq mm2, MMWORD [wk(0)] ; mm2=(22 32 23 33) + movq mm3, MMWORD [wk(1)] ; mm3=(24 34 25 35) + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp6 + movq MMWORD [wk(1)], mm6 ; wk(1)=tmp7 + + movq mm7, mm4 ; transpose coefficients(phase 2) + punpckldq mm4, mm2 ; mm4=(02 12 22 32)=data2 + punpckhdq mm7, mm2 ; mm7=(03 13 23 33)=data3 + movq mm6, mm1 ; transpose coefficients(phase 2) + punpckldq mm1, mm3 ; mm1=(04 14 24 34)=data4 + punpckhdq mm6, mm3 ; mm6=(05 15 25 35)=data5 + + movq mm2, mm7 + movq mm3, mm4 + paddw mm7, mm1 ; mm7=data3+data4=tmp3 + paddw mm4, mm6 ; mm4=data2+data5=tmp2 + psubw mm2, mm1 ; mm2=data3-data4=tmp4 + psubw mm3, mm6 ; mm3=data2-data5=tmp5 + + ; -- Even part + + movq mm1, mm5 + movq mm6, mm0 + paddw mm5, mm7 ; mm5=tmp10 + paddw mm0, mm4 ; mm0=tmp11 + psubw mm1, mm7 ; mm1=tmp13 + psubw mm6, mm4 ; mm6=tmp12 + + movq mm7, mm5 + paddw mm5, mm0 ; mm5=tmp10+tmp11 + psubw mm7, mm0 ; mm7=tmp10-tmp11 + + psllw mm5, PASS1_BITS ; mm5=data0 + psllw mm7, PASS1_BITS ; mm7=data4 + + movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)], mm5 + movq MMWORD [MMBLOCK(0,1,edx,SIZEOF_DCTELEM)], mm7 + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + movq mm4, mm1 ; mm1=tmp13 + movq mm0, mm1 + punpcklwd mm4, mm6 ; mm6=tmp12 + punpckhwd mm0, mm6 + movq mm1, mm4 + movq mm6, mm0 + pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=data2L + pmaddwd mm0, [GOTOFF(ebx,PW_F130_F054)] ; mm0=data2H + pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=data6L + pmaddwd mm6, [GOTOFF(ebx,PW_F054_MF130)] ; mm6=data6H + + paddd mm4, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd mm0, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad mm4, DESCALE_P1 + psrad mm0, DESCALE_P1 + paddd mm1, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd mm6, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad mm1, DESCALE_P1 + psrad mm6, DESCALE_P1 + + packssdw mm4, mm0 ; mm4=data2 + packssdw mm1, mm6 ; mm1=data6 + + movq MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)], mm4 + movq MMWORD [MMBLOCK(2,1,edx,SIZEOF_DCTELEM)], mm1 + + ; -- Odd part + + movq mm5, MMWORD [wk(0)] ; mm5=tmp6 + movq mm7, MMWORD [wk(1)] ; mm7=tmp7 + + movq mm0, mm2 ; mm2=tmp4 + movq mm6, mm3 ; mm3=tmp5 + paddw mm0, mm5 ; mm0=z3 + paddw mm6, mm7 ; mm6=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movq mm4, mm0 + movq mm1, mm0 + punpcklwd mm4, mm6 + punpckhwd mm1, mm6 + movq mm0, mm4 + movq mm6, mm1 + pmaddwd mm4, [GOTOFF(ebx,PW_MF078_F117)] ; mm4=z3L + pmaddwd mm1, [GOTOFF(ebx,PW_MF078_F117)] ; mm1=z3H + pmaddwd mm0, [GOTOFF(ebx,PW_F117_F078)] ; mm0=z4L + pmaddwd mm6, [GOTOFF(ebx,PW_F117_F078)] ; mm6=z4H + + movq MMWORD [wk(0)], mm4 ; wk(0)=z3L + movq MMWORD [wk(1)], mm1 ; wk(1)=z3H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + movq mm4, mm2 + movq mm1, mm2 + punpcklwd mm4, mm7 + punpckhwd mm1, mm7 + movq mm2, mm4 + movq mm7, mm1 + pmaddwd mm4, [GOTOFF(ebx,PW_MF060_MF089)] ; mm4=tmp4L + pmaddwd mm1, [GOTOFF(ebx,PW_MF060_MF089)] ; mm1=tmp4H + pmaddwd mm2, [GOTOFF(ebx,PW_MF089_F060)] ; mm2=tmp7L + pmaddwd mm7, [GOTOFF(ebx,PW_MF089_F060)] ; mm7=tmp7H + + paddd mm4, MMWORD [wk(0)] ; mm4=data7L + paddd mm1, MMWORD [wk(1)] ; mm1=data7H + paddd mm2, mm0 ; mm2=data1L + paddd mm7, mm6 ; mm7=data1H + + paddd mm4, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd mm1, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad mm4, DESCALE_P1 + psrad mm1, DESCALE_P1 + paddd mm2, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd mm7, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad mm2, DESCALE_P1 + psrad mm7, DESCALE_P1 + + packssdw mm4, mm1 ; mm4=data7 + packssdw mm2, mm7 ; mm2=data1 + + movq MMWORD [MMBLOCK(3,1,edx,SIZEOF_DCTELEM)], mm4 + movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)], mm2 + + movq mm1, mm3 + movq mm7, mm3 + punpcklwd mm1, mm5 + punpckhwd mm7, mm5 + movq mm3, mm1 + movq mm5, mm7 + pmaddwd mm1, [GOTOFF(ebx,PW_MF050_MF256)] ; mm1=tmp5L + pmaddwd mm7, [GOTOFF(ebx,PW_MF050_MF256)] ; mm7=tmp5H + pmaddwd mm3, [GOTOFF(ebx,PW_MF256_F050)] ; mm3=tmp6L + pmaddwd mm5, [GOTOFF(ebx,PW_MF256_F050)] ; mm5=tmp6H + + paddd mm1, mm0 ; mm1=data5L + paddd mm7, mm6 ; mm7=data5H + paddd mm3, MMWORD [wk(0)] ; mm3=data3L + paddd mm5, MMWORD [wk(1)] ; mm5=data3H + + paddd mm1, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd mm7, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad mm1, DESCALE_P1 + psrad mm7, DESCALE_P1 + paddd mm3, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd mm5, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad mm3, DESCALE_P1 + psrad mm5, DESCALE_P1 + + packssdw mm1, mm7 ; mm1=data5 + packssdw mm3, mm5 ; mm3=data3 + + movq MMWORD [MMBLOCK(1,1,edx,SIZEOF_DCTELEM)], mm1 + movq MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)], mm3 + + add edx, byte 4*DCTSIZE*SIZEOF_DCTELEM + dec ecx + jnz near .rowloop + + ; ---- Pass 2: process columns. + + mov edx, POINTER [data(eax)] ; (DCTELEM *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.columnloop: + + movq mm0, MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)] + movq mm2, MMWORD [MMBLOCK(6,0,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_DCTELEM)] + + ; mm0=(02 12 22 32), mm2=(42 52 62 72) + ; mm1=(03 13 23 33), mm3=(43 53 63 73) + + movq mm4, mm0 ; transpose coefficients(phase 1) + punpcklwd mm0, mm1 ; mm0=(02 03 12 13) + punpckhwd mm4, mm1 ; mm4=(22 23 32 33) + movq mm5, mm2 ; transpose coefficients(phase 1) + punpcklwd mm2, mm3 ; mm2=(42 43 52 53) + punpckhwd mm5, mm3 ; mm5=(62 63 72 73) + + movq mm6, MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)] + movq mm7, MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)] + movq mm1, MMWORD [MMBLOCK(4,0,edx,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(5,0,edx,SIZEOF_DCTELEM)] + + ; mm6=(00 10 20 30), mm1=(40 50 60 70) + ; mm7=(01 11 21 31), mm3=(41 51 61 71) + + movq MMWORD [wk(0)], mm4 ; wk(0)=(22 23 32 33) + movq MMWORD [wk(1)], mm2 ; wk(1)=(42 43 52 53) + + movq mm4, mm6 ; transpose coefficients(phase 1) + punpcklwd mm6, mm7 ; mm6=(00 01 10 11) + punpckhwd mm4, mm7 ; mm4=(20 21 30 31) + movq mm2, mm1 ; transpose coefficients(phase 1) + punpcklwd mm1, mm3 ; mm1=(40 41 50 51) + punpckhwd mm2, mm3 ; mm2=(60 61 70 71) + + movq mm7, mm6 ; transpose coefficients(phase 2) + punpckldq mm6, mm0 ; mm6=(00 01 02 03)=data0 + punpckhdq mm7, mm0 ; mm7=(10 11 12 13)=data1 + movq mm3, mm2 ; transpose coefficients(phase 2) + punpckldq mm2, mm5 ; mm2=(60 61 62 63)=data6 + punpckhdq mm3, mm5 ; mm3=(70 71 72 73)=data7 + + movq mm0, mm7 + movq mm5, mm6 + psubw mm7, mm2 ; mm7=data1-data6=tmp6 + psubw mm6, mm3 ; mm6=data0-data7=tmp7 + paddw mm0, mm2 ; mm0=data1+data6=tmp1 + paddw mm5, mm3 ; mm5=data0+data7=tmp0 + + movq mm2, MMWORD [wk(0)] ; mm2=(22 23 32 33) + movq mm3, MMWORD [wk(1)] ; mm3=(42 43 52 53) + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp6 + movq MMWORD [wk(1)], mm6 ; wk(1)=tmp7 + + movq mm7, mm4 ; transpose coefficients(phase 2) + punpckldq mm4, mm2 ; mm4=(20 21 22 23)=data2 + punpckhdq mm7, mm2 ; mm7=(30 31 32 33)=data3 + movq mm6, mm1 ; transpose coefficients(phase 2) + punpckldq mm1, mm3 ; mm1=(40 41 42 43)=data4 + punpckhdq mm6, mm3 ; mm6=(50 51 52 53)=data5 + + movq mm2, mm7 + movq mm3, mm4 + paddw mm7, mm1 ; mm7=data3+data4=tmp3 + paddw mm4, mm6 ; mm4=data2+data5=tmp2 + psubw mm2, mm1 ; mm2=data3-data4=tmp4 + psubw mm3, mm6 ; mm3=data2-data5=tmp5 + + ; -- Even part + + movq mm1, mm5 + movq mm6, mm0 + paddw mm5, mm7 ; mm5=tmp10 + paddw mm0, mm4 ; mm0=tmp11 + psubw mm1, mm7 ; mm1=tmp13 + psubw mm6, mm4 ; mm6=tmp12 + + movq mm7, mm5 + paddw mm5, mm0 ; mm5=tmp10+tmp11 + psubw mm7, mm0 ; mm7=tmp10-tmp11 + + paddw mm5, [GOTOFF(ebx,PW_DESCALE_P2X)] + paddw mm7, [GOTOFF(ebx,PW_DESCALE_P2X)] + psraw mm5, PASS1_BITS ; mm5=data0 + psraw mm7, PASS1_BITS ; mm7=data4 + + movq MMWORD [MMBLOCK(0,0,edx,SIZEOF_DCTELEM)], mm5 + movq MMWORD [MMBLOCK(4,0,edx,SIZEOF_DCTELEM)], mm7 + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + movq mm4, mm1 ; mm1=tmp13 + movq mm0, mm1 + punpcklwd mm4, mm6 ; mm6=tmp12 + punpckhwd mm0, mm6 + movq mm1, mm4 + movq mm6, mm0 + pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=data2L + pmaddwd mm0, [GOTOFF(ebx,PW_F130_F054)] ; mm0=data2H + pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=data6L + pmaddwd mm6, [GOTOFF(ebx,PW_F054_MF130)] ; mm6=data6H + + paddd mm4, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd mm0, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad mm4, DESCALE_P2 + psrad mm0, DESCALE_P2 + paddd mm1, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd mm6, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad mm1, DESCALE_P2 + psrad mm6, DESCALE_P2 + + packssdw mm4, mm0 ; mm4=data2 + packssdw mm1, mm6 ; mm1=data6 + + movq MMWORD [MMBLOCK(2,0,edx,SIZEOF_DCTELEM)], mm4 + movq MMWORD [MMBLOCK(6,0,edx,SIZEOF_DCTELEM)], mm1 + + ; -- Odd part + + movq mm5, MMWORD [wk(0)] ; mm5=tmp6 + movq mm7, MMWORD [wk(1)] ; mm7=tmp7 + + movq mm0, mm2 ; mm2=tmp4 + movq mm6, mm3 ; mm3=tmp5 + paddw mm0, mm5 ; mm0=z3 + paddw mm6, mm7 ; mm6=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movq mm4, mm0 + movq mm1, mm0 + punpcklwd mm4, mm6 + punpckhwd mm1, mm6 + movq mm0, mm4 + movq mm6, mm1 + pmaddwd mm4, [GOTOFF(ebx,PW_MF078_F117)] ; mm4=z3L + pmaddwd mm1, [GOTOFF(ebx,PW_MF078_F117)] ; mm1=z3H + pmaddwd mm0, [GOTOFF(ebx,PW_F117_F078)] ; mm0=z4L + pmaddwd mm6, [GOTOFF(ebx,PW_F117_F078)] ; mm6=z4H + + movq MMWORD [wk(0)], mm4 ; wk(0)=z3L + movq MMWORD [wk(1)], mm1 ; wk(1)=z3H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + movq mm4, mm2 + movq mm1, mm2 + punpcklwd mm4, mm7 + punpckhwd mm1, mm7 + movq mm2, mm4 + movq mm7, mm1 + pmaddwd mm4, [GOTOFF(ebx,PW_MF060_MF089)] ; mm4=tmp4L + pmaddwd mm1, [GOTOFF(ebx,PW_MF060_MF089)] ; mm1=tmp4H + pmaddwd mm2, [GOTOFF(ebx,PW_MF089_F060)] ; mm2=tmp7L + pmaddwd mm7, [GOTOFF(ebx,PW_MF089_F060)] ; mm7=tmp7H + + paddd mm4, MMWORD [wk(0)] ; mm4=data7L + paddd mm1, MMWORD [wk(1)] ; mm1=data7H + paddd mm2, mm0 ; mm2=data1L + paddd mm7, mm6 ; mm7=data1H + + paddd mm4, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd mm1, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad mm4, DESCALE_P2 + psrad mm1, DESCALE_P2 + paddd mm2, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd mm7, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad mm2, DESCALE_P2 + psrad mm7, DESCALE_P2 + + packssdw mm4, mm1 ; mm4=data7 + packssdw mm2, mm7 ; mm2=data1 + + movq MMWORD [MMBLOCK(7,0,edx,SIZEOF_DCTELEM)], mm4 + movq MMWORD [MMBLOCK(1,0,edx,SIZEOF_DCTELEM)], mm2 + + movq mm1, mm3 + movq mm7, mm3 + punpcklwd mm1, mm5 + punpckhwd mm7, mm5 + movq mm3, mm1 + movq mm5, mm7 + pmaddwd mm1, [GOTOFF(ebx,PW_MF050_MF256)] ; mm1=tmp5L + pmaddwd mm7, [GOTOFF(ebx,PW_MF050_MF256)] ; mm7=tmp5H + pmaddwd mm3, [GOTOFF(ebx,PW_MF256_F050)] ; mm3=tmp6L + pmaddwd mm5, [GOTOFF(ebx,PW_MF256_F050)] ; mm5=tmp6H + + paddd mm1, mm0 ; mm1=data5L + paddd mm7, mm6 ; mm7=data5H + paddd mm3, MMWORD [wk(0)] ; mm3=data3L + paddd mm5, MMWORD [wk(1)] ; mm5=data3H + + paddd mm1, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd mm7, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad mm1, DESCALE_P2 + psrad mm7, DESCALE_P2 + paddd mm3, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd mm5, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad mm3, DESCALE_P2 + psrad mm5, DESCALE_P2 + + packssdw mm1, mm7 ; mm1=data5 + packssdw mm3, mm5 ; mm3=data3 + + movq MMWORD [MMBLOCK(5,0,edx,SIZEOF_DCTELEM)], mm1 + movq MMWORD [MMBLOCK(3,0,edx,SIZEOF_DCTELEM)], mm3 + + add edx, byte 4*SIZEOF_DCTELEM + dec ecx + jnz near .columnloop + + emms ; empty MMX state + +; pop edi ; unused +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-sse2.asm new file mode 100644 index 0000000..7ed5c95 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jfdctint-sse2.asm @@ -0,0 +1,629 @@ +; +; jfdctint.asm - accurate integer FDCT (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2020, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; forward DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jfdctint.c; see the jfdctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_islow_sse2) + +EXTN(jconst_fdct_islow_sse2): + +PW_F130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 +PW_F054_MF130 times 4 dw F_0_541, (F_0_541 - F_1_847) +PW_MF078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 +PW_F117_F078 times 4 dw F_1_175, (F_1_175 - F_0_390) +PW_MF060_MF089 times 4 dw (F_0_298 - F_0_899), -F_0_899 +PW_MF089_F060 times 4 dw -F_0_899, (F_1_501 - F_0_899) +PW_MF050_MF256 times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF256_F050 times 4 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2 - 1) +PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_islow_sse2(DCTELEM *data) +; + +%define data(b) (b) + 8 ; DCTELEM *data + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 6 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_islow_sse2) + +EXTN(jsimd_fdct_islow_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; unused +; push edx ; need not be preserved +; push esi ; unused +; push edi ; unused + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process rows. + + mov edx, POINTER [data(eax)] ; (DCTELEM *) + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)] + movdqa xmm2, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)] + + ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27) + ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37) + + movdqa xmm4, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13) + punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17) + movdqa xmm5, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33) + punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37) + + movdqa xmm6, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)] + movdqa xmm7, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)] + + ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62) + ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63) + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33) + movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37) + + movdqa xmm2, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53) + punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57) + movdqa xmm5, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73) + punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77) + + movdqa xmm7, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71) + punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73) + movdqa xmm3, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75) + punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77) + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33) + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37) + movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73) + movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75) + + movdqa xmm7, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31) + punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33) + movdqa xmm2, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35) + punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37) + + movdqa xmm1, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0 + punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1 + movdqa xmm5, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6 + punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7 + + movdqa xmm6, xmm1 + movdqa xmm3, xmm0 + psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6 + psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7 + paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1 + paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0 + + movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73) + movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75) + movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7 + + movdqa xmm1, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2 + punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3 + movdqa xmm0, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4 + punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5 + + movdqa xmm2, xmm1 + movdqa xmm5, xmm7 + paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3 + paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2 + psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4 + psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm4, xmm3 + movdqa xmm0, xmm6 + paddw xmm3, xmm1 ; xmm3=tmp10 + paddw xmm6, xmm7 ; xmm6=tmp11 + psubw xmm4, xmm1 ; xmm4=tmp13 + psubw xmm0, xmm7 ; xmm0=tmp12 + + movdqa xmm1, xmm3 + paddw xmm3, xmm6 ; xmm3=tmp10+tmp11 + psubw xmm1, xmm6 ; xmm1=tmp10-tmp11 + + psllw xmm3, PASS1_BITS ; xmm3=data0 + psllw xmm1, PASS1_BITS ; xmm1=data4 + + movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0 + movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4 + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + movdqa xmm7, xmm4 ; xmm4=tmp13 + movdqa xmm6, xmm4 + punpcklwd xmm7, xmm0 ; xmm0=tmp12 + punpckhwd xmm6, xmm0 + movdqa xmm4, xmm7 + movdqa xmm0, xmm6 + pmaddwd xmm7, [GOTOFF(ebx,PW_F130_F054)] ; xmm7=data2L + pmaddwd xmm6, [GOTOFF(ebx,PW_F130_F054)] ; xmm6=data2H + pmaddwd xmm4, [GOTOFF(ebx,PW_F054_MF130)] ; xmm4=data6L + pmaddwd xmm0, [GOTOFF(ebx,PW_F054_MF130)] ; xmm0=data6H + + paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad xmm7, DESCALE_P1 + psrad xmm6, DESCALE_P1 + paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd xmm0, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad xmm4, DESCALE_P1 + psrad xmm0, DESCALE_P1 + + packssdw xmm7, xmm6 ; xmm7=data2 + packssdw xmm4, xmm0 ; xmm4=data6 + + movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2 + movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6 + + ; -- Odd part + + movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6 + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7 + + movdqa xmm6, xmm2 ; xmm2=tmp4 + movdqa xmm0, xmm5 ; xmm5=tmp5 + paddw xmm6, xmm3 ; xmm6=z3 + paddw xmm0, xmm1 ; xmm0=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm7, xmm6 + movdqa xmm4, xmm6 + punpcklwd xmm7, xmm0 + punpckhwd xmm4, xmm0 + movdqa xmm6, xmm7 + movdqa xmm0, xmm4 + pmaddwd xmm7, [GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3L + pmaddwd xmm4, [GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3H + pmaddwd xmm6, [GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4L + pmaddwd xmm0, [GOTOFF(ebx,PW_F117_F078)] ; xmm0=z4H + + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L + movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + movdqa xmm7, xmm2 + movdqa xmm4, xmm2 + punpcklwd xmm7, xmm1 + punpckhwd xmm4, xmm1 + movdqa xmm2, xmm7 + movdqa xmm1, xmm4 + pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp4L + pmaddwd xmm4, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4H + pmaddwd xmm2, [GOTOFF(ebx,PW_MF089_F060)] ; xmm2=tmp7L + pmaddwd xmm1, [GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp7H + + paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L + paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H + paddd xmm2, xmm6 ; xmm2=data1L + paddd xmm1, xmm0 ; xmm1=data1H + + paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad xmm7, DESCALE_P1 + psrad xmm4, DESCALE_P1 + paddd xmm2, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad xmm2, DESCALE_P1 + psrad xmm1, DESCALE_P1 + + packssdw xmm7, xmm4 ; xmm7=data7 + packssdw xmm2, xmm1 ; xmm2=data1 + + movdqa xmm4, xmm5 + movdqa xmm1, xmm5 + punpcklwd xmm4, xmm3 + punpckhwd xmm1, xmm3 + movdqa xmm5, xmm4 + movdqa xmm3, xmm1 + pmaddwd xmm4, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm4=tmp5L + pmaddwd xmm1, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5H + pmaddwd xmm5, [GOTOFF(ebx,PW_MF256_F050)] ; xmm5=tmp6L + pmaddwd xmm3, [GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6H + + paddd xmm4, xmm6 ; xmm4=data5L + paddd xmm1, xmm0 ; xmm1=data5H + paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L + paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H + + paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad xmm4, DESCALE_P1 + psrad xmm1, DESCALE_P1 + paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P1)] + paddd xmm3, [GOTOFF(ebx,PD_DESCALE_P1)] + psrad xmm5, DESCALE_P1 + psrad xmm3, DESCALE_P1 + + packssdw xmm4, xmm1 ; xmm4=data5 + packssdw xmm5, xmm3 ; xmm5=data3 + + ; ---- Pass 2: process columns. + +; mov edx, POINTER [data(eax)] ; (DCTELEM *) + + movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0 + movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2 + + ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72) + ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73) + + movdqa xmm1, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm2 ; xmm6=(00 01 10 11 20 21 30 31) + punpckhwd xmm1, xmm2 ; xmm1=(40 41 50 51 60 61 70 71) + movdqa xmm3, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm5 ; xmm0=(02 03 12 13 22 23 32 33) + punpckhwd xmm3, xmm5 ; xmm3=(42 43 52 53 62 63 72 73) + + movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4 + movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6 + + ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76) + ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77) + + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33) + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73) + + movdqa xmm0, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm4 ; xmm2=(04 05 14 15 24 25 34 35) + punpckhwd xmm0, xmm4 ; xmm0=(44 45 54 55 64 65 74 75) + movdqa xmm3, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm7 ; xmm5=(06 07 16 17 26 27 36 37) + punpckhwd xmm3, xmm7 ; xmm3=(46 47 56 57 66 67 76 77) + + movdqa xmm4, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm5 ; xmm2=(04 05 06 07 14 15 16 17) + punpckhdq xmm4, xmm5 ; xmm4=(24 25 26 27 34 35 36 37) + movdqa xmm7, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm3 ; xmm0=(44 45 46 47 54 55 56 57) + punpckhdq xmm7, xmm3 ; xmm7=(64 65 66 67 74 75 76 77) + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33) + movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73) + movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37) + movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57) + + movdqa xmm4, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm5 ; xmm6=(00 01 02 03 10 11 12 13) + punpckhdq xmm4, xmm5 ; xmm4=(20 21 22 23 30 31 32 33) + movdqa xmm0, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm3 ; xmm1=(40 41 42 43 50 51 52 53) + punpckhdq xmm0, xmm3 ; xmm0=(60 61 62 63 70 71 72 73) + + movdqa xmm5, xmm6 ; transpose coefficients(phase 3) + punpcklqdq xmm6, xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0 + punpckhqdq xmm5, xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1 + movdqa xmm3, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6 + punpckhqdq xmm3, xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7 + + movdqa xmm2, xmm5 + movdqa xmm7, xmm6 + psubw xmm5, xmm0 ; xmm5=data1-data6=tmp6 + psubw xmm6, xmm3 ; xmm6=data0-data7=tmp7 + paddw xmm2, xmm0 ; xmm2=data1+data6=tmp1 + paddw xmm7, xmm3 ; xmm7=data0+data7=tmp0 + + movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37) + movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57) + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 + + movdqa xmm5, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2 + punpckhqdq xmm5, xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3 + movdqa xmm6, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4 + punpckhqdq xmm6, xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5 + + movdqa xmm0, xmm5 + movdqa xmm3, xmm4 + paddw xmm5, xmm1 ; xmm5=data3+data4=tmp3 + paddw xmm4, xmm6 ; xmm4=data2+data5=tmp2 + psubw xmm0, xmm1 ; xmm0=data3-data4=tmp4 + psubw xmm3, xmm6 ; xmm3=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm1, xmm7 + movdqa xmm6, xmm2 + paddw xmm7, xmm5 ; xmm7=tmp10 + paddw xmm2, xmm4 ; xmm2=tmp11 + psubw xmm1, xmm5 ; xmm1=tmp13 + psubw xmm6, xmm4 ; xmm6=tmp12 + + movdqa xmm5, xmm7 + paddw xmm7, xmm2 ; xmm7=tmp10+tmp11 + psubw xmm5, xmm2 ; xmm5=tmp10-tmp11 + + paddw xmm7, [GOTOFF(ebx,PW_DESCALE_P2X)] + paddw xmm5, [GOTOFF(ebx,PW_DESCALE_P2X)] + psraw xmm7, PASS1_BITS ; xmm7=data0 + psraw xmm5, PASS1_BITS ; xmm5=data4 + + movdqa XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_DCTELEM)], xmm7 + movdqa XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_DCTELEM)], xmm5 + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + movdqa xmm4, xmm1 ; xmm1=tmp13 + movdqa xmm2, xmm1 + punpcklwd xmm4, xmm6 ; xmm6=tmp12 + punpckhwd xmm2, xmm6 + movdqa xmm1, xmm4 + movdqa xmm6, xmm2 + pmaddwd xmm4, [GOTOFF(ebx,PW_F130_F054)] ; xmm4=data2L + pmaddwd xmm2, [GOTOFF(ebx,PW_F130_F054)] ; xmm2=data2H + pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=data6L + pmaddwd xmm6, [GOTOFF(ebx,PW_F054_MF130)] ; xmm6=data6H + + paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd xmm2, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad xmm4, DESCALE_P2 + psrad xmm2, DESCALE_P2 + paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad xmm1, DESCALE_P2 + psrad xmm6, DESCALE_P2 + + packssdw xmm4, xmm2 ; xmm4=data2 + packssdw xmm1, xmm6 ; xmm1=data6 + + movdqa XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_DCTELEM)], xmm4 + movdqa XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_DCTELEM)], xmm1 + + ; -- Odd part + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6 + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7 + + movdqa xmm2, xmm0 ; xmm0=tmp4 + movdqa xmm6, xmm3 ; xmm3=tmp5 + paddw xmm2, xmm7 ; xmm2=z3 + paddw xmm6, xmm5 ; xmm6=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm4, xmm2 + movdqa xmm1, xmm2 + punpcklwd xmm4, xmm6 + punpckhwd xmm1, xmm6 + movdqa xmm2, xmm4 + movdqa xmm6, xmm1 + pmaddwd xmm4, [GOTOFF(ebx,PW_MF078_F117)] ; xmm4=z3L + pmaddwd xmm1, [GOTOFF(ebx,PW_MF078_F117)] ; xmm1=z3H + pmaddwd xmm2, [GOTOFF(ebx,PW_F117_F078)] ; xmm2=z4L + pmaddwd xmm6, [GOTOFF(ebx,PW_F117_F078)] ; xmm6=z4H + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L + movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + movdqa xmm4, xmm0 + movdqa xmm1, xmm0 + punpcklwd xmm4, xmm5 + punpckhwd xmm1, xmm5 + movdqa xmm0, xmm4 + movdqa xmm5, xmm1 + pmaddwd xmm4, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm4=tmp4L + pmaddwd xmm1, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm1=tmp4H + pmaddwd xmm0, [GOTOFF(ebx,PW_MF089_F060)] ; xmm0=tmp7L + pmaddwd xmm5, [GOTOFF(ebx,PW_MF089_F060)] ; xmm5=tmp7H + + paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L + paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H + paddd xmm0, xmm2 ; xmm0=data1L + paddd xmm5, xmm6 ; xmm5=data1H + + paddd xmm4, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad xmm4, DESCALE_P2 + psrad xmm1, DESCALE_P2 + paddd xmm0, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad xmm0, DESCALE_P2 + psrad xmm5, DESCALE_P2 + + packssdw xmm4, xmm1 ; xmm4=data7 + packssdw xmm0, xmm5 ; xmm0=data1 + + movdqa XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_DCTELEM)], xmm4 + movdqa XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_DCTELEM)], xmm0 + + movdqa xmm1, xmm3 + movdqa xmm5, xmm3 + punpcklwd xmm1, xmm7 + punpckhwd xmm5, xmm7 + movdqa xmm3, xmm1 + movdqa xmm7, xmm5 + pmaddwd xmm1, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm1=tmp5L + pmaddwd xmm5, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm5=tmp5H + pmaddwd xmm3, [GOTOFF(ebx,PW_MF256_F050)] ; xmm3=tmp6L + pmaddwd xmm7, [GOTOFF(ebx,PW_MF256_F050)] ; xmm7=tmp6H + + paddd xmm1, xmm2 ; xmm1=data5L + paddd xmm5, xmm6 ; xmm5=data5H + paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L + paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H + + paddd xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd xmm5, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad xmm1, DESCALE_P2 + psrad xmm5, DESCALE_P2 + paddd xmm3, [GOTOFF(ebx,PD_DESCALE_P2)] + paddd xmm7, [GOTOFF(ebx,PD_DESCALE_P2)] + psrad xmm3, DESCALE_P2 + psrad xmm7, DESCALE_P2 + + packssdw xmm1, xmm5 ; xmm1=data5 + packssdw xmm3, xmm7 ; xmm3=data3 + + movdqa XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_DCTELEM)], xmm1 + movdqa XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_DCTELEM)], xmm3 + +; pop edi ; unused +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; unused + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-3dn.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-3dn.asm new file mode 100644 index 0000000..8612eee --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-3dn.asm @@ -0,0 +1,447 @@ +; +; jidctflt.asm - floating-point IDCT (3DNow! & MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a floating-point implementation of the inverse DCT +; (Discrete Cosine Transform). The following code is based directly on +; the IJG's original jidctflt.c; see the jidctflt.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_float_3dnow) + +EXTN(jconst_idct_float_3dnow): + +PD_1_414 times 2 dd 1.414213562373095048801689 +PD_1_847 times 2 dd 1.847759065022573512256366 +PD_1_082 times 2 dd 1.082392200292393968799446 +PD_2_613 times 2 dd 2.613125929752753055713286 +PD_RNDINT_MAGIC times 2 dd 100663296.0 ; (float)(0x00C00000 << 3) +PB_CENTERJSAMP times 8 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_float_3dnow(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; void *dct_table +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD + ; mmword wk[WK_NUM] +%define WK_NUM 2 +%define workspace wk(0) - DCTSIZE2 * SIZEOF_FAST_FLOAT + ; FAST_FLOAT workspace[DCTSIZE2] + + align 32 + GLOBAL_FUNCTION(jsimd_idct_float_3dnow) + +EXTN(jsimd_idct_float_3dnow): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [workspace] + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input, store into work array. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + lea edi, [workspace] ; FAST_FLOAT *wsptr + mov ecx, DCTSIZE/2 ; ctr + ALIGNX 16, 7 +.columnloop: +%ifndef NO_ZERO_COLUMN_TEST_FLOAT_3DNOW + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz short .columnDCT + + PUSHPIC ebx ; save GOT address + mov ebx, dword [DWBLOCK(3,0,esi,SIZEOF_JCOEF)] + mov eax, dword [DWBLOCK(4,0,esi,SIZEOF_JCOEF)] + or ebx, dword [DWBLOCK(5,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(6,0,esi,SIZEOF_JCOEF)] + or ebx, dword [DWBLOCK(7,0,esi,SIZEOF_JCOEF)] + or eax, ebx + POPPIC ebx ; restore GOT address + jnz short .columnDCT + + ; -- AC terms all zero + + movd mm0, dword [DWBLOCK(0,0,esi,SIZEOF_JCOEF)] + + punpcklwd mm0, mm0 + psrad mm0, (DWORD_BIT-WORD_BIT) + pi2fd mm0, mm0 + + pfmul mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movq mm1, mm0 + punpckldq mm0, mm0 + punpckhdq mm1, mm1 + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], mm0 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], mm0 + movq MMWORD [MMBLOCK(0,2,edi,SIZEOF_FAST_FLOAT)], mm0 + movq MMWORD [MMBLOCK(0,3,edi,SIZEOF_FAST_FLOAT)], mm0 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], mm1 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], mm1 + movq MMWORD [MMBLOCK(1,2,edi,SIZEOF_FAST_FLOAT)], mm1 + movq MMWORD [MMBLOCK(1,3,edi,SIZEOF_FAST_FLOAT)], mm1 + jmp near .nextcolumn + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Even part + + movd mm0, dword [DWBLOCK(0,0,esi,SIZEOF_JCOEF)] + movd mm1, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + movd mm2, dword [DWBLOCK(4,0,esi,SIZEOF_JCOEF)] + movd mm3, dword [DWBLOCK(6,0,esi,SIZEOF_JCOEF)] + + punpcklwd mm0, mm0 + punpcklwd mm1, mm1 + psrad mm0, (DWORD_BIT-WORD_BIT) + psrad mm1, (DWORD_BIT-WORD_BIT) + pi2fd mm0, mm0 + pi2fd mm1, mm1 + + pfmul mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + pfmul mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + punpcklwd mm2, mm2 + punpcklwd mm3, mm3 + psrad mm2, (DWORD_BIT-WORD_BIT) + psrad mm3, (DWORD_BIT-WORD_BIT) + pi2fd mm2, mm2 + pi2fd mm3, mm3 + + pfmul mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + pfmul mm3, MMWORD [MMBLOCK(6,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movq mm4, mm0 + movq mm5, mm1 + pfsub mm0, mm2 ; mm0=tmp11 + pfsub mm1, mm3 + pfadd mm4, mm2 ; mm4=tmp10 + pfadd mm5, mm3 ; mm5=tmp13 + + pfmul mm1, [GOTOFF(ebx,PD_1_414)] + pfsub mm1, mm5 ; mm1=tmp12 + + movq mm6, mm4 + movq mm7, mm0 + pfsub mm4, mm5 ; mm4=tmp3 + pfsub mm0, mm1 ; mm0=tmp2 + pfadd mm6, mm5 ; mm6=tmp0 + pfadd mm7, mm1 ; mm7=tmp1 + + movq MMWORD [wk(1)], mm4 ; tmp3 + movq MMWORD [wk(0)], mm0 ; tmp2 + + ; -- Odd part + + movd mm2, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + movd mm3, dword [DWBLOCK(3,0,esi,SIZEOF_JCOEF)] + movd mm5, dword [DWBLOCK(5,0,esi,SIZEOF_JCOEF)] + movd mm1, dword [DWBLOCK(7,0,esi,SIZEOF_JCOEF)] + + punpcklwd mm2, mm2 + punpcklwd mm3, mm3 + psrad mm2, (DWORD_BIT-WORD_BIT) + psrad mm3, (DWORD_BIT-WORD_BIT) + pi2fd mm2, mm2 + pi2fd mm3, mm3 + + pfmul mm2, MMWORD [MMBLOCK(1,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + pfmul mm3, MMWORD [MMBLOCK(3,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + punpcklwd mm5, mm5 + punpcklwd mm1, mm1 + psrad mm5, (DWORD_BIT-WORD_BIT) + psrad mm1, (DWORD_BIT-WORD_BIT) + pi2fd mm5, mm5 + pi2fd mm1, mm1 + + pfmul mm5, MMWORD [MMBLOCK(5,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + pfmul mm1, MMWORD [MMBLOCK(7,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movq mm4, mm2 + movq mm0, mm5 + pfadd mm2, mm1 ; mm2=z11 + pfadd mm5, mm3 ; mm5=z13 + pfsub mm4, mm1 ; mm4=z12 + pfsub mm0, mm3 ; mm0=z10 + + movq mm1, mm2 + pfsub mm2, mm5 + pfadd mm1, mm5 ; mm1=tmp7 + + pfmul mm2, [GOTOFF(ebx,PD_1_414)] ; mm2=tmp11 + + movq mm3, mm0 + pfadd mm0, mm4 + pfmul mm0, [GOTOFF(ebx,PD_1_847)] ; mm0=z5 + pfmul mm3, [GOTOFF(ebx,PD_2_613)] ; mm3=(z10 * 2.613125930) + pfmul mm4, [GOTOFF(ebx,PD_1_082)] ; mm4=(z12 * 1.082392200) + pfsubr mm3, mm0 ; mm3=tmp12 + pfsub mm4, mm0 ; mm4=tmp10 + + ; -- Final output stage + + pfsub mm3, mm1 ; mm3=tmp6 + movq mm5, mm6 + movq mm0, mm7 + pfadd mm6, mm1 ; mm6=data0=(00 01) + pfadd mm7, mm3 ; mm7=data1=(10 11) + pfsub mm5, mm1 ; mm5=data7=(70 71) + pfsub mm0, mm3 ; mm0=data6=(60 61) + pfsub mm2, mm3 ; mm2=tmp5 + + movq mm1, mm6 ; transpose coefficients + punpckldq mm6, mm7 ; mm6=(00 10) + punpckhdq mm1, mm7 ; mm1=(01 11) + movq mm3, mm0 ; transpose coefficients + punpckldq mm0, mm5 ; mm0=(60 70) + punpckhdq mm3, mm5 ; mm3=(61 71) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], mm6 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], mm1 + movq MMWORD [MMBLOCK(0,3,edi,SIZEOF_FAST_FLOAT)], mm0 + movq MMWORD [MMBLOCK(1,3,edi,SIZEOF_FAST_FLOAT)], mm3 + + movq mm7, MMWORD [wk(0)] ; mm7=tmp2 + movq mm5, MMWORD [wk(1)] ; mm5=tmp3 + + pfadd mm4, mm2 ; mm4=tmp4 + movq mm6, mm7 + movq mm1, mm5 + pfadd mm7, mm2 ; mm7=data2=(20 21) + pfadd mm5, mm4 ; mm5=data4=(40 41) + pfsub mm6, mm2 ; mm6=data5=(50 51) + pfsub mm1, mm4 ; mm1=data3=(30 31) + + movq mm0, mm7 ; transpose coefficients + punpckldq mm7, mm1 ; mm7=(20 30) + punpckhdq mm0, mm1 ; mm0=(21 31) + movq mm3, mm5 ; transpose coefficients + punpckldq mm5, mm6 ; mm5=(40 50) + punpckhdq mm3, mm6 ; mm3=(41 51) + + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], mm7 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], mm0 + movq MMWORD [MMBLOCK(0,2,edi,SIZEOF_FAST_FLOAT)], mm5 + movq MMWORD [MMBLOCK(1,2,edi,SIZEOF_FAST_FLOAT)], mm3 + +.nextcolumn: + add esi, byte 2*SIZEOF_JCOEF ; coef_block + add edx, byte 2*SIZEOF_FLOAT_MULT_TYPE ; quantptr + add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr + dec ecx ; ctr + jnz near .columnloop + + ; -- Prefetch the next coefficient block + + prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32] + prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32] + prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32] + prefetch [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + lea esi, [workspace] ; FAST_FLOAT *wsptr + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + mov ecx, DCTSIZE/2 ; ctr + ALIGNX 16, 7 +.rowloop: + + ; -- Even part + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_FAST_FLOAT)] + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_FAST_FLOAT)] + + movq mm4, mm0 + movq mm5, mm1 + pfsub mm0, mm2 ; mm0=tmp11 + pfsub mm1, mm3 + pfadd mm4, mm2 ; mm4=tmp10 + pfadd mm5, mm3 ; mm5=tmp13 + + pfmul mm1, [GOTOFF(ebx,PD_1_414)] + pfsub mm1, mm5 ; mm1=tmp12 + + movq mm6, mm4 + movq mm7, mm0 + pfsub mm4, mm5 ; mm4=tmp3 + pfsub mm0, mm1 ; mm0=tmp2 + pfadd mm6, mm5 ; mm6=tmp0 + pfadd mm7, mm1 ; mm7=tmp1 + + movq MMWORD [wk(1)], mm4 ; tmp3 + movq MMWORD [wk(0)], mm0 ; tmp2 + + ; -- Odd part + + movq mm2, MMWORD [MMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(3,0,esi,SIZEOF_FAST_FLOAT)] + movq mm5, MMWORD [MMBLOCK(5,0,esi,SIZEOF_FAST_FLOAT)] + movq mm1, MMWORD [MMBLOCK(7,0,esi,SIZEOF_FAST_FLOAT)] + + movq mm4, mm2 + movq mm0, mm5 + pfadd mm2, mm1 ; mm2=z11 + pfadd mm5, mm3 ; mm5=z13 + pfsub mm4, mm1 ; mm4=z12 + pfsub mm0, mm3 ; mm0=z10 + + movq mm1, mm2 + pfsub mm2, mm5 + pfadd mm1, mm5 ; mm1=tmp7 + + pfmul mm2, [GOTOFF(ebx,PD_1_414)] ; mm2=tmp11 + + movq mm3, mm0 + pfadd mm0, mm4 + pfmul mm0, [GOTOFF(ebx,PD_1_847)] ; mm0=z5 + pfmul mm3, [GOTOFF(ebx,PD_2_613)] ; mm3=(z10 * 2.613125930) + pfmul mm4, [GOTOFF(ebx,PD_1_082)] ; mm4=(z12 * 1.082392200) + pfsubr mm3, mm0 ; mm3=tmp12 + pfsub mm4, mm0 ; mm4=tmp10 + + ; -- Final output stage + + pfsub mm3, mm1 ; mm3=tmp6 + movq mm5, mm6 + movq mm0, mm7 + pfadd mm6, mm1 ; mm6=data0=(00 10) + pfadd mm7, mm3 ; mm7=data1=(01 11) + pfsub mm5, mm1 ; mm5=data7=(07 17) + pfsub mm0, mm3 ; mm0=data6=(06 16) + pfsub mm2, mm3 ; mm2=tmp5 + + movq mm1, [GOTOFF(ebx,PD_RNDINT_MAGIC)] ; mm1=[PD_RNDINT_MAGIC] + pcmpeqd mm3, mm3 + psrld mm3, WORD_BIT ; mm3={0xFFFF 0x0000 0xFFFF 0x0000} + + pfadd mm6, mm1 ; mm6=roundint(data0/8)=(00 ** 10 **) + pfadd mm7, mm1 ; mm7=roundint(data1/8)=(01 ** 11 **) + pfadd mm0, mm1 ; mm0=roundint(data6/8)=(06 ** 16 **) + pfadd mm5, mm1 ; mm5=roundint(data7/8)=(07 ** 17 **) + + pand mm6, mm3 ; mm6=(00 -- 10 --) + pslld mm7, WORD_BIT ; mm7=(-- 01 -- 11) + pand mm0, mm3 ; mm0=(06 -- 16 --) + pslld mm5, WORD_BIT ; mm5=(-- 07 -- 17) + por mm6, mm7 ; mm6=(00 01 10 11) + por mm0, mm5 ; mm0=(06 07 16 17) + + movq mm1, MMWORD [wk(0)] ; mm1=tmp2 + movq mm3, MMWORD [wk(1)] ; mm3=tmp3 + + pfadd mm4, mm2 ; mm4=tmp4 + movq mm7, mm1 + movq mm5, mm3 + pfadd mm1, mm2 ; mm1=data2=(02 12) + pfadd mm3, mm4 ; mm3=data4=(04 14) + pfsub mm7, mm2 ; mm7=data5=(05 15) + pfsub mm5, mm4 ; mm5=data3=(03 13) + + movq mm2, [GOTOFF(ebx,PD_RNDINT_MAGIC)] ; mm2=[PD_RNDINT_MAGIC] + pcmpeqd mm4, mm4 + psrld mm4, WORD_BIT ; mm4={0xFFFF 0x0000 0xFFFF 0x0000} + + pfadd mm3, mm2 ; mm3=roundint(data4/8)=(04 ** 14 **) + pfadd mm7, mm2 ; mm7=roundint(data5/8)=(05 ** 15 **) + pfadd mm1, mm2 ; mm1=roundint(data2/8)=(02 ** 12 **) + pfadd mm5, mm2 ; mm5=roundint(data3/8)=(03 ** 13 **) + + pand mm3, mm4 ; mm3=(04 -- 14 --) + pslld mm7, WORD_BIT ; mm7=(-- 05 -- 15) + pand mm1, mm4 ; mm1=(02 -- 12 --) + pslld mm5, WORD_BIT ; mm5=(-- 03 -- 13) + por mm3, mm7 ; mm3=(04 05 14 15) + por mm1, mm5 ; mm1=(02 03 12 13) + + movq mm2, [GOTOFF(ebx,PB_CENTERJSAMP)] ; mm2=[PB_CENTERJSAMP] + + packsswb mm6, mm3 ; mm6=(00 01 10 11 04 05 14 15) + packsswb mm1, mm0 ; mm1=(02 03 12 13 06 07 16 17) + paddb mm6, mm2 + paddb mm1, mm2 + + movq mm4, mm6 ; transpose coefficients(phase 2) + punpcklwd mm6, mm1 ; mm6=(00 01 02 03 10 11 12 13) + punpckhwd mm4, mm1 ; mm4=(04 05 06 07 14 15 16 17) + + movq mm7, mm6 ; transpose coefficients(phase 3) + punpckldq mm6, mm4 ; mm6=(00 01 02 03 04 05 06 07) + punpckhdq mm7, mm4 ; mm7=(10 11 12 13 14 15 16 17) + + PUSHPIC ebx ; save GOT address + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm6 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm7 + + POPPIC ebx ; restore GOT address + + add esi, byte 2*SIZEOF_FAST_FLOAT ; wsptr + add edi, byte 2*SIZEOF_JSAMPROW + dec ecx ; ctr + jnz near .rowloop + + femms ; empty MMX/3DNow! state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse.asm new file mode 100644 index 0000000..caf636b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse.asm @@ -0,0 +1,567 @@ +; +; jidctflt.asm - floating-point IDCT (SSE & MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a floating-point implementation of the inverse DCT +; (Discrete Cosine Transform). The following code is based directly on +; the IJG's original jidctflt.c; see the jidctflt.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%macro UNPCKLPS2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5) + shufps %1, %2, 0x44 +%endmacro + +%macro UNPCKHPS2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7) + shufps %1, %2, 0xEE +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_float_sse) + +EXTN(jconst_idct_float_sse): + +PD_1_414 times 4 dd 1.414213562373095048801689 +PD_1_847 times 4 dd 1.847759065022573512256366 +PD_1_082 times 4 dd 1.082392200292393968799446 +PD_M2_613 times 4 dd -2.613125929752753055713286 +PD_0_125 times 4 dd 0.125 ; 1/8 +PB_CENTERJSAMP times 8 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_float_sse(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; void *dct_table +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 +%define workspace wk(0) - DCTSIZE2 * SIZEOF_FAST_FLOAT + ; FAST_FLOAT workspace[DCTSIZE2] + + align 32 + GLOBAL_FUNCTION(jsimd_idct_float_sse) + +EXTN(jsimd_idct_float_sse): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [workspace] + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input, store into work array. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + lea edi, [workspace] ; FAST_FLOAT *wsptr + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.columnloop: +%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por mm1, mm0 + packsswb mm1, mm1 + movd eax, mm1 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + + punpckhwd mm1, mm0 ; mm1=(** 02 ** 03) + punpcklwd mm0, mm0 ; mm0=(00 00 01 01) + psrad mm1, (DWORD_BIT-WORD_BIT) ; mm1=in0H=(02 03) + psrad mm0, (DWORD_BIT-WORD_BIT) ; mm0=in0L=(00 01) + cvtpi2ps xmm3, mm1 ; xmm3=(02 03 ** **) + cvtpi2ps xmm0, mm0 ; xmm0=(00 01 ** **) + movlhps xmm0, xmm3 ; xmm0=in0=(00 01 02 03) + + mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm1, xmm0 + movaps xmm2, xmm0 + movaps xmm3, xmm0 + + shufps xmm0, xmm0, 0x00 ; xmm0=(00 00 00 00) + shufps xmm1, xmm1, 0x55 ; xmm1=(01 01 01 01) + shufps xmm2, xmm2, 0xAA ; xmm2=(02 02 02 02) + shufps xmm3, xmm3, 0xFF ; xmm3=(03 03 03 03) + + movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3 + jmp near .nextcolumn + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Even part + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + + punpckhwd mm4, mm0 ; mm4=(** 02 ** 03) + punpcklwd mm0, mm0 ; mm0=(00 00 01 01) + punpckhwd mm5, mm1 ; mm5=(** 22 ** 23) + punpcklwd mm1, mm1 ; mm1=(20 20 21 21) + + psrad mm4, (DWORD_BIT-WORD_BIT) ; mm4=in0H=(02 03) + psrad mm0, (DWORD_BIT-WORD_BIT) ; mm0=in0L=(00 01) + cvtpi2ps xmm4, mm4 ; xmm4=(02 03 ** **) + cvtpi2ps xmm0, mm0 ; xmm0=(00 01 ** **) + psrad mm5, (DWORD_BIT-WORD_BIT) ; mm5=in2H=(22 23) + psrad mm1, (DWORD_BIT-WORD_BIT) ; mm1=in2L=(20 21) + cvtpi2ps xmm5, mm5 ; xmm5=(22 23 ** **) + cvtpi2ps xmm1, mm1 ; xmm1=(20 21 ** **) + + punpckhwd mm6, mm2 ; mm6=(** 42 ** 43) + punpcklwd mm2, mm2 ; mm2=(40 40 41 41) + punpckhwd mm7, mm3 ; mm7=(** 62 ** 63) + punpcklwd mm3, mm3 ; mm3=(60 60 61 61) + + psrad mm6, (DWORD_BIT-WORD_BIT) ; mm6=in4H=(42 43) + psrad mm2, (DWORD_BIT-WORD_BIT) ; mm2=in4L=(40 41) + cvtpi2ps xmm6, mm6 ; xmm6=(42 43 ** **) + cvtpi2ps xmm2, mm2 ; xmm2=(40 41 ** **) + psrad mm7, (DWORD_BIT-WORD_BIT) ; mm7=in6H=(62 63) + psrad mm3, (DWORD_BIT-WORD_BIT) ; mm3=in6L=(60 61) + cvtpi2ps xmm7, mm7 ; xmm7=(62 63 ** **) + cvtpi2ps xmm3, mm3 ; xmm3=(60 61 ** **) + + movlhps xmm0, xmm4 ; xmm0=in0=(00 01 02 03) + movlhps xmm1, xmm5 ; xmm1=in2=(20 21 22 23) + mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movlhps xmm2, xmm6 ; xmm2=in4=(40 41 42 43) + movlhps xmm3, xmm7 ; xmm3=in6=(60 61 62 63) + mulps xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm4, xmm0 + movaps xmm5, xmm1 + subps xmm0, xmm2 ; xmm0=tmp11 + subps xmm1, xmm3 + addps xmm4, xmm2 ; xmm4=tmp10 + addps xmm5, xmm3 ; xmm5=tmp13 + + mulps xmm1, [GOTOFF(ebx,PD_1_414)] + subps xmm1, xmm5 ; xmm1=tmp12 + + movaps xmm6, xmm4 + movaps xmm7, xmm0 + subps xmm4, xmm5 ; xmm4=tmp3 + subps xmm0, xmm1 ; xmm0=tmp2 + addps xmm6, xmm5 ; xmm6=tmp0 + addps xmm7, xmm1 ; xmm7=tmp1 + + movaps XMMWORD [wk(1)], xmm4 ; tmp3 + movaps XMMWORD [wk(0)], xmm0 ; tmp2 + + ; -- Odd part + + movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + movq mm5, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + + punpckhwd mm6, mm4 ; mm6=(** 12 ** 13) + punpcklwd mm4, mm4 ; mm4=(10 10 11 11) + punpckhwd mm2, mm0 ; mm2=(** 32 ** 33) + punpcklwd mm0, mm0 ; mm0=(30 30 31 31) + + psrad mm6, (DWORD_BIT-WORD_BIT) ; mm6=in1H=(12 13) + psrad mm4, (DWORD_BIT-WORD_BIT) ; mm4=in1L=(10 11) + cvtpi2ps xmm4, mm6 ; xmm4=(12 13 ** **) + cvtpi2ps xmm2, mm4 ; xmm2=(10 11 ** **) + psrad mm2, (DWORD_BIT-WORD_BIT) ; mm2=in3H=(32 33) + psrad mm0, (DWORD_BIT-WORD_BIT) ; mm0=in3L=(30 31) + cvtpi2ps xmm0, mm2 ; xmm0=(32 33 ** **) + cvtpi2ps xmm3, mm0 ; xmm3=(30 31 ** **) + + punpckhwd mm7, mm5 ; mm7=(** 52 ** 53) + punpcklwd mm5, mm5 ; mm5=(50 50 51 51) + punpckhwd mm3, mm1 ; mm3=(** 72 ** 73) + punpcklwd mm1, mm1 ; mm1=(70 70 71 71) + + movlhps xmm2, xmm4 ; xmm2=in1=(10 11 12 13) + movlhps xmm3, xmm0 ; xmm3=in3=(30 31 32 33) + + psrad mm7, (DWORD_BIT-WORD_BIT) ; mm7=in5H=(52 53) + psrad mm5, (DWORD_BIT-WORD_BIT) ; mm5=in5L=(50 51) + cvtpi2ps xmm4, mm7 ; xmm4=(52 53 ** **) + cvtpi2ps xmm5, mm5 ; xmm5=(50 51 ** **) + psrad mm3, (DWORD_BIT-WORD_BIT) ; mm3=in7H=(72 73) + psrad mm1, (DWORD_BIT-WORD_BIT) ; mm1=in7L=(70 71) + cvtpi2ps xmm0, mm3 ; xmm0=(72 73 ** **) + cvtpi2ps xmm1, mm1 ; xmm1=(70 71 ** **) + + mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movlhps xmm5, xmm4 ; xmm5=in5=(50 51 52 53) + movlhps xmm1, xmm0 ; xmm1=in7=(70 71 72 73) + mulps xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm4, xmm2 + movaps xmm0, xmm5 + addps xmm2, xmm1 ; xmm2=z11 + addps xmm5, xmm3 ; xmm5=z13 + subps xmm4, xmm1 ; xmm4=z12 + subps xmm0, xmm3 ; xmm0=z10 + + movaps xmm1, xmm2 + subps xmm2, xmm5 + addps xmm1, xmm5 ; xmm1=tmp7 + + mulps xmm2, [GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11 + + movaps xmm3, xmm0 + addps xmm0, xmm4 + mulps xmm0, [GOTOFF(ebx,PD_1_847)] ; xmm0=z5 + mulps xmm3, [GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930) + mulps xmm4, [GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200) + addps xmm3, xmm0 ; xmm3=tmp12 + subps xmm4, xmm0 ; xmm4=tmp10 + + ; -- Final output stage + + subps xmm3, xmm1 ; xmm3=tmp6 + movaps xmm5, xmm6 + movaps xmm0, xmm7 + addps xmm6, xmm1 ; xmm6=data0=(00 01 02 03) + addps xmm7, xmm3 ; xmm7=data1=(10 11 12 13) + subps xmm5, xmm1 ; xmm5=data7=(70 71 72 73) + subps xmm0, xmm3 ; xmm0=data6=(60 61 62 63) + subps xmm2, xmm3 ; xmm2=tmp5 + + movaps xmm1, xmm6 ; transpose coefficients(phase 1) + unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11) + unpckhps xmm1, xmm7 ; xmm1=(02 12 03 13) + movaps xmm3, xmm0 ; transpose coefficients(phase 1) + unpcklps xmm0, xmm5 ; xmm0=(60 70 61 71) + unpckhps xmm3, xmm5 ; xmm3=(62 72 63 73) + + movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2 + movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3 + + movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71) + movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73) + + addps xmm4, xmm2 ; xmm4=tmp4 + movaps xmm0, xmm7 + movaps xmm3, xmm5 + addps xmm7, xmm2 ; xmm7=data2=(20 21 22 23) + addps xmm5, xmm4 ; xmm5=data4=(40 41 42 43) + subps xmm0, xmm2 ; xmm0=data5=(50 51 52 53) + subps xmm3, xmm4 ; xmm3=data3=(30 31 32 33) + + movaps xmm2, xmm7 ; transpose coefficients(phase 1) + unpcklps xmm7, xmm3 ; xmm7=(20 30 21 31) + unpckhps xmm2, xmm3 ; xmm2=(22 32 23 33) + movaps xmm4, xmm5 ; transpose coefficients(phase 1) + unpcklps xmm5, xmm0 ; xmm5=(40 50 41 51) + unpckhps xmm4, xmm0 ; xmm4=(42 52 43 53) + + movaps xmm3, xmm6 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm6, xmm7 ; xmm6=(00 10 20 30) + UNPCKHPS2 xmm3, xmm7 ; xmm3=(01 11 21 31) + movaps xmm0, xmm1 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm1, xmm2 ; xmm1=(02 12 22 32) + UNPCKHPS2 xmm0, xmm2 ; xmm0=(03 13 23 33) + + movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71) + movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73) + + movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm6 + movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm0 + + movaps xmm6, xmm5 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm5, xmm7 ; xmm5=(40 50 60 70) + UNPCKHPS2 xmm6, xmm7 ; xmm6=(41 51 61 71) + movaps xmm3, xmm4 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm4, xmm2 ; xmm4=(42 52 62 72) + UNPCKHPS2 xmm3, xmm2 ; xmm3=(43 53 63 73) + + movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm6 + movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm4 + movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3 + +.nextcolumn: + add esi, byte 4*SIZEOF_JCOEF ; coef_block + add edx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr + add edi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr + dec ecx ; ctr + jnz near .columnloop + + ; -- Prefetch the next coefficient block + + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32] + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32] + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32] + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + lea esi, [workspace] ; FAST_FLOAT *wsptr + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.rowloop: + + ; -- Even part + + movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_FAST_FLOAT)] + + movaps xmm4, xmm0 + movaps xmm5, xmm1 + subps xmm0, xmm2 ; xmm0=tmp11 + subps xmm1, xmm3 + addps xmm4, xmm2 ; xmm4=tmp10 + addps xmm5, xmm3 ; xmm5=tmp13 + + mulps xmm1, [GOTOFF(ebx,PD_1_414)] + subps xmm1, xmm5 ; xmm1=tmp12 + + movaps xmm6, xmm4 + movaps xmm7, xmm0 + subps xmm4, xmm5 ; xmm4=tmp3 + subps xmm0, xmm1 ; xmm0=tmp2 + addps xmm6, xmm5 ; xmm6=tmp0 + addps xmm7, xmm1 ; xmm7=tmp1 + + movaps XMMWORD [wk(1)], xmm4 ; tmp3 + movaps XMMWORD [wk(0)], xmm0 ; tmp2 + + ; -- Odd part + + movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_FAST_FLOAT)] + + movaps xmm4, xmm2 + movaps xmm0, xmm5 + addps xmm2, xmm1 ; xmm2=z11 + addps xmm5, xmm3 ; xmm5=z13 + subps xmm4, xmm1 ; xmm4=z12 + subps xmm0, xmm3 ; xmm0=z10 + + movaps xmm1, xmm2 + subps xmm2, xmm5 + addps xmm1, xmm5 ; xmm1=tmp7 + + mulps xmm2, [GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11 + + movaps xmm3, xmm0 + addps xmm0, xmm4 + mulps xmm0, [GOTOFF(ebx,PD_1_847)] ; xmm0=z5 + mulps xmm3, [GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930) + mulps xmm4, [GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200) + addps xmm3, xmm0 ; xmm3=tmp12 + subps xmm4, xmm0 ; xmm4=tmp10 + + ; -- Final output stage + + subps xmm3, xmm1 ; xmm3=tmp6 + movaps xmm5, xmm6 + movaps xmm0, xmm7 + addps xmm6, xmm1 ; xmm6=data0=(00 10 20 30) + addps xmm7, xmm3 ; xmm7=data1=(01 11 21 31) + subps xmm5, xmm1 ; xmm5=data7=(07 17 27 37) + subps xmm0, xmm3 ; xmm0=data6=(06 16 26 36) + subps xmm2, xmm3 ; xmm2=tmp5 + + movaps xmm1, [GOTOFF(ebx,PD_0_125)] ; xmm1=[PD_0_125] + + mulps xmm6, xmm1 ; descale(1/8) + mulps xmm7, xmm1 ; descale(1/8) + mulps xmm5, xmm1 ; descale(1/8) + mulps xmm0, xmm1 ; descale(1/8) + + movhlps xmm3, xmm6 + movhlps xmm1, xmm7 + cvtps2pi mm0, xmm6 ; round to int32, mm0=data0L=(00 10) + cvtps2pi mm1, xmm7 ; round to int32, mm1=data1L=(01 11) + cvtps2pi mm2, xmm3 ; round to int32, mm2=data0H=(20 30) + cvtps2pi mm3, xmm1 ; round to int32, mm3=data1H=(21 31) + packssdw mm0, mm2 ; mm0=data0=(00 10 20 30) + packssdw mm1, mm3 ; mm1=data1=(01 11 21 31) + + movhlps xmm6, xmm5 + movhlps xmm7, xmm0 + cvtps2pi mm4, xmm5 ; round to int32, mm4=data7L=(07 17) + cvtps2pi mm5, xmm0 ; round to int32, mm5=data6L=(06 16) + cvtps2pi mm6, xmm6 ; round to int32, mm6=data7H=(27 37) + cvtps2pi mm7, xmm7 ; round to int32, mm7=data6H=(26 36) + packssdw mm4, mm6 ; mm4=data7=(07 17 27 37) + packssdw mm5, mm7 ; mm5=data6=(06 16 26 36) + + packsswb mm0, mm5 ; mm0=(00 10 20 30 06 16 26 36) + packsswb mm1, mm4 ; mm1=(01 11 21 31 07 17 27 37) + + movaps xmm3, XMMWORD [wk(0)] ; xmm3=tmp2 + movaps xmm1, XMMWORD [wk(1)] ; xmm1=tmp3 + + movaps xmm6, [GOTOFF(ebx,PD_0_125)] ; xmm6=[PD_0_125] + + addps xmm4, xmm2 ; xmm4=tmp4 + movaps xmm5, xmm3 + movaps xmm0, xmm1 + addps xmm3, xmm2 ; xmm3=data2=(02 12 22 32) + addps xmm1, xmm4 ; xmm1=data4=(04 14 24 34) + subps xmm5, xmm2 ; xmm5=data5=(05 15 25 35) + subps xmm0, xmm4 ; xmm0=data3=(03 13 23 33) + + mulps xmm3, xmm6 ; descale(1/8) + mulps xmm1, xmm6 ; descale(1/8) + mulps xmm5, xmm6 ; descale(1/8) + mulps xmm0, xmm6 ; descale(1/8) + + movhlps xmm7, xmm3 + movhlps xmm2, xmm1 + cvtps2pi mm2, xmm3 ; round to int32, mm2=data2L=(02 12) + cvtps2pi mm3, xmm1 ; round to int32, mm3=data4L=(04 14) + cvtps2pi mm6, xmm7 ; round to int32, mm6=data2H=(22 32) + cvtps2pi mm7, xmm2 ; round to int32, mm7=data4H=(24 34) + packssdw mm2, mm6 ; mm2=data2=(02 12 22 32) + packssdw mm3, mm7 ; mm3=data4=(04 14 24 34) + + movhlps xmm4, xmm5 + movhlps xmm6, xmm0 + cvtps2pi mm5, xmm5 ; round to int32, mm5=data5L=(05 15) + cvtps2pi mm4, xmm0 ; round to int32, mm4=data3L=(03 13) + cvtps2pi mm6, xmm4 ; round to int32, mm6=data5H=(25 35) + cvtps2pi mm7, xmm6 ; round to int32, mm7=data3H=(23 33) + packssdw mm5, mm6 ; mm5=data5=(05 15 25 35) + packssdw mm4, mm7 ; mm4=data3=(03 13 23 33) + + movq mm6, [GOTOFF(ebx,PB_CENTERJSAMP)] ; mm6=[PB_CENTERJSAMP] + + packsswb mm2, mm3 ; mm2=(02 12 22 32 04 14 24 34) + packsswb mm4, mm5 ; mm4=(03 13 23 33 05 15 25 35) + + paddb mm0, mm6 + paddb mm1, mm6 + paddb mm2, mm6 + paddb mm4, mm6 + + movq mm7, mm0 ; transpose coefficients(phase 1) + punpcklbw mm0, mm1 ; mm0=(00 01 10 11 20 21 30 31) + punpckhbw mm7, mm1 ; mm7=(06 07 16 17 26 27 36 37) + movq mm3, mm2 ; transpose coefficients(phase 1) + punpcklbw mm2, mm4 ; mm2=(02 03 12 13 22 23 32 33) + punpckhbw mm3, mm4 ; mm3=(04 05 14 15 24 25 34 35) + + movq mm5, mm0 ; transpose coefficients(phase 2) + punpcklwd mm0, mm2 ; mm0=(00 01 02 03 10 11 12 13) + punpckhwd mm5, mm2 ; mm5=(20 21 22 23 30 31 32 33) + movq mm6, mm3 ; transpose coefficients(phase 2) + punpcklwd mm3, mm7 ; mm3=(04 05 06 07 14 15 16 17) + punpckhwd mm6, mm7 ; mm6=(24 25 26 27 34 35 36 37) + + movq mm1, mm0 ; transpose coefficients(phase 3) + punpckldq mm0, mm3 ; mm0=(00 01 02 03 04 05 06 07) + punpckhdq mm1, mm3 ; mm1=(10 11 12 13 14 15 16 17) + movq mm4, mm5 ; transpose coefficients(phase 3) + punpckldq mm5, mm6 ; mm5=(20 21 22 23 24 25 26 27) + punpckhdq mm4, mm6 ; mm4=(30 31 32 33 34 35 36 37) + + PUSHPIC ebx ; save GOT address + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm0 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm1 + mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm5 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm4 + + POPPIC ebx ; restore GOT address + + add esi, byte 4*SIZEOF_FAST_FLOAT ; wsptr + add edi, byte 4*SIZEOF_JSAMPROW + dec ecx ; ctr + jnz near .rowloop + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse2.asm new file mode 100644 index 0000000..42703a8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctflt-sse2.asm @@ -0,0 +1,493 @@ +; +; jidctflt.asm - floating-point IDCT (SSE & SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a floating-point implementation of the inverse DCT +; (Discrete Cosine Transform). The following code is based directly on +; the IJG's original jidctflt.c; see the jidctflt.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%macro UNPCKLPS2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5) + shufps %1, %2, 0x44 +%endmacro + +%macro UNPCKHPS2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7) + shufps %1, %2, 0xEE +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_float_sse2) + +EXTN(jconst_idct_float_sse2): + +PD_1_414 times 4 dd 1.414213562373095048801689 +PD_1_847 times 4 dd 1.847759065022573512256366 +PD_1_082 times 4 dd 1.082392200292393968799446 +PD_M2_613 times 4 dd -2.613125929752753055713286 +PD_RNDINT_MAGIC times 4 dd 100663296.0 ; (float)(0x00C00000 << 3) +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_float_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; void *dct_table +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 +%define workspace wk(0) - DCTSIZE2 * SIZEOF_FAST_FLOAT + ; FAST_FLOAT workspace[DCTSIZE2] + + align 32 + GLOBAL_FUNCTION(jsimd_idct_float_sse2) + +EXTN(jsimd_idct_float_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [workspace] + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input, store into work array. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + lea edi, [workspace] ; FAST_FLOAT *wsptr + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.columnloop: +%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movq xmm1, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq xmm2, XMM_MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movq xmm3, XMM_MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + movq xmm4, XMM_MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movq xmm5, XMM_MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq xmm6, XMM_MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + movq xmm7, XMM_MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por xmm1, xmm2 + por xmm3, xmm4 + por xmm5, xmm6 + por xmm1, xmm3 + por xmm5, xmm7 + por xmm1, xmm5 + packsswb xmm1, xmm1 + movd eax, xmm1 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movq xmm0, XMM_MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03) + cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03) + + mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm1, xmm0 + movaps xmm2, xmm0 + movaps xmm3, xmm0 + + shufps xmm0, xmm0, 0x00 ; xmm0=(00 00 00 00) + shufps xmm1, xmm1, 0x55 ; xmm1=(01 01 01 01) + shufps xmm2, xmm2, 0xAA ; xmm2=(02 02 02 02) + shufps xmm3, xmm3, 0xFF ; xmm3=(03 03 03 03) + + movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3 + jmp near .nextcolumn + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Even part + + movq xmm0, XMM_MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq xmm1, XMM_MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movq xmm2, XMM_MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movq xmm3, XMM_MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + punpcklwd xmm1, xmm1 ; xmm1=(20 20 21 21 22 22 23 23) + psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03) + psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in2=(20 21 22 23) + cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03) + cvtdq2ps xmm1, xmm1 ; xmm1=in2=(20 21 22 23) + + punpcklwd xmm2, xmm2 ; xmm2=(40 40 41 41 42 42 43 43) + punpcklwd xmm3, xmm3 ; xmm3=(60 60 61 61 62 62 63 63) + psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in4=(40 41 42 43) + psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in6=(60 61 62 63) + cvtdq2ps xmm2, xmm2 ; xmm2=in4=(40 41 42 43) + cvtdq2ps xmm3, xmm3 ; xmm3=in6=(60 61 62 63) + + mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm4, xmm0 + movaps xmm5, xmm1 + subps xmm0, xmm2 ; xmm0=tmp11 + subps xmm1, xmm3 + addps xmm4, xmm2 ; xmm4=tmp10 + addps xmm5, xmm3 ; xmm5=tmp13 + + mulps xmm1, [GOTOFF(ebx,PD_1_414)] + subps xmm1, xmm5 ; xmm1=tmp12 + + movaps xmm6, xmm4 + movaps xmm7, xmm0 + subps xmm4, xmm5 ; xmm4=tmp3 + subps xmm0, xmm1 ; xmm0=tmp2 + addps xmm6, xmm5 ; xmm6=tmp0 + addps xmm7, xmm1 ; xmm7=tmp1 + + movaps XMMWORD [wk(1)], xmm4 ; tmp3 + movaps XMMWORD [wk(0)], xmm0 ; tmp2 + + ; -- Odd part + + movq xmm2, XMM_MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq xmm3, XMM_MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + movq xmm5, XMM_MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq xmm1, XMM_MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + + punpcklwd xmm2, xmm2 ; xmm2=(10 10 11 11 12 12 13 13) + punpcklwd xmm3, xmm3 ; xmm3=(30 30 31 31 32 32 33 33) + psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in1=(10 11 12 13) + psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in3=(30 31 32 33) + cvtdq2ps xmm2, xmm2 ; xmm2=in1=(10 11 12 13) + cvtdq2ps xmm3, xmm3 ; xmm3=in3=(30 31 32 33) + + punpcklwd xmm5, xmm5 ; xmm5=(50 50 51 51 52 52 53 53) + punpcklwd xmm1, xmm1 ; xmm1=(70 70 71 71 72 72 73 73) + psrad xmm5, (DWORD_BIT-WORD_BIT) ; xmm5=in5=(50 51 52 53) + psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in7=(70 71 72 73) + cvtdq2ps xmm5, xmm5 ; xmm5=in5=(50 51 52 53) + cvtdq2ps xmm1, xmm1 ; xmm1=in7=(70 71 72 73) + + mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm4, xmm2 + movaps xmm0, xmm5 + addps xmm2, xmm1 ; xmm2=z11 + addps xmm5, xmm3 ; xmm5=z13 + subps xmm4, xmm1 ; xmm4=z12 + subps xmm0, xmm3 ; xmm0=z10 + + movaps xmm1, xmm2 + subps xmm2, xmm5 + addps xmm1, xmm5 ; xmm1=tmp7 + + mulps xmm2, [GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11 + + movaps xmm3, xmm0 + addps xmm0, xmm4 + mulps xmm0, [GOTOFF(ebx,PD_1_847)] ; xmm0=z5 + mulps xmm3, [GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930) + mulps xmm4, [GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200) + addps xmm3, xmm0 ; xmm3=tmp12 + subps xmm4, xmm0 ; xmm4=tmp10 + + ; -- Final output stage + + subps xmm3, xmm1 ; xmm3=tmp6 + movaps xmm5, xmm6 + movaps xmm0, xmm7 + addps xmm6, xmm1 ; xmm6=data0=(00 01 02 03) + addps xmm7, xmm3 ; xmm7=data1=(10 11 12 13) + subps xmm5, xmm1 ; xmm5=data7=(70 71 72 73) + subps xmm0, xmm3 ; xmm0=data6=(60 61 62 63) + subps xmm2, xmm3 ; xmm2=tmp5 + + movaps xmm1, xmm6 ; transpose coefficients(phase 1) + unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11) + unpckhps xmm1, xmm7 ; xmm1=(02 12 03 13) + movaps xmm3, xmm0 ; transpose coefficients(phase 1) + unpcklps xmm0, xmm5 ; xmm0=(60 70 61 71) + unpckhps xmm3, xmm5 ; xmm3=(62 72 63 73) + + movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2 + movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3 + + movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71) + movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73) + + addps xmm4, xmm2 ; xmm4=tmp4 + movaps xmm0, xmm7 + movaps xmm3, xmm5 + addps xmm7, xmm2 ; xmm7=data2=(20 21 22 23) + addps xmm5, xmm4 ; xmm5=data4=(40 41 42 43) + subps xmm0, xmm2 ; xmm0=data5=(50 51 52 53) + subps xmm3, xmm4 ; xmm3=data3=(30 31 32 33) + + movaps xmm2, xmm7 ; transpose coefficients(phase 1) + unpcklps xmm7, xmm3 ; xmm7=(20 30 21 31) + unpckhps xmm2, xmm3 ; xmm2=(22 32 23 33) + movaps xmm4, xmm5 ; transpose coefficients(phase 1) + unpcklps xmm5, xmm0 ; xmm5=(40 50 41 51) + unpckhps xmm4, xmm0 ; xmm4=(42 52 43 53) + + movaps xmm3, xmm6 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm6, xmm7 ; xmm6=(00 10 20 30) + UNPCKHPS2 xmm3, xmm7 ; xmm3=(01 11 21 31) + movaps xmm0, xmm1 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm1, xmm2 ; xmm1=(02 12 22 32) + UNPCKHPS2 xmm0, xmm2 ; xmm0=(03 13 23 33) + + movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71) + movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73) + + movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm6 + movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_FAST_FLOAT)], xmm0 + + movaps xmm6, xmm5 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm5, xmm7 ; xmm5=(40 50 60 70) + UNPCKHPS2 xmm6, xmm7 ; xmm6=(41 51 61 71) + movaps xmm3, xmm4 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm4, xmm2 ; xmm4=(42 52 62 72) + UNPCKHPS2 xmm3, xmm2 ; xmm3=(43 53 63 73) + + movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm6 + movaps XMMWORD [XMMBLOCK(2,1,edi,SIZEOF_FAST_FLOAT)], xmm4 + movaps XMMWORD [XMMBLOCK(3,1,edi,SIZEOF_FAST_FLOAT)], xmm3 + +.nextcolumn: + add esi, byte 4*SIZEOF_JCOEF ; coef_block + add edx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr + add edi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr + dec ecx ; ctr + jnz near .columnloop + + ; -- Prefetch the next coefficient block + + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32] + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32] + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32] + prefetchnta [esi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + lea esi, [workspace] ; FAST_FLOAT *wsptr + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.rowloop: + + ; -- Even part + + movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_FAST_FLOAT)] + + movaps xmm4, xmm0 + movaps xmm5, xmm1 + subps xmm0, xmm2 ; xmm0=tmp11 + subps xmm1, xmm3 + addps xmm4, xmm2 ; xmm4=tmp10 + addps xmm5, xmm3 ; xmm5=tmp13 + + mulps xmm1, [GOTOFF(ebx,PD_1_414)] + subps xmm1, xmm5 ; xmm1=tmp12 + + movaps xmm6, xmm4 + movaps xmm7, xmm0 + subps xmm4, xmm5 ; xmm4=tmp3 + subps xmm0, xmm1 ; xmm0=tmp2 + addps xmm6, xmm5 ; xmm6=tmp0 + addps xmm7, xmm1 ; xmm7=tmp1 + + movaps XMMWORD [wk(1)], xmm4 ; tmp3 + movaps XMMWORD [wk(0)], xmm0 ; tmp2 + + ; -- Odd part + + movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_FAST_FLOAT)] + + movaps xmm4, xmm2 + movaps xmm0, xmm5 + addps xmm2, xmm1 ; xmm2=z11 + addps xmm5, xmm3 ; xmm5=z13 + subps xmm4, xmm1 ; xmm4=z12 + subps xmm0, xmm3 ; xmm0=z10 + + movaps xmm1, xmm2 + subps xmm2, xmm5 + addps xmm1, xmm5 ; xmm1=tmp7 + + mulps xmm2, [GOTOFF(ebx,PD_1_414)] ; xmm2=tmp11 + + movaps xmm3, xmm0 + addps xmm0, xmm4 + mulps xmm0, [GOTOFF(ebx,PD_1_847)] ; xmm0=z5 + mulps xmm3, [GOTOFF(ebx,PD_M2_613)] ; xmm3=(z10 * -2.613125930) + mulps xmm4, [GOTOFF(ebx,PD_1_082)] ; xmm4=(z12 * 1.082392200) + addps xmm3, xmm0 ; xmm3=tmp12 + subps xmm4, xmm0 ; xmm4=tmp10 + + ; -- Final output stage + + subps xmm3, xmm1 ; xmm3=tmp6 + movaps xmm5, xmm6 + movaps xmm0, xmm7 + addps xmm6, xmm1 ; xmm6=data0=(00 10 20 30) + addps xmm7, xmm3 ; xmm7=data1=(01 11 21 31) + subps xmm5, xmm1 ; xmm5=data7=(07 17 27 37) + subps xmm0, xmm3 ; xmm0=data6=(06 16 26 36) + subps xmm2, xmm3 ; xmm2=tmp5 + + movaps xmm1, [GOTOFF(ebx,PD_RNDINT_MAGIC)] ; xmm1=[PD_RNDINT_MAGIC] + pcmpeqd xmm3, xmm3 + psrld xmm3, WORD_BIT ; xmm3={0xFFFF 0x0000 0xFFFF 0x0000 ..} + + addps xmm6, xmm1 ; xmm6=roundint(data0/8)=(00 ** 10 ** 20 ** 30 **) + addps xmm7, xmm1 ; xmm7=roundint(data1/8)=(01 ** 11 ** 21 ** 31 **) + addps xmm0, xmm1 ; xmm0=roundint(data6/8)=(06 ** 16 ** 26 ** 36 **) + addps xmm5, xmm1 ; xmm5=roundint(data7/8)=(07 ** 17 ** 27 ** 37 **) + + pand xmm6, xmm3 ; xmm6=(00 -- 10 -- 20 -- 30 --) + pslld xmm7, WORD_BIT ; xmm7=(-- 01 -- 11 -- 21 -- 31) + pand xmm0, xmm3 ; xmm0=(06 -- 16 -- 26 -- 36 --) + pslld xmm5, WORD_BIT ; xmm5=(-- 07 -- 17 -- 27 -- 37) + por xmm6, xmm7 ; xmm6=(00 01 10 11 20 21 30 31) + por xmm0, xmm5 ; xmm0=(06 07 16 17 26 27 36 37) + + movaps xmm1, XMMWORD [wk(0)] ; xmm1=tmp2 + movaps xmm3, XMMWORD [wk(1)] ; xmm3=tmp3 + + addps xmm4, xmm2 ; xmm4=tmp4 + movaps xmm7, xmm1 + movaps xmm5, xmm3 + addps xmm1, xmm2 ; xmm1=data2=(02 12 22 32) + addps xmm3, xmm4 ; xmm3=data4=(04 14 24 34) + subps xmm7, xmm2 ; xmm7=data5=(05 15 25 35) + subps xmm5, xmm4 ; xmm5=data3=(03 13 23 33) + + movaps xmm2, [GOTOFF(ebx,PD_RNDINT_MAGIC)] ; xmm2=[PD_RNDINT_MAGIC] + pcmpeqd xmm4, xmm4 + psrld xmm4, WORD_BIT ; xmm4={0xFFFF 0x0000 0xFFFF 0x0000 ..} + + addps xmm3, xmm2 ; xmm3=roundint(data4/8)=(04 ** 14 ** 24 ** 34 **) + addps xmm7, xmm2 ; xmm7=roundint(data5/8)=(05 ** 15 ** 25 ** 35 **) + addps xmm1, xmm2 ; xmm1=roundint(data2/8)=(02 ** 12 ** 22 ** 32 **) + addps xmm5, xmm2 ; xmm5=roundint(data3/8)=(03 ** 13 ** 23 ** 33 **) + + pand xmm3, xmm4 ; xmm3=(04 -- 14 -- 24 -- 34 --) + pslld xmm7, WORD_BIT ; xmm7=(-- 05 -- 15 -- 25 -- 35) + pand xmm1, xmm4 ; xmm1=(02 -- 12 -- 22 -- 32 --) + pslld xmm5, WORD_BIT ; xmm5=(-- 03 -- 13 -- 23 -- 33) + por xmm3, xmm7 ; xmm3=(04 05 14 15 24 25 34 35) + por xmm1, xmm5 ; xmm1=(02 03 12 13 22 23 32 33) + + movdqa xmm2, [GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm2=[PB_CENTERJSAMP] + + packsswb xmm6, xmm3 ; xmm6=(00 01 10 11 20 21 30 31 04 05 14 15 24 25 34 35) + packsswb xmm1, xmm0 ; xmm1=(02 03 12 13 22 23 32 33 06 07 16 17 26 27 36 37) + paddb xmm6, xmm2 + paddb xmm1, xmm2 + + movdqa xmm4, xmm6 ; transpose coefficients(phase 2) + punpcklwd xmm6, xmm1 ; xmm6=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) + punpckhwd xmm4, xmm1 ; xmm4=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) + + movdqa xmm7, xmm6 ; transpose coefficients(phase 3) + punpckldq xmm6, xmm4 ; xmm6=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + punpckhdq xmm7, xmm4 ; xmm7=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + + pshufd xmm5, xmm6, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + pshufd xmm3, xmm7, 0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + + PUSHPIC ebx ; save GOT address + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE], xmm7 + mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm5 + movq XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE], xmm3 + + POPPIC ebx ; restore GOT address + + add esi, byte 4*SIZEOF_FAST_FLOAT ; wsptr + add edi, byte 4*SIZEOF_JSAMPROW + dec ecx ; ctr + jnz near .rowloop + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-mmx.asm new file mode 100644 index 0000000..77d4613 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-mmx.asm @@ -0,0 +1,495 @@ +; +; jidctfst.asm - fast integer IDCT (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a fast, not so accurate integer implementation of +; the inverse DCT (Discrete Cosine Transform). The following code is +; based directly on the IJG's original jidctfst.c; see the jidctfst.c +; for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 8 ; 14 is also OK. +%define PASS1_BITS 2 + +%if IFAST_SCALE_BITS != PASS1_BITS +%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'." +%endif + +%if CONST_BITS == 8 +F_1_082 equ 277 ; FIX(1.082392200) +F_1_414 equ 362 ; FIX(1.414213562) +F_1_847 equ 473 ; FIX(1.847759065) +F_2_613 equ 669 ; FIX(2.613125930) +F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_1_082 equ DESCALE(1162209775, 30 - CONST_BITS) ; FIX(1.082392200) +F_1_414 equ DESCALE(1518500249, 30 - CONST_BITS) ; FIX(1.414213562) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_2_613 equ DESCALE(2805822602, 30 - CONST_BITS) ; FIX(2.613125930) +F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + +; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow) +; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw) + +%define PRE_MULTIPLY_SCALE_BITS 2 +%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS) + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_ifast_mmx) + +EXTN(jconst_idct_ifast_mmx): + +PW_F1414 times 4 dw F_1_414 << CONST_SHIFT +PW_F1847 times 4 dw F_1_847 << CONST_SHIFT +PW_MF1613 times 4 dw -F_1_613 << CONST_SHIFT +PW_F1082 times 4 dw F_1_082 << CONST_SHIFT +PB_CENTERJSAMP times 8 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_ifast_mmx(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; jpeg_component_info *compptr +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD + ; mmword wk[WK_NUM] +%define WK_NUM 2 +%define workspace wk(0) - DCTSIZE2 * SIZEOF_JCOEF + ; JCOEF workspace[DCTSIZE2] + + align 32 + GLOBAL_FUNCTION(jsimd_idct_ifast_mmx) + +EXTN(jsimd_idct_ifast_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [workspace] + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input, store into work array. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + lea edi, [workspace] ; JCOEF *wsptr + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.columnloop: +%ifndef NO_ZERO_COLUMN_TEST_IFAST_MMX + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz short .columnDCT + + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por mm1, mm0 + packsswb mm1, mm1 + movd eax, mm1 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_IFAST_MULT_TYPE)] + + movq mm2, mm0 ; mm0=in0=(00 01 02 03) + punpcklwd mm0, mm0 ; mm0=(00 00 01 01) + punpckhwd mm2, mm2 ; mm2=(02 02 03 03) + + movq mm1, mm0 + punpckldq mm0, mm0 ; mm0=(00 00 00 00) + punpckhdq mm1, mm1 ; mm1=(01 01 01 01) + movq mm3, mm2 + punpckldq mm2, mm2 ; mm2=(02 02 02 02) + punpckhdq mm3, mm3 ; mm3=(03 03 03 03) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm3 + movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm3 + jmp near .nextcolumn + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Even part + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_IFAST_MULT_TYPE)] + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + pmullw mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw mm3, MMWORD [MMBLOCK(6,0,edx,SIZEOF_IFAST_MULT_TYPE)] + + movq mm4, mm0 + movq mm5, mm1 + psubw mm0, mm2 ; mm0=tmp11 + psubw mm1, mm3 + paddw mm4, mm2 ; mm4=tmp10 + paddw mm5, mm3 ; mm5=tmp13 + + psllw mm1, PRE_MULTIPLY_SCALE_BITS + pmulhw mm1, [GOTOFF(ebx,PW_F1414)] + psubw mm1, mm5 ; mm1=tmp12 + + movq mm6, mm4 + movq mm7, mm0 + psubw mm4, mm5 ; mm4=tmp3 + psubw mm0, mm1 ; mm0=tmp2 + paddw mm6, mm5 ; mm6=tmp0 + paddw mm7, mm1 ; mm7=tmp1 + + movq MMWORD [wk(1)], mm4 ; wk(1)=tmp3 + movq MMWORD [wk(0)], mm0 ; wk(0)=tmp2 + + ; -- Odd part + + movq mm2, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw mm2, MMWORD [MMBLOCK(1,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw mm3, MMWORD [MMBLOCK(3,0,edx,SIZEOF_IFAST_MULT_TYPE)] + movq mm5, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw mm5, MMWORD [MMBLOCK(5,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw mm1, MMWORD [MMBLOCK(7,0,edx,SIZEOF_IFAST_MULT_TYPE)] + + movq mm4, mm2 + movq mm0, mm5 + psubw mm2, mm1 ; mm2=z12 + psubw mm5, mm3 ; mm5=z10 + paddw mm4, mm1 ; mm4=z11 + paddw mm0, mm3 ; mm0=z13 + + movq mm1, mm5 ; mm1=z10(unscaled) + psllw mm2, PRE_MULTIPLY_SCALE_BITS + psllw mm5, PRE_MULTIPLY_SCALE_BITS + + movq mm3, mm4 + psubw mm4, mm0 + paddw mm3, mm0 ; mm3=tmp7 + + psllw mm4, PRE_MULTIPLY_SCALE_BITS + pmulhw mm4, [GOTOFF(ebx,PW_F1414)] ; mm4=tmp11 + + ; To avoid overflow... + ; + ; (Original) + ; tmp12 = -2.613125930 * z10 + z5; + ; + ; (This implementation) + ; tmp12 = (-1.613125930 - 1) * z10 + z5; + ; = -1.613125930 * z10 - z10 + z5; + + movq mm0, mm5 + paddw mm5, mm2 + pmulhw mm5, [GOTOFF(ebx,PW_F1847)] ; mm5=z5 + pmulhw mm0, [GOTOFF(ebx,PW_MF1613)] + pmulhw mm2, [GOTOFF(ebx,PW_F1082)] + psubw mm0, mm1 + psubw mm2, mm5 ; mm2=tmp10 + paddw mm0, mm5 ; mm0=tmp12 + + ; -- Final output stage + + psubw mm0, mm3 ; mm0=tmp6 + movq mm1, mm6 + movq mm5, mm7 + paddw mm6, mm3 ; mm6=data0=(00 01 02 03) + paddw mm7, mm0 ; mm7=data1=(10 11 12 13) + psubw mm1, mm3 ; mm1=data7=(70 71 72 73) + psubw mm5, mm0 ; mm5=data6=(60 61 62 63) + psubw mm4, mm0 ; mm4=tmp5 + + movq mm3, mm6 ; transpose coefficients(phase 1) + punpcklwd mm6, mm7 ; mm6=(00 10 01 11) + punpckhwd mm3, mm7 ; mm3=(02 12 03 13) + movq mm0, mm5 ; transpose coefficients(phase 1) + punpcklwd mm5, mm1 ; mm5=(60 70 61 71) + punpckhwd mm0, mm1 ; mm0=(62 72 63 73) + + movq mm7, MMWORD [wk(0)] ; mm7=tmp2 + movq mm1, MMWORD [wk(1)] ; mm1=tmp3 + + movq MMWORD [wk(0)], mm5 ; wk(0)=(60 70 61 71) + movq MMWORD [wk(1)], mm0 ; wk(1)=(62 72 63 73) + + paddw mm2, mm4 ; mm2=tmp4 + movq mm5, mm7 + movq mm0, mm1 + paddw mm7, mm4 ; mm7=data2=(20 21 22 23) + paddw mm1, mm2 ; mm1=data4=(40 41 42 43) + psubw mm5, mm4 ; mm5=data5=(50 51 52 53) + psubw mm0, mm2 ; mm0=data3=(30 31 32 33) + + movq mm4, mm7 ; transpose coefficients(phase 1) + punpcklwd mm7, mm0 ; mm7=(20 30 21 31) + punpckhwd mm4, mm0 ; mm4=(22 32 23 33) + movq mm2, mm1 ; transpose coefficients(phase 1) + punpcklwd mm1, mm5 ; mm1=(40 50 41 51) + punpckhwd mm2, mm5 ; mm2=(42 52 43 53) + + movq mm0, mm6 ; transpose coefficients(phase 2) + punpckldq mm6, mm7 ; mm6=(00 10 20 30) + punpckhdq mm0, mm7 ; mm0=(01 11 21 31) + movq mm5, mm3 ; transpose coefficients(phase 2) + punpckldq mm3, mm4 ; mm3=(02 12 22 32) + punpckhdq mm5, mm4 ; mm5=(03 13 23 33) + + movq mm7, MMWORD [wk(0)] ; mm7=(60 70 61 71) + movq mm4, MMWORD [wk(1)] ; mm4=(62 72 63 73) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm6 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm3 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm5 + + movq mm6, mm1 ; transpose coefficients(phase 2) + punpckldq mm1, mm7 ; mm1=(40 50 60 70) + punpckhdq mm6, mm7 ; mm6=(41 51 61 71) + movq mm0, mm2 ; transpose coefficients(phase 2) + punpckldq mm2, mm4 ; mm2=(42 52 62 72) + punpckhdq mm0, mm4 ; mm0=(43 53 63 73) + + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm6 + movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm0 + +.nextcolumn: + add esi, byte 4*SIZEOF_JCOEF ; coef_block + add edx, byte 4*SIZEOF_IFAST_MULT_TYPE ; quantptr + add edi, byte 4*DCTSIZE*SIZEOF_JCOEF ; wsptr + dec ecx ; ctr + jnz near .columnloop + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + lea esi, [workspace] ; JCOEF *wsptr + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.rowloop: + + ; -- Even part + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + + movq mm4, mm0 + movq mm5, mm1 + psubw mm0, mm2 ; mm0=tmp11 + psubw mm1, mm3 + paddw mm4, mm2 ; mm4=tmp10 + paddw mm5, mm3 ; mm5=tmp13 + + psllw mm1, PRE_MULTIPLY_SCALE_BITS + pmulhw mm1, [GOTOFF(ebx,PW_F1414)] + psubw mm1, mm5 ; mm1=tmp12 + + movq mm6, mm4 + movq mm7, mm0 + psubw mm4, mm5 ; mm4=tmp3 + psubw mm0, mm1 ; mm0=tmp2 + paddw mm6, mm5 ; mm6=tmp0 + paddw mm7, mm1 ; mm7=tmp1 + + movq MMWORD [wk(1)], mm4 ; wk(1)=tmp3 + movq MMWORD [wk(0)], mm0 ; wk(0)=tmp2 + + ; -- Odd part + + movq mm2, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + movq mm5, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + + movq mm4, mm2 + movq mm0, mm5 + psubw mm2, mm1 ; mm2=z12 + psubw mm5, mm3 ; mm5=z10 + paddw mm4, mm1 ; mm4=z11 + paddw mm0, mm3 ; mm0=z13 + + movq mm1, mm5 ; mm1=z10(unscaled) + psllw mm2, PRE_MULTIPLY_SCALE_BITS + psllw mm5, PRE_MULTIPLY_SCALE_BITS + + movq mm3, mm4 + psubw mm4, mm0 + paddw mm3, mm0 ; mm3=tmp7 + + psllw mm4, PRE_MULTIPLY_SCALE_BITS + pmulhw mm4, [GOTOFF(ebx,PW_F1414)] ; mm4=tmp11 + + ; To avoid overflow... + ; + ; (Original) + ; tmp12 = -2.613125930 * z10 + z5; + ; + ; (This implementation) + ; tmp12 = (-1.613125930 - 1) * z10 + z5; + ; = -1.613125930 * z10 - z10 + z5; + + movq mm0, mm5 + paddw mm5, mm2 + pmulhw mm5, [GOTOFF(ebx,PW_F1847)] ; mm5=z5 + pmulhw mm0, [GOTOFF(ebx,PW_MF1613)] + pmulhw mm2, [GOTOFF(ebx,PW_F1082)] + psubw mm0, mm1 + psubw mm2, mm5 ; mm2=tmp10 + paddw mm0, mm5 ; mm0=tmp12 + + ; -- Final output stage + + psubw mm0, mm3 ; mm0=tmp6 + movq mm1, mm6 + movq mm5, mm7 + paddw mm6, mm3 ; mm6=data0=(00 10 20 30) + paddw mm7, mm0 ; mm7=data1=(01 11 21 31) + psraw mm6, (PASS1_BITS+3) ; descale + psraw mm7, (PASS1_BITS+3) ; descale + psubw mm1, mm3 ; mm1=data7=(07 17 27 37) + psubw mm5, mm0 ; mm5=data6=(06 16 26 36) + psraw mm1, (PASS1_BITS+3) ; descale + psraw mm5, (PASS1_BITS+3) ; descale + psubw mm4, mm0 ; mm4=tmp5 + + packsswb mm6, mm5 ; mm6=(00 10 20 30 06 16 26 36) + packsswb mm7, mm1 ; mm7=(01 11 21 31 07 17 27 37) + + movq mm3, MMWORD [wk(0)] ; mm3=tmp2 + movq mm0, MMWORD [wk(1)] ; mm0=tmp3 + + paddw mm2, mm4 ; mm2=tmp4 + movq mm5, mm3 + movq mm1, mm0 + paddw mm3, mm4 ; mm3=data2=(02 12 22 32) + paddw mm0, mm2 ; mm0=data4=(04 14 24 34) + psraw mm3, (PASS1_BITS+3) ; descale + psraw mm0, (PASS1_BITS+3) ; descale + psubw mm5, mm4 ; mm5=data5=(05 15 25 35) + psubw mm1, mm2 ; mm1=data3=(03 13 23 33) + psraw mm5, (PASS1_BITS+3) ; descale + psraw mm1, (PASS1_BITS+3) ; descale + + movq mm4, [GOTOFF(ebx,PB_CENTERJSAMP)] ; mm4=[PB_CENTERJSAMP] + + packsswb mm3, mm0 ; mm3=(02 12 22 32 04 14 24 34) + packsswb mm1, mm5 ; mm1=(03 13 23 33 05 15 25 35) + + paddb mm6, mm4 + paddb mm7, mm4 + paddb mm3, mm4 + paddb mm1, mm4 + + movq mm2, mm6 ; transpose coefficients(phase 1) + punpcklbw mm6, mm7 ; mm6=(00 01 10 11 20 21 30 31) + punpckhbw mm2, mm7 ; mm2=(06 07 16 17 26 27 36 37) + movq mm0, mm3 ; transpose coefficients(phase 1) + punpcklbw mm3, mm1 ; mm3=(02 03 12 13 22 23 32 33) + punpckhbw mm0, mm1 ; mm0=(04 05 14 15 24 25 34 35) + + movq mm5, mm6 ; transpose coefficients(phase 2) + punpcklwd mm6, mm3 ; mm6=(00 01 02 03 10 11 12 13) + punpckhwd mm5, mm3 ; mm5=(20 21 22 23 30 31 32 33) + movq mm4, mm0 ; transpose coefficients(phase 2) + punpcklwd mm0, mm2 ; mm0=(04 05 06 07 14 15 16 17) + punpckhwd mm4, mm2 ; mm4=(24 25 26 27 34 35 36 37) + + movq mm7, mm6 ; transpose coefficients(phase 3) + punpckldq mm6, mm0 ; mm6=(00 01 02 03 04 05 06 07) + punpckhdq mm7, mm0 ; mm7=(10 11 12 13 14 15 16 17) + movq mm1, mm5 ; transpose coefficients(phase 3) + punpckldq mm5, mm4 ; mm5=(20 21 22 23 24 25 26 27) + punpckhdq mm1, mm4 ; mm1=(30 31 32 33 34 35 36 37) + + PUSHPIC ebx ; save GOT address + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm6 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm7 + mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm5 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm1 + + POPPIC ebx ; restore GOT address + + add esi, byte 4*SIZEOF_JCOEF ; wsptr + add edi, byte 4*SIZEOF_JSAMPROW + dec ecx ; ctr + jnz near .rowloop + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-sse2.asm new file mode 100644 index 0000000..c2fe34b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctfst-sse2.asm @@ -0,0 +1,497 @@ +; +; jidctfst.asm - fast integer IDCT (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a fast, not so accurate integer implementation of +; the inverse DCT (Discrete Cosine Transform). The following code is +; based directly on the IJG's original jidctfst.c; see the jidctfst.c +; for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 8 ; 14 is also OK. +%define PASS1_BITS 2 + +%if IFAST_SCALE_BITS != PASS1_BITS +%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'." +%endif + +%if CONST_BITS == 8 +F_1_082 equ 277 ; FIX(1.082392200) +F_1_414 equ 362 ; FIX(1.414213562) +F_1_847 equ 473 ; FIX(1.847759065) +F_2_613 equ 669 ; FIX(2.613125930) +F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_1_082 equ DESCALE(1162209775, 30 - CONST_BITS) ; FIX(1.082392200) +F_1_414 equ DESCALE(1518500249, 30 - CONST_BITS) ; FIX(1.414213562) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_2_613 equ DESCALE(2805822602, 30 - CONST_BITS) ; FIX(2.613125930) +F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + +; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow) +; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw) + +%define PRE_MULTIPLY_SCALE_BITS 2 +%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS) + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_ifast_sse2) + +EXTN(jconst_idct_ifast_sse2): + +PW_F1414 times 8 dw F_1_414 << CONST_SHIFT +PW_F1847 times 8 dw F_1_847 << CONST_SHIFT +PW_MF1613 times 8 dw -F_1_613 << CONST_SHIFT +PW_F1082 times 8 dw F_1_082 << CONST_SHIFT +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_ifast_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; jpeg_component_info *compptr +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_idct_ifast_sse2) + +EXTN(jsimd_idct_ifast_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + +%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2 + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por xmm1, xmm0 + packsswb xmm1, xmm1 + packsswb xmm1, xmm1 + movd eax, xmm1 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + movdqa xmm7, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07) + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + punpckhwd xmm7, xmm7 ; xmm7=(04 04 05 05 06 06 07 07) + + pshufd xmm6, xmm0, 0x00 ; xmm6=col0=(00 00 00 00 00 00 00 00) + pshufd xmm2, xmm0, 0x55 ; xmm2=col1=(01 01 01 01 01 01 01 01) + pshufd xmm5, xmm0, 0xAA ; xmm5=col2=(02 02 02 02 02 02 02 02) + pshufd xmm0, xmm0, 0xFF ; xmm0=col3=(03 03 03 03 03 03 03 03) + pshufd xmm1, xmm7, 0x00 ; xmm1=col4=(04 04 04 04 04 04 04 04) + pshufd xmm4, xmm7, 0x55 ; xmm4=col5=(05 05 05 05 05 05 05 05) + pshufd xmm3, xmm7, 0xAA ; xmm3=col6=(06 06 06 06 06 06 06 06) + pshufd xmm7, xmm7, 0xFF ; xmm7=col7=(07 07 07 07 07 07 07 07) + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=col1 + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=col3 + jmp near .column_end + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Even part + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_IFAST_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_IFAST_MULT_TYPE)] + + movdqa xmm4, xmm0 + movdqa xmm5, xmm1 + psubw xmm0, xmm2 ; xmm0=tmp11 + psubw xmm1, xmm3 + paddw xmm4, xmm2 ; xmm4=tmp10 + paddw xmm5, xmm3 ; xmm5=tmp13 + + psllw xmm1, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm1, [GOTOFF(ebx,PW_F1414)] + psubw xmm1, xmm5 ; xmm1=tmp12 + + movdqa xmm6, xmm4 + movdqa xmm7, xmm0 + psubw xmm4, xmm5 ; xmm4=tmp3 + psubw xmm0, xmm1 ; xmm0=tmp2 + paddw xmm6, xmm5 ; xmm6=tmp0 + paddw xmm7, xmm1 ; xmm7=tmp1 + + movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=tmp3 + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=tmp2 + + ; -- Odd part + + movdqa xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_IFAST_MULT_TYPE)] + movdqa xmm5, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw xmm5, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_IFAST_MULT_TYPE)] + + movdqa xmm4, xmm2 + movdqa xmm0, xmm5 + psubw xmm2, xmm1 ; xmm2=z12 + psubw xmm5, xmm3 ; xmm5=z10 + paddw xmm4, xmm1 ; xmm4=z11 + paddw xmm0, xmm3 ; xmm0=z13 + + movdqa xmm1, xmm5 ; xmm1=z10(unscaled) + psllw xmm2, PRE_MULTIPLY_SCALE_BITS + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + + movdqa xmm3, xmm4 + psubw xmm4, xmm0 + paddw xmm3, xmm0 ; xmm3=tmp7 + + psllw xmm4, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm4, [GOTOFF(ebx,PW_F1414)] ; xmm4=tmp11 + + ; To avoid overflow... + ; + ; (Original) + ; tmp12 = -2.613125930 * z10 + z5; + ; + ; (This implementation) + ; tmp12 = (-1.613125930 - 1) * z10 + z5; + ; = -1.613125930 * z10 - z10 + z5; + + movdqa xmm0, xmm5 + paddw xmm5, xmm2 + pmulhw xmm5, [GOTOFF(ebx,PW_F1847)] ; xmm5=z5 + pmulhw xmm0, [GOTOFF(ebx,PW_MF1613)] + pmulhw xmm2, [GOTOFF(ebx,PW_F1082)] + psubw xmm0, xmm1 + psubw xmm2, xmm5 ; xmm2=tmp10 + paddw xmm0, xmm5 ; xmm0=tmp12 + + ; -- Final output stage + + psubw xmm0, xmm3 ; xmm0=tmp6 + movdqa xmm1, xmm6 + movdqa xmm5, xmm7 + paddw xmm6, xmm3 ; xmm6=data0=(00 01 02 03 04 05 06 07) + paddw xmm7, xmm0 ; xmm7=data1=(10 11 12 13 14 15 16 17) + psubw xmm1, xmm3 ; xmm1=data7=(70 71 72 73 74 75 76 77) + psubw xmm5, xmm0 ; xmm5=data6=(60 61 62 63 64 65 66 67) + psubw xmm4, xmm0 ; xmm4=tmp5 + + movdqa xmm3, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm7 ; xmm6=(00 10 01 11 02 12 03 13) + punpckhwd xmm3, xmm7 ; xmm3=(04 14 05 15 06 16 07 17) + movdqa xmm0, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm1 ; xmm5=(60 70 61 71 62 72 63 73) + punpckhwd xmm0, xmm1 ; xmm0=(64 74 65 75 66 76 67 77) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp2 + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp3 + + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(60 70 61 71 62 72 63 73) + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(64 74 65 75 66 76 67 77) + + paddw xmm2, xmm4 ; xmm2=tmp4 + movdqa xmm5, xmm7 + movdqa xmm0, xmm1 + paddw xmm7, xmm4 ; xmm7=data2=(20 21 22 23 24 25 26 27) + paddw xmm1, xmm2 ; xmm1=data4=(40 41 42 43 44 45 46 47) + psubw xmm5, xmm4 ; xmm5=data5=(50 51 52 53 54 55 56 57) + psubw xmm0, xmm2 ; xmm0=data3=(30 31 32 33 34 35 36 37) + + movdqa xmm4, xmm7 ; transpose coefficients(phase 1) + punpcklwd xmm7, xmm0 ; xmm7=(20 30 21 31 22 32 23 33) + punpckhwd xmm4, xmm0 ; xmm4=(24 34 25 35 26 36 27 37) + movdqa xmm2, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm5 ; xmm1=(40 50 41 51 42 52 43 53) + punpckhwd xmm2, xmm5 ; xmm2=(44 54 45 55 46 56 47 57) + + movdqa xmm0, xmm3 ; transpose coefficients(phase 2) + punpckldq xmm3, xmm4 ; xmm3=(04 14 24 34 05 15 25 35) + punpckhdq xmm0, xmm4 ; xmm0=(06 16 26 36 07 17 27 37) + movdqa xmm5, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm7 ; xmm6=(00 10 20 30 01 11 21 31) + punpckhdq xmm5, xmm7 ; xmm5=(02 12 22 32 03 13 23 33) + + movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(60 70 61 71 62 72 63 73) + movdqa xmm7, XMMWORD [wk(1)] ; xmm7=(64 74 65 75 66 76 67 77) + + movdqa XMMWORD [wk(0)], xmm3 ; wk(0)=(04 14 24 34 05 15 25 35) + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(06 16 26 36 07 17 27 37) + + movdqa xmm3, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm4 ; xmm1=(40 50 60 70 41 51 61 71) + punpckhdq xmm3, xmm4 ; xmm3=(42 52 62 72 43 53 63 73) + movdqa xmm0, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm7 ; xmm2=(44 54 64 74 45 55 65 75) + punpckhdq xmm0, xmm7 ; xmm0=(46 56 66 76 47 57 67 77) + + movdqa xmm4, xmm6 ; transpose coefficients(phase 3) + punpcklqdq xmm6, xmm1 ; xmm6=col0=(00 10 20 30 40 50 60 70) + punpckhqdq xmm4, xmm1 ; xmm4=col1=(01 11 21 31 41 51 61 71) + movdqa xmm7, xmm5 ; transpose coefficients(phase 3) + punpcklqdq xmm5, xmm3 ; xmm5=col2=(02 12 22 32 42 52 62 72) + punpckhqdq xmm7, xmm3 ; xmm7=col3=(03 13 23 33 43 53 63 73) + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(04 14 24 34 05 15 25 35) + movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(06 16 26 36 07 17 27 37) + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=col1 + movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=col3 + + movdqa xmm4, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm2 ; xmm1=col4=(04 14 24 34 44 54 64 74) + punpckhqdq xmm4, xmm2 ; xmm4=col5=(05 15 25 35 45 55 65 75) + movdqa xmm7, xmm3 ; transpose coefficients(phase 3) + punpcklqdq xmm3, xmm0 ; xmm3=col6=(06 16 26 36 46 56 66 76) + punpckhqdq xmm7, xmm0 ; xmm7=col7=(07 17 27 37 47 57 67 77) +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + + ; -- Even part + + ; xmm6=col0, xmm5=col2, xmm1=col4, xmm3=col6 + + movdqa xmm2, xmm6 + movdqa xmm0, xmm5 + psubw xmm6, xmm1 ; xmm6=tmp11 + psubw xmm5, xmm3 + paddw xmm2, xmm1 ; xmm2=tmp10 + paddw xmm0, xmm3 ; xmm0=tmp13 + + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm5, [GOTOFF(ebx,PW_F1414)] + psubw xmm5, xmm0 ; xmm5=tmp12 + + movdqa xmm1, xmm2 + movdqa xmm3, xmm6 + psubw xmm2, xmm0 ; xmm2=tmp3 + psubw xmm6, xmm5 ; xmm6=tmp2 + paddw xmm1, xmm0 ; xmm1=tmp0 + paddw xmm3, xmm5 ; xmm3=tmp1 + + movdqa xmm0, XMMWORD [wk(0)] ; xmm0=col1 + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=col3 + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp3 + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp2 + + ; -- Odd part + + ; xmm0=col1, xmm5=col3, xmm4=col5, xmm7=col7 + + movdqa xmm2, xmm0 + movdqa xmm6, xmm4 + psubw xmm0, xmm7 ; xmm0=z12 + psubw xmm4, xmm5 ; xmm4=z10 + paddw xmm2, xmm7 ; xmm2=z11 + paddw xmm6, xmm5 ; xmm6=z13 + + movdqa xmm7, xmm4 ; xmm7=z10(unscaled) + psllw xmm0, PRE_MULTIPLY_SCALE_BITS + psllw xmm4, PRE_MULTIPLY_SCALE_BITS + + movdqa xmm5, xmm2 + psubw xmm2, xmm6 + paddw xmm5, xmm6 ; xmm5=tmp7 + + psllw xmm2, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm2, [GOTOFF(ebx,PW_F1414)] ; xmm2=tmp11 + + ; To avoid overflow... + ; + ; (Original) + ; tmp12 = -2.613125930 * z10 + z5; + ; + ; (This implementation) + ; tmp12 = (-1.613125930 - 1) * z10 + z5; + ; = -1.613125930 * z10 - z10 + z5; + + movdqa xmm6, xmm4 + paddw xmm4, xmm0 + pmulhw xmm4, [GOTOFF(ebx,PW_F1847)] ; xmm4=z5 + pmulhw xmm6, [GOTOFF(ebx,PW_MF1613)] + pmulhw xmm0, [GOTOFF(ebx,PW_F1082)] + psubw xmm6, xmm7 + psubw xmm0, xmm4 ; xmm0=tmp10 + paddw xmm6, xmm4 ; xmm6=tmp12 + + ; -- Final output stage + + psubw xmm6, xmm5 ; xmm6=tmp6 + movdqa xmm7, xmm1 + movdqa xmm4, xmm3 + paddw xmm1, xmm5 ; xmm1=data0=(00 10 20 30 40 50 60 70) + paddw xmm3, xmm6 ; xmm3=data1=(01 11 21 31 41 51 61 71) + psraw xmm1, (PASS1_BITS+3) ; descale + psraw xmm3, (PASS1_BITS+3) ; descale + psubw xmm7, xmm5 ; xmm7=data7=(07 17 27 37 47 57 67 77) + psubw xmm4, xmm6 ; xmm4=data6=(06 16 26 36 46 56 66 76) + psraw xmm7, (PASS1_BITS+3) ; descale + psraw xmm4, (PASS1_BITS+3) ; descale + psubw xmm2, xmm6 ; xmm2=tmp5 + + packsswb xmm1, xmm4 ; xmm1=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + packsswb xmm3, xmm7 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp2 + movdqa xmm6, XMMWORD [wk(0)] ; xmm6=tmp3 + + paddw xmm0, xmm2 ; xmm0=tmp4 + movdqa xmm4, xmm5 + movdqa xmm7, xmm6 + paddw xmm5, xmm2 ; xmm5=data2=(02 12 22 32 42 52 62 72) + paddw xmm6, xmm0 ; xmm6=data4=(04 14 24 34 44 54 64 74) + psraw xmm5, (PASS1_BITS+3) ; descale + psraw xmm6, (PASS1_BITS+3) ; descale + psubw xmm4, xmm2 ; xmm4=data5=(05 15 25 35 45 55 65 75) + psubw xmm7, xmm0 ; xmm7=data3=(03 13 23 33 43 53 63 73) + psraw xmm4, (PASS1_BITS+3) ; descale + psraw xmm7, (PASS1_BITS+3) ; descale + + movdqa xmm2, [GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm2=[PB_CENTERJSAMP] + + packsswb xmm5, xmm6 ; xmm5=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74) + packsswb xmm7, xmm4 ; xmm7=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75) + + paddb xmm1, xmm2 + paddb xmm3, xmm2 + paddb xmm5, xmm2 + paddb xmm7, xmm2 + + movdqa xmm0, xmm1 ; transpose coefficients(phase 1) + punpcklbw xmm1, xmm3 ; xmm1=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71) + punpckhbw xmm0, xmm3 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77) + movdqa xmm6, xmm5 ; transpose coefficients(phase 1) + punpcklbw xmm5, xmm7 ; xmm5=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73) + punpckhbw xmm6, xmm7 ; xmm6=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75) + + movdqa xmm4, xmm1 ; transpose coefficients(phase 2) + punpcklwd xmm1, xmm5 ; xmm1=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) + punpckhwd xmm4, xmm5 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73) + movdqa xmm2, xmm6 ; transpose coefficients(phase 2) + punpcklwd xmm6, xmm0 ; xmm6=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) + punpckhwd xmm2, xmm0 ; xmm2=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77) + + movdqa xmm3, xmm1 ; transpose coefficients(phase 3) + punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + punpckhdq xmm3, xmm6 ; xmm3=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + movdqa xmm7, xmm4 ; transpose coefficients(phase 3) + punpckldq xmm4, xmm2 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) + punpckhdq xmm7, xmm2 ; xmm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) + + pshufd xmm5, xmm1, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + pshufd xmm0, xmm3, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + pshufd xmm6, xmm4, 0x4E ; xmm6=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) + pshufd xmm2, xmm7, 0x4E ; xmm2=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm1 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3 + mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+6*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm7 + + mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm5 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm0 + mov edx, JSAMPROW [edi+5*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm2 + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-avx2.asm new file mode 100644 index 0000000..cb119d3 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-avx2.asm @@ -0,0 +1,449 @@ +; +; jidctint.asm - accurate integer IDCT (AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2018, 2020, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; inverse DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jidctint.c; see the jidctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit inverse matrix transpose using AVX2 instructions +; %1-%4: Input/output registers +; %5-%8: Temp registers + +%macro DOTRANSPOSE 8 + ; %5=(00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71) + ; %6=(03 13 23 33 43 53 63 73 02 12 22 32 42 52 62 72) + ; %7=(04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75) + ; %8=(07 17 27 37 47 57 67 77 06 16 26 36 46 56 66 76) + + vpermq %5, %1, 0xD8 + vpermq %6, %2, 0x72 + vpermq %7, %3, 0xD8 + vpermq %8, %4, 0x72 + ; transpose coefficients(phase 1) + ; %5=(00 10 20 30 01 11 21 31 40 50 60 70 41 51 61 71) + ; %6=(02 12 22 32 03 13 23 33 42 52 62 72 43 53 63 73) + ; %7=(04 14 24 34 05 15 25 35 44 54 64 74 45 55 65 75) + ; %8=(06 16 26 36 07 17 27 37 46 56 66 76 47 57 67 77) + + vpunpcklwd %1, %5, %6 + vpunpckhwd %2, %5, %6 + vpunpcklwd %3, %7, %8 + vpunpckhwd %4, %7, %8 + ; transpose coefficients(phase 2) + ; %1=(00 02 10 12 20 22 30 32 40 42 50 52 60 62 70 72) + ; %2=(01 03 11 13 21 23 31 33 41 43 51 53 61 63 71 73) + ; %3=(04 06 14 16 24 26 34 36 44 46 54 56 64 66 74 76) + ; %4=(05 07 15 17 25 27 35 37 45 47 55 57 65 67 75 77) + + vpunpcklwd %5, %1, %2 + vpunpcklwd %6, %3, %4 + vpunpckhwd %7, %1, %2 + vpunpckhwd %8, %3, %4 + ; transpose coefficients(phase 3) + ; %5=(00 01 02 03 10 11 12 13 40 41 42 43 50 51 52 53) + ; %6=(04 05 06 07 14 15 16 17 44 45 46 47 54 55 56 57) + ; %7=(20 21 22 23 30 31 32 33 60 61 62 63 70 71 72 73) + ; %8=(24 25 26 27 34 35 36 37 64 65 66 67 74 75 76 77) + + vpunpcklqdq %1, %5, %6 + vpunpckhqdq %2, %5, %6 + vpunpcklqdq %3, %7, %8 + vpunpckhqdq %4, %7, %8 + ; transpose coefficients(phase 4) + ; %1=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47) + ; %2=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57) + ; %3=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67) + ; %4=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77) +%endmacro + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit accurate integer inverse DCT using AVX2 instructions +; %1-%4: Input/output registers +; %5-%12: Temp registers +; %9: Pass (1 or 2) + +%macro DODCT 13 + ; -- Even part + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + vperm2i128 %6, %3, %3, 0x01 ; %6=in6_2 + vpunpcklwd %5, %3, %6 ; %5=in26_62L + vpunpckhwd %6, %3, %6 ; %6=in26_62H + vpmaddwd %5, %5, [GOTOFF(ebx,PW_F130_F054_MF130_F054)] ; %5=tmp3_2L + vpmaddwd %6, %6, [GOTOFF(ebx,PW_F130_F054_MF130_F054)] ; %6=tmp3_2H + + vperm2i128 %7, %1, %1, 0x01 ; %7=in4_0 + vpsignw %1, %1, [GOTOFF(ebx,PW_1_NEG1)] + vpaddw %7, %7, %1 ; %7=(in0+in4)_(in0-in4) + + vpxor %1, %1, %1 + vpunpcklwd %8, %1, %7 ; %8=tmp0_1L + vpunpckhwd %1, %1, %7 ; %1=tmp0_1H + vpsrad %8, %8, (16-CONST_BITS) ; vpsrad %8,16 & vpslld %8,CONST_BITS + vpsrad %1, %1, (16-CONST_BITS) ; vpsrad %1,16 & vpslld %1,CONST_BITS + + vpsubd %3, %8, %5 + vmovdqu %11, %3 ; %11=tmp0_1L-tmp3_2L=tmp13_12L + vpaddd %3, %8, %5 + vmovdqu %9, %3 ; %9=tmp0_1L+tmp3_2L=tmp10_11L + vpsubd %3, %1, %6 + vmovdqu %12, %3 ; %12=tmp0_1H-tmp3_2H=tmp13_12H + vpaddd %3, %1, %6 + vmovdqu %10, %3 ; %10=tmp0_1H+tmp3_2H=tmp10_11H + + ; -- Odd part + + vpaddw %1, %4, %2 ; %1=in7_5+in3_1=z3_4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + vperm2i128 %8, %1, %1, 0x01 ; %8=z4_3 + vpunpcklwd %7, %1, %8 ; %7=z34_43L + vpunpckhwd %8, %1, %8 ; %8=z34_43H + vpmaddwd %7, %7, [GOTOFF(ebx,PW_MF078_F117_F078_F117)] ; %7=z3_4L + vpmaddwd %8, %8, [GOTOFF(ebx,PW_MF078_F117_F078_F117)] ; %8=z3_4H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + vperm2i128 %2, %2, %2, 0x01 ; %2=in1_3 + vpunpcklwd %3, %4, %2 ; %3=in71_53L + vpunpckhwd %4, %4, %2 ; %4=in71_53H + + vpmaddwd %5, %3, [GOTOFF(ebx,PW_MF060_MF089_MF050_MF256)] ; %5=tmp0_1L + vpmaddwd %6, %4, [GOTOFF(ebx,PW_MF060_MF089_MF050_MF256)] ; %6=tmp0_1H + vpaddd %5, %5, %7 ; %5=tmp0_1L+z3_4L=tmp0_1L + vpaddd %6, %6, %8 ; %6=tmp0_1H+z3_4H=tmp0_1H + + vpmaddwd %3, %3, [GOTOFF(ebx,PW_MF089_F060_MF256_F050)] ; %3=tmp3_2L + vpmaddwd %4, %4, [GOTOFF(ebx,PW_MF089_F060_MF256_F050)] ; %4=tmp3_2H + vperm2i128 %7, %7, %7, 0x01 ; %7=z4_3L + vperm2i128 %8, %8, %8, 0x01 ; %8=z4_3H + vpaddd %7, %3, %7 ; %7=tmp3_2L+z4_3L=tmp3_2L + vpaddd %8, %4, %8 ; %8=tmp3_2H+z4_3H=tmp3_2H + + ; -- Final output stage + + vmovdqu %3, %9 + vmovdqu %4, %10 + + vpaddd %1, %3, %7 ; %1=tmp10_11L+tmp3_2L=data0_1L + vpaddd %2, %4, %8 ; %2=tmp10_11H+tmp3_2H=data0_1H + vpaddd %1, %1, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpaddd %2, %2, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpsrad %1, %1, DESCALE_P %+ %13 + vpsrad %2, %2, DESCALE_P %+ %13 + vpackssdw %1, %1, %2 ; %1=data0_1 + + vpsubd %3, %3, %7 ; %3=tmp10_11L-tmp3_2L=data7_6L + vpsubd %4, %4, %8 ; %4=tmp10_11H-tmp3_2H=data7_6H + vpaddd %3, %3, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpaddd %4, %4, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpsrad %3, %3, DESCALE_P %+ %13 + vpsrad %4, %4, DESCALE_P %+ %13 + vpackssdw %4, %3, %4 ; %4=data7_6 + + vmovdqu %7, %11 + vmovdqu %8, %12 + + vpaddd %2, %7, %5 ; %7=tmp13_12L+tmp0_1L=data3_2L + vpaddd %3, %8, %6 ; %8=tmp13_12H+tmp0_1H=data3_2H + vpaddd %2, %2, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpaddd %3, %3, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpsrad %2, %2, DESCALE_P %+ %13 + vpsrad %3, %3, DESCALE_P %+ %13 + vpackssdw %2, %2, %3 ; %2=data3_2 + + vpsubd %3, %7, %5 ; %7=tmp13_12L-tmp0_1L=data4_5L + vpsubd %6, %8, %6 ; %8=tmp13_12H-tmp0_1H=data4_5H + vpaddd %3, %3, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpaddd %6, %6, [GOTOFF(ebx,PD_DESCALE_P %+ %13)] + vpsrad %3, %3, DESCALE_P %+ %13 + vpsrad %6, %6, DESCALE_P %+ %13 + vpackssdw %3, %3, %6 ; %3=data4_5 +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_islow_avx2) + +EXTN(jconst_idct_islow_avx2): + +PW_F130_F054_MF130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 + times 4 dw (F_0_541 - F_1_847), F_0_541 +PW_MF078_F117_F078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 + times 4 dw (F_1_175 - F_0_390), F_1_175 +PW_MF060_MF089_MF050_MF256 times 4 dw (F_0_298 - F_0_899), -F_0_899 + times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF089_F060_MF256_F050 times 4 dw -F_0_899, (F_1_501 - F_0_899) + times 4 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 8 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 8 dd 1 << (DESCALE_P2 - 1) +PB_CENTERJSAMP times 32 db CENTERJSAMPLE +PW_1_NEG1 times 8 dw 1 + times 8 dw -1 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_islow_avx2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; jpeg_component_info *compptr +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_YMMWORD + ; ymmword wk[WK_NUM] +%define WK_NUM 4 + + align 32 + GLOBAL_FUNCTION(jsimd_idct_islow_avx2) + +EXTN(jsimd_idct_islow_avx2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + +%ifndef NO_ZERO_COLUMN_TEST_ISLOW_AVX2 + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] + vpor xmm0, xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + vpor xmm1, xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)] + vpor xmm0, xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + vpor xmm1, xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] + vpor xmm0, xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + vpor xmm1, xmm1, xmm0 + vpacksswb xmm1, xmm1, xmm1 + vpacksswb xmm1, xmm1, xmm1 + movd eax, xmm1 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm5, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + vpmullw xmm5, xmm5, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + vpsllw xmm5, xmm5, PASS1_BITS + + vpunpcklwd xmm4, xmm5, xmm5 ; xmm4=(00 00 01 01 02 02 03 03) + vpunpckhwd xmm5, xmm5, xmm5 ; xmm5=(04 04 05 05 06 06 07 07) + vinserti128 ymm4, ymm4, xmm5, 1 + + vpshufd ymm0, ymm4, 0x00 ; ymm0=col0_4=(00 00 00 00 00 00 00 00 04 04 04 04 04 04 04 04) + vpshufd ymm1, ymm4, 0x55 ; ymm1=col1_5=(01 01 01 01 01 01 01 01 05 05 05 05 05 05 05 05) + vpshufd ymm2, ymm4, 0xAA ; ymm2=col2_6=(02 02 02 02 02 02 02 02 06 06 06 06 06 06 06 06) + vpshufd ymm3, ymm4, 0xFF ; ymm3=col3_7=(03 03 03 03 03 03 03 03 07 07 07 07 07 07 07 07) + + jmp near .column_end + ALIGNX 16, 7 +%endif +.columnDCT: + + vmovdqu ymm4, YMMWORD [YMMBLOCK(0,0,esi,SIZEOF_JCOEF)] ; ymm4=in0_1 + vmovdqu ymm5, YMMWORD [YMMBLOCK(2,0,esi,SIZEOF_JCOEF)] ; ymm5=in2_3 + vmovdqu ymm6, YMMWORD [YMMBLOCK(4,0,esi,SIZEOF_JCOEF)] ; ymm6=in4_5 + vmovdqu ymm7, YMMWORD [YMMBLOCK(6,0,esi,SIZEOF_JCOEF)] ; ymm7=in6_7 + vpmullw ymm4, ymm4, YMMWORD [YMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + vpmullw ymm5, ymm5, YMMWORD [YMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + vpmullw ymm6, ymm6, YMMWORD [YMMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + vpmullw ymm7, ymm7, YMMWORD [YMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + vperm2i128 ymm0, ymm4, ymm6, 0x20 ; ymm0=in0_4 + vperm2i128 ymm1, ymm5, ymm4, 0x31 ; ymm1=in3_1 + vperm2i128 ymm2, ymm5, ymm7, 0x20 ; ymm2=in2_6 + vperm2i128 ymm3, ymm7, ymm6, 0x31 ; ymm3=in7_5 + + DODCT ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7, XMMWORD [wk(0)], XMMWORD [wk(1)], XMMWORD [wk(2)], XMMWORD [wk(3)], 1 + ; ymm0=data0_1, ymm1=data3_2, ymm2=data4_5, ymm3=data7_6 + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7 + ; ymm0=data0_4, ymm1=data1_5, ymm2=data2_6, ymm3=data3_7 + +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows. + + mov eax, [original_ebp] + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + + vperm2i128 ymm4, ymm3, ymm1, 0x31 ; ymm3=in7_5 + vperm2i128 ymm1, ymm3, ymm1, 0x20 ; ymm1=in3_1 + + DODCT ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7, XMMWORD [wk(0)], XMMWORD [wk(1)], XMMWORD [wk(2)], XMMWORD [wk(3)], 2 + ; ymm0=data0_1, ymm1=data3_2, ymm2=data4_5, ymm4=data7_6 + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7 + ; ymm0=data0_4, ymm1=data1_5, ymm2=data2_6, ymm4=data3_7 + + vpacksswb ymm0, ymm0, ymm1 ; ymm0=data01_45 + vpacksswb ymm1, ymm2, ymm4 ; ymm1=data23_67 + vpaddb ymm0, ymm0, [GOTOFF(ebx,PB_CENTERJSAMP)] + vpaddb ymm1, ymm1, [GOTOFF(ebx,PB_CENTERJSAMP)] + + vextracti128 xmm6, ymm1, 1 ; xmm3=data67 + vextracti128 xmm4, ymm0, 1 ; xmm2=data45 + vextracti128 xmm2, ymm1, 0 ; xmm1=data23 + vextracti128 xmm0, ymm0, 0 ; xmm0=data01 + + vpshufd xmm1, xmm0, 0x4E ; xmm1=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + vpshufd xmm3, xmm2, 0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + vpshufd xmm5, xmm4, 0x4E ; xmm5=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) + vpshufd xmm7, xmm6, 0x4E ; xmm7=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) + + vzeroupper + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm0 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm1 + + mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm2 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3 + + mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov esi, JSAMPROW [edi+5*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm5 + + mov edx, JSAMPROW [edi+6*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm7 + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-mmx.asm new file mode 100644 index 0000000..c2c17f4 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-mmx.asm @@ -0,0 +1,847 @@ +; +; jidctint.asm - accurate integer IDCT (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2020, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; inverse DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jidctint.c; see the jidctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_islow_mmx) + +EXTN(jconst_idct_islow_mmx): + +PW_F130_F054 times 2 dw (F_0_541 + F_0_765), F_0_541 +PW_F054_MF130 times 2 dw F_0_541, (F_0_541 - F_1_847) +PW_MF078_F117 times 2 dw (F_1_175 - F_1_961), F_1_175 +PW_F117_F078 times 2 dw F_1_175, (F_1_175 - F_0_390) +PW_MF060_MF089 times 2 dw (F_0_298 - F_0_899), -F_0_899 +PW_MF089_F060 times 2 dw -F_0_899, (F_1_501 - F_0_899) +PW_MF050_MF256 times 2 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF256_F050 times 2 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 2 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 2 dd 1 << (DESCALE_P2 - 1) +PB_CENTERJSAMP times 8 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_islow_mmx(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; jpeg_component_info *compptr +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD + ; mmword wk[WK_NUM] +%define WK_NUM 12 +%define workspace wk(0) - DCTSIZE2 * SIZEOF_JCOEF + ; JCOEF workspace[DCTSIZE2] + + align 32 + GLOBAL_FUNCTION(jsimd_idct_islow_mmx) + +EXTN(jsimd_idct_islow_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [workspace] + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input, store into work array. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + lea edi, [workspace] ; JCOEF *wsptr + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.columnloop: +%ifndef NO_ZERO_COLUMN_TEST_ISLOW_MMX + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz short .columnDCT + + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por mm1, mm0 + packsswb mm1, mm1 + movd eax, mm1 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + psllw mm0, PASS1_BITS + + movq mm2, mm0 ; mm0=in0=(00 01 02 03) + punpcklwd mm0, mm0 ; mm0=(00 00 01 01) + punpckhwd mm2, mm2 ; mm2=(02 02 03 03) + + movq mm1, mm0 + punpckldq mm0, mm0 ; mm0=(00 00 00 00) + punpckhdq mm1, mm1 ; mm1=(01 01 01 01) + movq mm3, mm2 + punpckldq mm2, mm2 ; mm2=(02 02 02 02) + punpckhdq mm3, mm3 ; mm3=(03 03 03 03) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm3 + movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm3 + jmp near .nextcolumn + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Even part + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm1, MMWORD [MMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + pmullw mm2, MMWORD [MMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm3, MMWORD [MMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + movq mm4, mm1 ; mm1=in2=z2 + movq mm5, mm1 + punpcklwd mm4, mm3 ; mm3=in6=z3 + punpckhwd mm5, mm3 + movq mm1, mm4 + movq mm3, mm5 + pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L + pmaddwd mm5, [GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H + pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L + pmaddwd mm3, [GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H + + movq mm6, mm0 + paddw mm0, mm2 ; mm0=in0+in4 + psubw mm6, mm2 ; mm6=in0-in4 + + pxor mm7, mm7 + pxor mm2, mm2 + punpcklwd mm7, mm0 ; mm7=tmp0L + punpckhwd mm2, mm0 ; mm2=tmp0H + psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS + psrad mm2, (16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS + + movq mm0, mm7 + paddd mm7, mm4 ; mm7=tmp10L + psubd mm0, mm4 ; mm0=tmp13L + movq mm4, mm2 + paddd mm2, mm5 ; mm2=tmp10H + psubd mm4, mm5 ; mm4=tmp13H + + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L + movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H + movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L + movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H + + pxor mm5, mm5 + pxor mm7, mm7 + punpcklwd mm5, mm6 ; mm5=tmp1L + punpckhwd mm7, mm6 ; mm7=tmp1H + psrad mm5, (16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS + psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS + + movq mm2, mm5 + paddd mm5, mm1 ; mm5=tmp11L + psubd mm2, mm1 ; mm2=tmp12L + movq mm0, mm7 + paddd mm7, mm3 ; mm7=tmp11H + psubd mm0, mm3 ; mm0=tmp12H + + movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L + movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H + movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L + movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H + + ; -- Odd part + + movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw mm4, MMWORD [MMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm6, MMWORD [MMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw mm1, MMWORD [MMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + movq mm5, mm6 + movq mm7, mm4 + paddw mm5, mm3 ; mm5=z3 + paddw mm7, mm1 ; mm7=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movq mm2, mm5 + movq mm0, mm5 + punpcklwd mm2, mm7 + punpckhwd mm0, mm7 + movq mm5, mm2 + movq mm7, mm0 + pmaddwd mm2, [GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L + pmaddwd mm0, [GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H + pmaddwd mm5, [GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L + pmaddwd mm7, [GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H + + movq MMWORD [wk(10)], mm2 ; wk(10)=z3L + movq MMWORD [wk(11)], mm0 ; wk(11)=z3H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + movq mm2, mm3 + movq mm0, mm3 + punpcklwd mm2, mm4 + punpckhwd mm0, mm4 + movq mm3, mm2 + movq mm4, mm0 + pmaddwd mm2, [GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L + pmaddwd mm0, [GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H + pmaddwd mm3, [GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L + pmaddwd mm4, [GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H + + paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L + paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H + paddd mm3, mm5 ; mm3=tmp3L + paddd mm4, mm7 ; mm4=tmp3H + + movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L + movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H + + movq mm2, mm1 + movq mm0, mm1 + punpcklwd mm2, mm6 + punpckhwd mm0, mm6 + movq mm1, mm2 + movq mm6, mm0 + pmaddwd mm2, [GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L + pmaddwd mm0, [GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H + pmaddwd mm1, [GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L + pmaddwd mm6, [GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H + + paddd mm2, mm5 ; mm2=tmp1L + paddd mm0, mm7 ; mm0=tmp1H + paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L + paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H + + movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L + movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H + + ; -- Final output stage + + movq mm5, MMWORD [wk(0)] ; mm5=tmp10L + movq mm7, MMWORD [wk(1)] ; mm7=tmp10H + + movq mm2, mm5 + movq mm0, mm7 + paddd mm5, mm3 ; mm5=data0L + paddd mm7, mm4 ; mm7=data0H + psubd mm2, mm3 ; mm2=data7L + psubd mm0, mm4 ; mm0=data7H + + movq mm3, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm3=[PD_DESCALE_P1] + + paddd mm5, mm3 + paddd mm7, mm3 + psrad mm5, DESCALE_P1 + psrad mm7, DESCALE_P1 + paddd mm2, mm3 + paddd mm0, mm3 + psrad mm2, DESCALE_P1 + psrad mm0, DESCALE_P1 + + packssdw mm5, mm7 ; mm5=data0=(00 01 02 03) + packssdw mm2, mm0 ; mm2=data7=(70 71 72 73) + + movq mm4, MMWORD [wk(4)] ; mm4=tmp11L + movq mm3, MMWORD [wk(5)] ; mm3=tmp11H + + movq mm7, mm4 + movq mm0, mm3 + paddd mm4, mm1 ; mm4=data1L + paddd mm3, mm6 ; mm3=data1H + psubd mm7, mm1 ; mm7=data6L + psubd mm0, mm6 ; mm0=data6H + + movq mm1, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm1=[PD_DESCALE_P1] + + paddd mm4, mm1 + paddd mm3, mm1 + psrad mm4, DESCALE_P1 + psrad mm3, DESCALE_P1 + paddd mm7, mm1 + paddd mm0, mm1 + psrad mm7, DESCALE_P1 + psrad mm0, DESCALE_P1 + + packssdw mm4, mm3 ; mm4=data1=(10 11 12 13) + packssdw mm7, mm0 ; mm7=data6=(60 61 62 63) + + movq mm6, mm5 ; transpose coefficients(phase 1) + punpcklwd mm5, mm4 ; mm5=(00 10 01 11) + punpckhwd mm6, mm4 ; mm6=(02 12 03 13) + movq mm1, mm7 ; transpose coefficients(phase 1) + punpcklwd mm7, mm2 ; mm7=(60 70 61 71) + punpckhwd mm1, mm2 ; mm1=(62 72 63 73) + + movq mm3, MMWORD [wk(6)] ; mm3=tmp12L + movq mm0, MMWORD [wk(7)] ; mm0=tmp12H + movq mm4, MMWORD [wk(10)] ; mm4=tmp1L + movq mm2, MMWORD [wk(11)] ; mm2=tmp1H + + movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 01 11) + movq MMWORD [wk(1)], mm6 ; wk(1)=(02 12 03 13) + movq MMWORD [wk(4)], mm7 ; wk(4)=(60 70 61 71) + movq MMWORD [wk(5)], mm1 ; wk(5)=(62 72 63 73) + + movq mm5, mm3 + movq mm6, mm0 + paddd mm3, mm4 ; mm3=data2L + paddd mm0, mm2 ; mm0=data2H + psubd mm5, mm4 ; mm5=data5L + psubd mm6, mm2 ; mm6=data5H + + movq mm7, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm7=[PD_DESCALE_P1] + + paddd mm3, mm7 + paddd mm0, mm7 + psrad mm3, DESCALE_P1 + psrad mm0, DESCALE_P1 + paddd mm5, mm7 + paddd mm6, mm7 + psrad mm5, DESCALE_P1 + psrad mm6, DESCALE_P1 + + packssdw mm3, mm0 ; mm3=data2=(20 21 22 23) + packssdw mm5, mm6 ; mm5=data5=(50 51 52 53) + + movq mm1, MMWORD [wk(2)] ; mm1=tmp13L + movq mm4, MMWORD [wk(3)] ; mm4=tmp13H + movq mm2, MMWORD [wk(8)] ; mm2=tmp0L + movq mm7, MMWORD [wk(9)] ; mm7=tmp0H + + movq mm0, mm1 + movq mm6, mm4 + paddd mm1, mm2 ; mm1=data3L + paddd mm4, mm7 ; mm4=data3H + psubd mm0, mm2 ; mm0=data4L + psubd mm6, mm7 ; mm6=data4H + + movq mm2, [GOTOFF(ebx,PD_DESCALE_P1)] ; mm2=[PD_DESCALE_P1] + + paddd mm1, mm2 + paddd mm4, mm2 + psrad mm1, DESCALE_P1 + psrad mm4, DESCALE_P1 + paddd mm0, mm2 + paddd mm6, mm2 + psrad mm0, DESCALE_P1 + psrad mm6, DESCALE_P1 + + packssdw mm1, mm4 ; mm1=data3=(30 31 32 33) + packssdw mm0, mm6 ; mm0=data4=(40 41 42 43) + + movq mm7, MMWORD [wk(0)] ; mm7=(00 10 01 11) + movq mm2, MMWORD [wk(1)] ; mm2=(02 12 03 13) + + movq mm4, mm3 ; transpose coefficients(phase 1) + punpcklwd mm3, mm1 ; mm3=(20 30 21 31) + punpckhwd mm4, mm1 ; mm4=(22 32 23 33) + movq mm6, mm0 ; transpose coefficients(phase 1) + punpcklwd mm0, mm5 ; mm0=(40 50 41 51) + punpckhwd mm6, mm5 ; mm6=(42 52 43 53) + + movq mm1, mm7 ; transpose coefficients(phase 2) + punpckldq mm7, mm3 ; mm7=(00 10 20 30) + punpckhdq mm1, mm3 ; mm1=(01 11 21 31) + movq mm5, mm2 ; transpose coefficients(phase 2) + punpckldq mm2, mm4 ; mm2=(02 12 22 32) + punpckhdq mm5, mm4 ; mm5=(03 13 23 33) + + movq mm3, MMWORD [wk(4)] ; mm3=(60 70 61 71) + movq mm4, MMWORD [wk(5)] ; mm4=(62 72 63 73) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm7 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm5 + + movq mm7, mm0 ; transpose coefficients(phase 2) + punpckldq mm0, mm3 ; mm0=(40 50 60 70) + punpckhdq mm7, mm3 ; mm7=(41 51 61 71) + movq mm1, mm6 ; transpose coefficients(phase 2) + punpckldq mm6, mm4 ; mm6=(42 52 62 72) + punpckhdq mm1, mm4 ; mm1=(43 53 63 73) + + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm7 + movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_JCOEF)], mm6 + movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_JCOEF)], mm1 + +.nextcolumn: + add esi, byte 4*SIZEOF_JCOEF ; coef_block + add edx, byte 4*SIZEOF_ISLOW_MULT_TYPE ; quantptr + add edi, byte 4*DCTSIZE*SIZEOF_JCOEF ; wsptr + dec ecx ; ctr + jnz near .columnloop + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + lea esi, [workspace] ; JCOEF *wsptr + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.rowloop: + + ; -- Even part + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movq mm2, MMWORD [MMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + movq mm4, mm1 ; mm1=in2=z2 + movq mm5, mm1 + punpcklwd mm4, mm3 ; mm3=in6=z3 + punpckhwd mm5, mm3 + movq mm1, mm4 + movq mm3, mm5 + pmaddwd mm4, [GOTOFF(ebx,PW_F130_F054)] ; mm4=tmp3L + pmaddwd mm5, [GOTOFF(ebx,PW_F130_F054)] ; mm5=tmp3H + pmaddwd mm1, [GOTOFF(ebx,PW_F054_MF130)] ; mm1=tmp2L + pmaddwd mm3, [GOTOFF(ebx,PW_F054_MF130)] ; mm3=tmp2H + + movq mm6, mm0 + paddw mm0, mm2 ; mm0=in0+in4 + psubw mm6, mm2 ; mm6=in0-in4 + + pxor mm7, mm7 + pxor mm2, mm2 + punpcklwd mm7, mm0 ; mm7=tmp0L + punpckhwd mm2, mm0 ; mm2=tmp0H + psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS + psrad mm2, (16-CONST_BITS) ; psrad mm2,16 & pslld mm2,CONST_BITS + + movq mm0, mm7 + paddd mm7, mm4 ; mm7=tmp10L + psubd mm0, mm4 ; mm0=tmp13L + movq mm4, mm2 + paddd mm2, mm5 ; mm2=tmp10H + psubd mm4, mm5 ; mm4=tmp13H + + movq MMWORD [wk(0)], mm7 ; wk(0)=tmp10L + movq MMWORD [wk(1)], mm2 ; wk(1)=tmp10H + movq MMWORD [wk(2)], mm0 ; wk(2)=tmp13L + movq MMWORD [wk(3)], mm4 ; wk(3)=tmp13H + + pxor mm5, mm5 + pxor mm7, mm7 + punpcklwd mm5, mm6 ; mm5=tmp1L + punpckhwd mm7, mm6 ; mm7=tmp1H + psrad mm5, (16-CONST_BITS) ; psrad mm5,16 & pslld mm5,CONST_BITS + psrad mm7, (16-CONST_BITS) ; psrad mm7,16 & pslld mm7,CONST_BITS + + movq mm2, mm5 + paddd mm5, mm1 ; mm5=tmp11L + psubd mm2, mm1 ; mm2=tmp12L + movq mm0, mm7 + paddd mm7, mm3 ; mm7=tmp11H + psubd mm0, mm3 ; mm0=tmp12H + + movq MMWORD [wk(4)], mm5 ; wk(4)=tmp11L + movq MMWORD [wk(5)], mm7 ; wk(5)=tmp11H + movq MMWORD [wk(6)], mm2 ; wk(6)=tmp12L + movq MMWORD [wk(7)], mm0 ; wk(7)=tmp12H + + ; -- Odd part + + movq mm4, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm6, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + + movq mm5, mm6 + movq mm7, mm4 + paddw mm5, mm3 ; mm5=z3 + paddw mm7, mm1 ; mm7=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movq mm2, mm5 + movq mm0, mm5 + punpcklwd mm2, mm7 + punpckhwd mm0, mm7 + movq mm5, mm2 + movq mm7, mm0 + pmaddwd mm2, [GOTOFF(ebx,PW_MF078_F117)] ; mm2=z3L + pmaddwd mm0, [GOTOFF(ebx,PW_MF078_F117)] ; mm0=z3H + pmaddwd mm5, [GOTOFF(ebx,PW_F117_F078)] ; mm5=z4L + pmaddwd mm7, [GOTOFF(ebx,PW_F117_F078)] ; mm7=z4H + + movq MMWORD [wk(10)], mm2 ; wk(10)=z3L + movq MMWORD [wk(11)], mm0 ; wk(11)=z3H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + movq mm2, mm3 + movq mm0, mm3 + punpcklwd mm2, mm4 + punpckhwd mm0, mm4 + movq mm3, mm2 + movq mm4, mm0 + pmaddwd mm2, [GOTOFF(ebx,PW_MF060_MF089)] ; mm2=tmp0L + pmaddwd mm0, [GOTOFF(ebx,PW_MF060_MF089)] ; mm0=tmp0H + pmaddwd mm3, [GOTOFF(ebx,PW_MF089_F060)] ; mm3=tmp3L + pmaddwd mm4, [GOTOFF(ebx,PW_MF089_F060)] ; mm4=tmp3H + + paddd mm2, MMWORD [wk(10)] ; mm2=tmp0L + paddd mm0, MMWORD [wk(11)] ; mm0=tmp0H + paddd mm3, mm5 ; mm3=tmp3L + paddd mm4, mm7 ; mm4=tmp3H + + movq MMWORD [wk(8)], mm2 ; wk(8)=tmp0L + movq MMWORD [wk(9)], mm0 ; wk(9)=tmp0H + + movq mm2, mm1 + movq mm0, mm1 + punpcklwd mm2, mm6 + punpckhwd mm0, mm6 + movq mm1, mm2 + movq mm6, mm0 + pmaddwd mm2, [GOTOFF(ebx,PW_MF050_MF256)] ; mm2=tmp1L + pmaddwd mm0, [GOTOFF(ebx,PW_MF050_MF256)] ; mm0=tmp1H + pmaddwd mm1, [GOTOFF(ebx,PW_MF256_F050)] ; mm1=tmp2L + pmaddwd mm6, [GOTOFF(ebx,PW_MF256_F050)] ; mm6=tmp2H + + paddd mm2, mm5 ; mm2=tmp1L + paddd mm0, mm7 ; mm0=tmp1H + paddd mm1, MMWORD [wk(10)] ; mm1=tmp2L + paddd mm6, MMWORD [wk(11)] ; mm6=tmp2H + + movq MMWORD [wk(10)], mm2 ; wk(10)=tmp1L + movq MMWORD [wk(11)], mm0 ; wk(11)=tmp1H + + ; -- Final output stage + + movq mm5, MMWORD [wk(0)] ; mm5=tmp10L + movq mm7, MMWORD [wk(1)] ; mm7=tmp10H + + movq mm2, mm5 + movq mm0, mm7 + paddd mm5, mm3 ; mm5=data0L + paddd mm7, mm4 ; mm7=data0H + psubd mm2, mm3 ; mm2=data7L + psubd mm0, mm4 ; mm0=data7H + + movq mm3, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm3=[PD_DESCALE_P2] + + paddd mm5, mm3 + paddd mm7, mm3 + psrad mm5, DESCALE_P2 + psrad mm7, DESCALE_P2 + paddd mm2, mm3 + paddd mm0, mm3 + psrad mm2, DESCALE_P2 + psrad mm0, DESCALE_P2 + + packssdw mm5, mm7 ; mm5=data0=(00 10 20 30) + packssdw mm2, mm0 ; mm2=data7=(07 17 27 37) + + movq mm4, MMWORD [wk(4)] ; mm4=tmp11L + movq mm3, MMWORD [wk(5)] ; mm3=tmp11H + + movq mm7, mm4 + movq mm0, mm3 + paddd mm4, mm1 ; mm4=data1L + paddd mm3, mm6 ; mm3=data1H + psubd mm7, mm1 ; mm7=data6L + psubd mm0, mm6 ; mm0=data6H + + movq mm1, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm1=[PD_DESCALE_P2] + + paddd mm4, mm1 + paddd mm3, mm1 + psrad mm4, DESCALE_P2 + psrad mm3, DESCALE_P2 + paddd mm7, mm1 + paddd mm0, mm1 + psrad mm7, DESCALE_P2 + psrad mm0, DESCALE_P2 + + packssdw mm4, mm3 ; mm4=data1=(01 11 21 31) + packssdw mm7, mm0 ; mm7=data6=(06 16 26 36) + + packsswb mm5, mm7 ; mm5=(00 10 20 30 06 16 26 36) + packsswb mm4, mm2 ; mm4=(01 11 21 31 07 17 27 37) + + movq mm6, MMWORD [wk(6)] ; mm6=tmp12L + movq mm1, MMWORD [wk(7)] ; mm1=tmp12H + movq mm3, MMWORD [wk(10)] ; mm3=tmp1L + movq mm0, MMWORD [wk(11)] ; mm0=tmp1H + + movq MMWORD [wk(0)], mm5 ; wk(0)=(00 10 20 30 06 16 26 36) + movq MMWORD [wk(1)], mm4 ; wk(1)=(01 11 21 31 07 17 27 37) + + movq mm7, mm6 + movq mm2, mm1 + paddd mm6, mm3 ; mm6=data2L + paddd mm1, mm0 ; mm1=data2H + psubd mm7, mm3 ; mm7=data5L + psubd mm2, mm0 ; mm2=data5H + + movq mm5, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm5=[PD_DESCALE_P2] + + paddd mm6, mm5 + paddd mm1, mm5 + psrad mm6, DESCALE_P2 + psrad mm1, DESCALE_P2 + paddd mm7, mm5 + paddd mm2, mm5 + psrad mm7, DESCALE_P2 + psrad mm2, DESCALE_P2 + + packssdw mm6, mm1 ; mm6=data2=(02 12 22 32) + packssdw mm7, mm2 ; mm7=data5=(05 15 25 35) + + movq mm4, MMWORD [wk(2)] ; mm4=tmp13L + movq mm3, MMWORD [wk(3)] ; mm3=tmp13H + movq mm0, MMWORD [wk(8)] ; mm0=tmp0L + movq mm5, MMWORD [wk(9)] ; mm5=tmp0H + + movq mm1, mm4 + movq mm2, mm3 + paddd mm4, mm0 ; mm4=data3L + paddd mm3, mm5 ; mm3=data3H + psubd mm1, mm0 ; mm1=data4L + psubd mm2, mm5 ; mm2=data4H + + movq mm0, [GOTOFF(ebx,PD_DESCALE_P2)] ; mm0=[PD_DESCALE_P2] + + paddd mm4, mm0 + paddd mm3, mm0 + psrad mm4, DESCALE_P2 + psrad mm3, DESCALE_P2 + paddd mm1, mm0 + paddd mm2, mm0 + psrad mm1, DESCALE_P2 + psrad mm2, DESCALE_P2 + + movq mm5, [GOTOFF(ebx,PB_CENTERJSAMP)] ; mm5=[PB_CENTERJSAMP] + + packssdw mm4, mm3 ; mm4=data3=(03 13 23 33) + packssdw mm1, mm2 ; mm1=data4=(04 14 24 34) + + movq mm0, MMWORD [wk(0)] ; mm0=(00 10 20 30 06 16 26 36) + movq mm3, MMWORD [wk(1)] ; mm3=(01 11 21 31 07 17 27 37) + + packsswb mm6, mm1 ; mm6=(02 12 22 32 04 14 24 34) + packsswb mm4, mm7 ; mm4=(03 13 23 33 05 15 25 35) + + paddb mm0, mm5 + paddb mm3, mm5 + paddb mm6, mm5 + paddb mm4, mm5 + + movq mm2, mm0 ; transpose coefficients(phase 1) + punpcklbw mm0, mm3 ; mm0=(00 01 10 11 20 21 30 31) + punpckhbw mm2, mm3 ; mm2=(06 07 16 17 26 27 36 37) + movq mm1, mm6 ; transpose coefficients(phase 1) + punpcklbw mm6, mm4 ; mm6=(02 03 12 13 22 23 32 33) + punpckhbw mm1, mm4 ; mm1=(04 05 14 15 24 25 34 35) + + movq mm7, mm0 ; transpose coefficients(phase 2) + punpcklwd mm0, mm6 ; mm0=(00 01 02 03 10 11 12 13) + punpckhwd mm7, mm6 ; mm7=(20 21 22 23 30 31 32 33) + movq mm5, mm1 ; transpose coefficients(phase 2) + punpcklwd mm1, mm2 ; mm1=(04 05 06 07 14 15 16 17) + punpckhwd mm5, mm2 ; mm5=(24 25 26 27 34 35 36 37) + + movq mm3, mm0 ; transpose coefficients(phase 3) + punpckldq mm0, mm1 ; mm0=(00 01 02 03 04 05 06 07) + punpckhdq mm3, mm1 ; mm3=(10 11 12 13 14 15 16 17) + movq mm4, mm7 ; transpose coefficients(phase 3) + punpckldq mm7, mm5 ; mm7=(20 21 22 23 24 25 26 27) + punpckhdq mm4, mm5 ; mm4=(30 31 32 33 34 35 36 37) + + PUSHPIC ebx ; save GOT address + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm0 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm3 + mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + mov ebx, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movq MMWORD [edx+eax*SIZEOF_JSAMPLE], mm7 + movq MMWORD [ebx+eax*SIZEOF_JSAMPLE], mm4 + + POPPIC ebx ; restore GOT address + + add esi, byte 4*SIZEOF_JCOEF ; wsptr + add edi, byte 4*SIZEOF_JSAMPROW + dec ecx ; ctr + jnz near .rowloop + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-sse2.asm new file mode 100644 index 0000000..70516ca --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctint-sse2.asm @@ -0,0 +1,854 @@ +; +; jidctint.asm - accurate integer IDCT (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2020, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; inverse DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jidctint.c; see the jidctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_islow_sse2) + +EXTN(jconst_idct_islow_sse2): + +PW_F130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 +PW_F054_MF130 times 4 dw F_0_541, (F_0_541 - F_1_847) +PW_MF078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 +PW_F117_F078 times 4 dw F_1_175, (F_1_175 - F_0_390) +PW_MF060_MF089 times 4 dw (F_0_298 - F_0_899), -F_0_899 +PW_MF089_F060 times 4 dw -F_0_899, (F_1_501 - F_0_899) +PW_MF050_MF256 times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF256_F050 times 4 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2 - 1) +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_islow_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; jpeg_component_info *compptr +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 12 + + align 32 + GLOBAL_FUNCTION(jsimd_idct_islow_sse2) + +EXTN(jsimd_idct_islow_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + +%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2 + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por xmm1, xmm0 + packsswb xmm1, xmm1 + packsswb xmm1, xmm1 + movd eax, xmm1 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm5, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + pmullw xmm5, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + psllw xmm5, PASS1_BITS + + movdqa xmm4, xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07) + punpcklwd xmm5, xmm5 ; xmm5=(00 00 01 01 02 02 03 03) + punpckhwd xmm4, xmm4 ; xmm4=(04 04 05 05 06 06 07 07) + + pshufd xmm7, xmm5, 0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00) + pshufd xmm6, xmm5, 0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01) + pshufd xmm1, xmm5, 0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02) + pshufd xmm5, xmm5, 0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03) + pshufd xmm0, xmm4, 0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04) + pshufd xmm3, xmm4, 0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05) + pshufd xmm2, xmm4, 0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06) + pshufd xmm4, xmm4, 0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07) + + movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1 + movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3 + movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 + movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 + jmp near .column_end + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Even part + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(4,0,esi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(4,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + movdqa xmm4, xmm1 ; xmm1=in2=z2 + movdqa xmm5, xmm1 + punpcklwd xmm4, xmm3 ; xmm3=in6=z3 + punpckhwd xmm5, xmm3 + movdqa xmm1, xmm4 + movdqa xmm3, xmm5 + pmaddwd xmm4, [GOTOFF(ebx,PW_F130_F054)] ; xmm4=tmp3L + pmaddwd xmm5, [GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H + pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L + pmaddwd xmm3, [GOTOFF(ebx,PW_F054_MF130)] ; xmm3=tmp2H + + movdqa xmm6, xmm0 + paddw xmm0, xmm2 ; xmm0=in0+in4 + psubw xmm6, xmm2 ; xmm6=in0-in4 + + pxor xmm7, xmm7 + pxor xmm2, xmm2 + punpcklwd xmm7, xmm0 ; xmm7=tmp0L + punpckhwd xmm2, xmm0 ; xmm2=tmp0H + psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS + psrad xmm2, (16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS + + movdqa xmm0, xmm7 + paddd xmm7, xmm4 ; xmm7=tmp10L + psubd xmm0, xmm4 ; xmm0=tmp13L + movdqa xmm4, xmm2 + paddd xmm2, xmm5 ; xmm2=tmp10H + psubd xmm4, xmm5 ; xmm4=tmp13H + + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L + movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H + movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L + movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H + + pxor xmm5, xmm5 + pxor xmm7, xmm7 + punpcklwd xmm5, xmm6 ; xmm5=tmp1L + punpckhwd xmm7, xmm6 ; xmm7=tmp1H + psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS + psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS + + movdqa xmm2, xmm5 + paddd xmm5, xmm1 ; xmm5=tmp11L + psubd xmm2, xmm1 ; xmm2=tmp12L + movdqa xmm0, xmm7 + paddd xmm7, xmm3 ; xmm7=tmp11H + psubd xmm0, xmm3 ; xmm0=tmp12H + + movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L + movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H + movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L + movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H + + ; -- Odd part + + movdqa xmm4, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm6, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw xmm4, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm6, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw xmm1, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + movdqa xmm5, xmm6 + movdqa xmm7, xmm4 + paddw xmm5, xmm3 ; xmm5=z3 + paddw xmm7, xmm1 ; xmm7=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm2, xmm5 + movdqa xmm0, xmm5 + punpcklwd xmm2, xmm7 + punpckhwd xmm0, xmm7 + movdqa xmm5, xmm2 + movdqa xmm7, xmm0 + pmaddwd xmm2, [GOTOFF(ebx,PW_MF078_F117)] ; xmm2=z3L + pmaddwd xmm0, [GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3H + pmaddwd xmm5, [GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L + pmaddwd xmm7, [GOTOFF(ebx,PW_F117_F078)] ; xmm7=z4H + + movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L + movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + movdqa xmm2, xmm3 + movdqa xmm0, xmm3 + punpcklwd xmm2, xmm4 + punpckhwd xmm0, xmm4 + movdqa xmm3, xmm2 + movdqa xmm4, xmm0 + pmaddwd xmm2, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm2=tmp0L + pmaddwd xmm0, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0H + pmaddwd xmm3, [GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3L + pmaddwd xmm4, [GOTOFF(ebx,PW_MF089_F060)] ; xmm4=tmp3H + + paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L + paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H + paddd xmm3, xmm5 ; xmm3=tmp3L + paddd xmm4, xmm7 ; xmm4=tmp3H + + movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L + movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H + + movdqa xmm2, xmm1 + movdqa xmm0, xmm1 + punpcklwd xmm2, xmm6 + punpckhwd xmm0, xmm6 + movdqa xmm1, xmm2 + movdqa xmm6, xmm0 + pmaddwd xmm2, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm2=tmp1L + pmaddwd xmm0, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1H + pmaddwd xmm1, [GOTOFF(ebx,PW_MF256_F050)] ; xmm1=tmp2L + pmaddwd xmm6, [GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H + + paddd xmm2, xmm5 ; xmm2=tmp1L + paddd xmm0, xmm7 ; xmm0=tmp1H + paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L + paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H + + movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L + movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H + + ; -- Final output stage + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L + movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H + + movdqa xmm2, xmm5 + movdqa xmm0, xmm7 + paddd xmm5, xmm3 ; xmm5=data0L + paddd xmm7, xmm4 ; xmm7=data0H + psubd xmm2, xmm3 ; xmm2=data7L + psubd xmm0, xmm4 ; xmm0=data7H + + movdqa xmm3, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm3=[PD_DESCALE_P1] + + paddd xmm5, xmm3 + paddd xmm7, xmm3 + psrad xmm5, DESCALE_P1 + psrad xmm7, DESCALE_P1 + paddd xmm2, xmm3 + paddd xmm0, xmm3 + psrad xmm2, DESCALE_P1 + psrad xmm0, DESCALE_P1 + + packssdw xmm5, xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07) + packssdw xmm2, xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77) + + movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L + movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H + + movdqa xmm7, xmm4 + movdqa xmm0, xmm3 + paddd xmm4, xmm1 ; xmm4=data1L + paddd xmm3, xmm6 ; xmm3=data1H + psubd xmm7, xmm1 ; xmm7=data6L + psubd xmm0, xmm6 ; xmm0=data6H + + movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm1=[PD_DESCALE_P1] + + paddd xmm4, xmm1 + paddd xmm3, xmm1 + psrad xmm4, DESCALE_P1 + psrad xmm3, DESCALE_P1 + paddd xmm7, xmm1 + paddd xmm0, xmm1 + psrad xmm7, DESCALE_P1 + psrad xmm0, DESCALE_P1 + + packssdw xmm4, xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17) + packssdw xmm7, xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67) + + movdqa xmm6, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm4 ; xmm5=(00 10 01 11 02 12 03 13) + punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17) + movdqa xmm1, xmm7 ; transpose coefficients(phase 1) + punpcklwd xmm7, xmm2 ; xmm7=(60 70 61 71 62 72 63 73) + punpckhwd xmm1, xmm2 ; xmm1=(64 74 65 75 66 76 67 77) + + movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L + movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H + movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L + movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H + + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13) + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17) + movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73) + movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77) + + movdqa xmm5, xmm3 + movdqa xmm6, xmm0 + paddd xmm3, xmm4 ; xmm3=data2L + paddd xmm0, xmm2 ; xmm0=data2H + psubd xmm5, xmm4 ; xmm5=data5L + psubd xmm6, xmm2 ; xmm6=data5H + + movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm7=[PD_DESCALE_P1] + + paddd xmm3, xmm7 + paddd xmm0, xmm7 + psrad xmm3, DESCALE_P1 + psrad xmm0, DESCALE_P1 + paddd xmm5, xmm7 + paddd xmm6, xmm7 + psrad xmm5, DESCALE_P1 + psrad xmm6, DESCALE_P1 + + packssdw xmm3, xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27) + packssdw xmm5, xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57) + + movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L + movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H + movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L + movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H + + movdqa xmm0, xmm1 + movdqa xmm6, xmm4 + paddd xmm1, xmm2 ; xmm1=data3L + paddd xmm4, xmm7 ; xmm4=data3H + psubd xmm0, xmm2 ; xmm0=data4L + psubd xmm6, xmm7 ; xmm6=data4H + + movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P1)] ; xmm2=[PD_DESCALE_P1] + + paddd xmm1, xmm2 + paddd xmm4, xmm2 + psrad xmm1, DESCALE_P1 + psrad xmm4, DESCALE_P1 + paddd xmm0, xmm2 + paddd xmm6, xmm2 + psrad xmm0, DESCALE_P1 + psrad xmm6, DESCALE_P1 + + packssdw xmm1, xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37) + packssdw xmm0, xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13) + movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17) + + movdqa xmm4, xmm3 ; transpose coefficients(phase 1) + punpcklwd xmm3, xmm1 ; xmm3=(20 30 21 31 22 32 23 33) + punpckhwd xmm4, xmm1 ; xmm4=(24 34 25 35 26 36 27 37) + movdqa xmm6, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm5 ; xmm0=(40 50 41 51 42 52 43 53) + punpckhwd xmm6, xmm5 ; xmm6=(44 54 45 55 46 56 47 57) + + movdqa xmm1, xmm7 ; transpose coefficients(phase 2) + punpckldq xmm7, xmm3 ; xmm7=(00 10 20 30 01 11 21 31) + punpckhdq xmm1, xmm3 ; xmm1=(02 12 22 32 03 13 23 33) + movdqa xmm5, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm4 ; xmm2=(04 14 24 34 05 15 25 35) + punpckhdq xmm5, xmm4 ; xmm5=(06 16 26 36 07 17 27 37) + + movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73) + movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77) + + movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35) + movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37) + + movdqa xmm2, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm3 ; xmm0=(40 50 60 70 41 51 61 71) + punpckhdq xmm2, xmm3 ; xmm2=(42 52 62 72 43 53 63 73) + movdqa xmm5, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm4 ; xmm6=(44 54 64 74 45 55 65 75) + punpckhdq xmm5, xmm4 ; xmm5=(46 56 66 76 47 57 67 77) + + movdqa xmm3, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70) + punpckhqdq xmm3, xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71) + movdqa xmm4, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72) + punpckhqdq xmm4, xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73) + + movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35) + movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37) + + movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1 + movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3 + + movdqa xmm3, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74) + punpckhqdq xmm3, xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75) + movdqa xmm4, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76) + punpckhqdq xmm4, xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77) + + movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 + movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + + ; -- Even part + + ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6 + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + movdqa xmm6, xmm1 ; xmm1=in2=z2 + movdqa xmm5, xmm1 + punpcklwd xmm6, xmm2 ; xmm2=in6=z3 + punpckhwd xmm5, xmm2 + movdqa xmm1, xmm6 + movdqa xmm2, xmm5 + pmaddwd xmm6, [GOTOFF(ebx,PW_F130_F054)] ; xmm6=tmp3L + pmaddwd xmm5, [GOTOFF(ebx,PW_F130_F054)] ; xmm5=tmp3H + pmaddwd xmm1, [GOTOFF(ebx,PW_F054_MF130)] ; xmm1=tmp2L + pmaddwd xmm2, [GOTOFF(ebx,PW_F054_MF130)] ; xmm2=tmp2H + + movdqa xmm3, xmm7 + paddw xmm7, xmm0 ; xmm7=in0+in4 + psubw xmm3, xmm0 ; xmm3=in0-in4 + + pxor xmm4, xmm4 + pxor xmm0, xmm0 + punpcklwd xmm4, xmm7 ; xmm4=tmp0L + punpckhwd xmm0, xmm7 ; xmm0=tmp0H + psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS + psrad xmm0, (16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS + + movdqa xmm7, xmm4 + paddd xmm4, xmm6 ; xmm4=tmp10L + psubd xmm7, xmm6 ; xmm7=tmp13L + movdqa xmm6, xmm0 + paddd xmm0, xmm5 ; xmm0=tmp10H + psubd xmm6, xmm5 ; xmm6=tmp13H + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H + movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L + movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H + + pxor xmm5, xmm5 + pxor xmm4, xmm4 + punpcklwd xmm5, xmm3 ; xmm5=tmp1L + punpckhwd xmm4, xmm3 ; xmm4=tmp1H + psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS + psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS + + movdqa xmm0, xmm5 + paddd xmm5, xmm1 ; xmm5=tmp11L + psubd xmm0, xmm1 ; xmm0=tmp12L + movdqa xmm7, xmm4 + paddd xmm4, xmm2 ; xmm4=tmp11H + psubd xmm7, xmm2 ; xmm7=tmp12H + + movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L + movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H + movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L + movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H + + ; -- Odd part + + movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3 + movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1 + movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7 + movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5 + + movdqa xmm5, xmm6 + movdqa xmm4, xmm3 + paddw xmm5, xmm1 ; xmm5=z3 + paddw xmm4, xmm2 ; xmm4=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm0, xmm5 + movdqa xmm7, xmm5 + punpcklwd xmm0, xmm4 + punpckhwd xmm7, xmm4 + movdqa xmm5, xmm0 + movdqa xmm4, xmm7 + pmaddwd xmm0, [GOTOFF(ebx,PW_MF078_F117)] ; xmm0=z3L + pmaddwd xmm7, [GOTOFF(ebx,PW_MF078_F117)] ; xmm7=z3H + pmaddwd xmm5, [GOTOFF(ebx,PW_F117_F078)] ; xmm5=z4L + pmaddwd xmm4, [GOTOFF(ebx,PW_F117_F078)] ; xmm4=z4H + + movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L + movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + movdqa xmm0, xmm1 + movdqa xmm7, xmm1 + punpcklwd xmm0, xmm3 + punpckhwd xmm7, xmm3 + movdqa xmm1, xmm0 + movdqa xmm3, xmm7 + pmaddwd xmm0, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm0=tmp0L + pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF089)] ; xmm7=tmp0H + pmaddwd xmm1, [GOTOFF(ebx,PW_MF089_F060)] ; xmm1=tmp3L + pmaddwd xmm3, [GOTOFF(ebx,PW_MF089_F060)] ; xmm3=tmp3H + + paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L + paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H + paddd xmm1, xmm5 ; xmm1=tmp3L + paddd xmm3, xmm4 ; xmm3=tmp3H + + movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L + movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H + + movdqa xmm0, xmm2 + movdqa xmm7, xmm2 + punpcklwd xmm0, xmm6 + punpckhwd xmm7, xmm6 + movdqa xmm2, xmm0 + movdqa xmm6, xmm7 + pmaddwd xmm0, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm0=tmp1L + pmaddwd xmm7, [GOTOFF(ebx,PW_MF050_MF256)] ; xmm7=tmp1H + pmaddwd xmm2, [GOTOFF(ebx,PW_MF256_F050)] ; xmm2=tmp2L + pmaddwd xmm6, [GOTOFF(ebx,PW_MF256_F050)] ; xmm6=tmp2H + + paddd xmm0, xmm5 ; xmm0=tmp1L + paddd xmm7, xmm4 ; xmm7=tmp1H + paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L + paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H + + movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L + movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H + + ; -- Final output stage + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L + movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H + + movdqa xmm0, xmm5 + movdqa xmm7, xmm4 + paddd xmm5, xmm1 ; xmm5=data0L + paddd xmm4, xmm3 ; xmm4=data0H + psubd xmm0, xmm1 ; xmm0=data7L + psubd xmm7, xmm3 ; xmm7=data7H + + movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm1=[PD_DESCALE_P2] + + paddd xmm5, xmm1 + paddd xmm4, xmm1 + psrad xmm5, DESCALE_P2 + psrad xmm4, DESCALE_P2 + paddd xmm0, xmm1 + paddd xmm7, xmm1 + psrad xmm0, DESCALE_P2 + psrad xmm7, DESCALE_P2 + + packssdw xmm5, xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70) + packssdw xmm0, xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77) + + movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L + movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H + + movdqa xmm4, xmm3 + movdqa xmm7, xmm1 + paddd xmm3, xmm2 ; xmm3=data1L + paddd xmm1, xmm6 ; xmm1=data1H + psubd xmm4, xmm2 ; xmm4=data6L + psubd xmm7, xmm6 ; xmm7=data6H + + movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm2=[PD_DESCALE_P2] + + paddd xmm3, xmm2 + paddd xmm1, xmm2 + psrad xmm3, DESCALE_P2 + psrad xmm1, DESCALE_P2 + paddd xmm4, xmm2 + paddd xmm7, xmm2 + psrad xmm4, DESCALE_P2 + psrad xmm7, DESCALE_P2 + + packssdw xmm3, xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71) + packssdw xmm4, xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76) + + packsswb xmm5, xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + packsswb xmm3, xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L + movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H + movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L + movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H + + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + movdqa xmm4, xmm6 + movdqa xmm0, xmm2 + paddd xmm6, xmm1 ; xmm6=data2L + paddd xmm2, xmm7 ; xmm2=data2H + psubd xmm4, xmm1 ; xmm4=data5L + psubd xmm0, xmm7 ; xmm0=data5H + + movdqa xmm5, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm5=[PD_DESCALE_P2] + + paddd xmm6, xmm5 + paddd xmm2, xmm5 + psrad xmm6, DESCALE_P2 + psrad xmm2, DESCALE_P2 + paddd xmm4, xmm5 + paddd xmm0, xmm5 + psrad xmm4, DESCALE_P2 + psrad xmm0, DESCALE_P2 + + packssdw xmm6, xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72) + packssdw xmm4, xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75) + + movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L + movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H + movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L + movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H + + movdqa xmm2, xmm3 + movdqa xmm0, xmm1 + paddd xmm3, xmm7 ; xmm3=data3L + paddd xmm1, xmm5 ; xmm1=data3H + psubd xmm2, xmm7 ; xmm2=data4L + psubd xmm0, xmm5 ; xmm0=data4H + + movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P2)] ; xmm7=[PD_DESCALE_P2] + + paddd xmm3, xmm7 + paddd xmm1, xmm7 + psrad xmm3, DESCALE_P2 + psrad xmm1, DESCALE_P2 + paddd xmm2, xmm7 + paddd xmm0, xmm7 + psrad xmm2, DESCALE_P2 + psrad xmm0, DESCALE_P2 + + movdqa xmm5, [GOTOFF(ebx,PB_CENTERJSAMP)] ; xmm5=[PB_CENTERJSAMP] + + packssdw xmm3, xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73) + packssdw xmm2, xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + packsswb xmm6, xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74) + packsswb xmm3, xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75) + + paddb xmm7, xmm5 + paddb xmm1, xmm5 + paddb xmm6, xmm5 + paddb xmm3, xmm5 + + movdqa xmm0, xmm7 ; transpose coefficients(phase 1) + punpcklbw xmm7, xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71) + punpckhbw xmm0, xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77) + movdqa xmm2, xmm6 ; transpose coefficients(phase 1) + punpcklbw xmm6, xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73) + punpckhbw xmm2, xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75) + + movdqa xmm4, xmm7 ; transpose coefficients(phase 2) + punpcklwd xmm7, xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) + punpckhwd xmm4, xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73) + movdqa xmm5, xmm2 ; transpose coefficients(phase 2) + punpcklwd xmm2, xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) + punpckhwd xmm5, xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77) + + movdqa xmm1, xmm7 ; transpose coefficients(phase 3) + punpckldq xmm7, xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + punpckhdq xmm1, xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + movdqa xmm3, xmm4 ; transpose coefficients(phase 3) + punpckldq xmm4, xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) + punpckhdq xmm3, xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) + + pshufd xmm6, xmm7, 0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + pshufd xmm0, xmm1, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + pshufd xmm2, xmm4, 0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) + pshufd xmm5, xmm3, 0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm7 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm1 + mov edx, JSAMPROW [edi+4*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+6*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm4 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm3 + + mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm0 + mov edx, JSAMPROW [edi+5*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+7*SIZEOF_JSAMPROW] + movq XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE], xmm2 + movq XMM_MMWORD [esi+eax*SIZEOF_JSAMPLE], xmm5 + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-mmx.asm new file mode 100644 index 0000000..96cda65 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-mmx.asm @@ -0,0 +1,700 @@ +; +; jidctred.asm - reduced-size IDCT (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains inverse-DCT routines that produce reduced-size +; output: either 4x4 or 2x2 pixels from an 8x8 DCT block. +; The following code is based directly on the IJG's original jidctred.c; +; see the jidctred.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1_4 (CONST_BITS - PASS1_BITS + 1) +%define DESCALE_P2_4 (CONST_BITS + PASS1_BITS + 3 + 1) +%define DESCALE_P1_2 (CONST_BITS - PASS1_BITS + 2) +%define DESCALE_P2_2 (CONST_BITS + PASS1_BITS + 3 + 2) + +%if CONST_BITS == 13 +F_0_211 equ 1730 ; FIX(0.211164243) +F_0_509 equ 4176 ; FIX(0.509795579) +F_0_601 equ 4926 ; FIX(0.601344887) +F_0_720 equ 5906 ; FIX(0.720959822) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_850 equ 6967 ; FIX(0.850430095) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_061 equ 8697 ; FIX(1.061594337) +F_1_272 equ 10426 ; FIX(1.272758580) +F_1_451 equ 11893 ; FIX(1.451774981) +F_1_847 equ 15137 ; FIX(1.847759065) +F_2_172 equ 17799 ; FIX(2.172734803) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_624 equ 29692 ; FIX(3.624509785) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_211 equ DESCALE( 226735879, 30 - CONST_BITS) ; FIX(0.211164243) +F_0_509 equ DESCALE( 547388834, 30 - CONST_BITS) ; FIX(0.509795579) +F_0_601 equ DESCALE( 645689155, 30 - CONST_BITS) ; FIX(0.601344887) +F_0_720 equ DESCALE( 774124714, 30 - CONST_BITS) ; FIX(0.720959822) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_850 equ DESCALE( 913142361, 30 - CONST_BITS) ; FIX(0.850430095) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_061 equ DESCALE(1139878239, 30 - CONST_BITS) ; FIX(1.061594337) +F_1_272 equ DESCALE(1366614119, 30 - CONST_BITS) ; FIX(1.272758580) +F_1_451 equ DESCALE(1558831516, 30 - CONST_BITS) ; FIX(1.451774981) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_2_172 equ DESCALE(2332956230, 30 - CONST_BITS) ; FIX(2.172734803) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_624 equ DESCALE(3891787747, 30 - CONST_BITS) ; FIX(3.624509785) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_red_mmx) + +EXTN(jconst_idct_red_mmx): + +PW_F184_MF076 times 2 dw F_1_847, -F_0_765 +PW_F256_F089 times 2 dw F_2_562, F_0_899 +PW_F106_MF217 times 2 dw F_1_061, -F_2_172 +PW_MF060_MF050 times 2 dw -F_0_601, -F_0_509 +PW_F145_MF021 times 2 dw F_1_451, -F_0_211 +PW_F362_MF127 times 2 dw F_3_624, -F_1_272 +PW_F085_MF072 times 2 dw F_0_850, -F_0_720 +PD_DESCALE_P1_4 times 2 dd 1 << (DESCALE_P1_4 - 1) +PD_DESCALE_P2_4 times 2 dd 1 << (DESCALE_P2_4 - 1) +PD_DESCALE_P1_2 times 2 dd 1 << (DESCALE_P1_2 - 1) +PD_DESCALE_P2_2 times 2 dd 1 << (DESCALE_P2_2 - 1) +PB_CENTERJSAMP times 8 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients, +; producing a reduced-size 4x4 output block. +; +; GLOBAL(void) +; jsimd_idct_4x4_mmx(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; void *dct_table +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_MMWORD + ; mmword wk[WK_NUM] +%define WK_NUM 2 +%define workspace wk(0) - DCTSIZE2 * SIZEOF_JCOEF + ; JCOEF workspace[DCTSIZE2] + + align 32 + GLOBAL_FUNCTION(jsimd_idct_4x4_mmx) + +EXTN(jsimd_idct_4x4_mmx): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_MMWORD) ; align to 64 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [workspace] + PUSHPIC ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input, store into work array. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + lea edi, [workspace] ; JCOEF *wsptr + mov ecx, DCTSIZE/4 ; ctr + ALIGNX 16, 7 +.columnloop: +%ifndef NO_ZERO_COLUMN_TEST_4X4_MMX + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz short .columnDCT + + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + por mm0, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + por mm1, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por mm0, mm1 + packsswb mm0, mm0 + movd eax, mm0 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + pmullw mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + psllw mm0, PASS1_BITS + + movq mm2, mm0 ; mm0=in0=(00 01 02 03) + punpcklwd mm0, mm0 ; mm0=(00 00 01 01) + punpckhwd mm2, mm2 ; mm2=(02 02 03 03) + + movq mm1, mm0 + punpckldq mm0, mm0 ; mm0=(00 00 00 00) + punpckhdq mm1, mm1 ; mm1=(01 01 01 01) + movq mm3, mm2 + punpckldq mm2, mm2 ; mm2=(02 02 02 02) + punpckhdq mm3, mm3 ; mm3=(03 03 03 03) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm3 + jmp near .nextcolumn + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Odd part + + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw mm0, MMWORD [MMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movq mm2, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw mm2, MMWORD [MMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + movq mm4, mm0 + movq mm5, mm0 + punpcklwd mm4, mm1 + punpckhwd mm5, mm1 + movq mm0, mm4 + movq mm1, mm5 + pmaddwd mm4, [GOTOFF(ebx,PW_F256_F089)] ; mm4=(tmp2L) + pmaddwd mm5, [GOTOFF(ebx,PW_F256_F089)] ; mm5=(tmp2H) + pmaddwd mm0, [GOTOFF(ebx,PW_F106_MF217)] ; mm0=(tmp0L) + pmaddwd mm1, [GOTOFF(ebx,PW_F106_MF217)] ; mm1=(tmp0H) + + movq mm6, mm2 + movq mm7, mm2 + punpcklwd mm6, mm3 + punpckhwd mm7, mm3 + movq mm2, mm6 + movq mm3, mm7 + pmaddwd mm6, [GOTOFF(ebx,PW_MF060_MF050)] ; mm6=(tmp2L) + pmaddwd mm7, [GOTOFF(ebx,PW_MF060_MF050)] ; mm7=(tmp2H) + pmaddwd mm2, [GOTOFF(ebx,PW_F145_MF021)] ; mm2=(tmp0L) + pmaddwd mm3, [GOTOFF(ebx,PW_F145_MF021)] ; mm3=(tmp0H) + + paddd mm6, mm4 ; mm6=tmp2L + paddd mm7, mm5 ; mm7=tmp2H + paddd mm2, mm0 ; mm2=tmp0L + paddd mm3, mm1 ; mm3=tmp0H + + movq MMWORD [wk(0)], mm2 ; wk(0)=tmp0L + movq MMWORD [wk(1)], mm3 ; wk(1)=tmp0H + + ; -- Even part + + movq mm4, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm5, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movq mm0, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + pmullw mm4, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm5, MMWORD [MMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm0, MMWORD [MMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + pxor mm1, mm1 + pxor mm2, mm2 + punpcklwd mm1, mm4 ; mm1=tmp0L + punpckhwd mm2, mm4 ; mm2=tmp0H + psrad mm1, (16-CONST_BITS-1) ; psrad mm1,16 & pslld mm1,CONST_BITS+1 + psrad mm2, (16-CONST_BITS-1) ; psrad mm2,16 & pslld mm2,CONST_BITS+1 + + movq mm3, mm5 ; mm5=in2=z2 + punpcklwd mm5, mm0 ; mm0=in6=z3 + punpckhwd mm3, mm0 + pmaddwd mm5, [GOTOFF(ebx,PW_F184_MF076)] ; mm5=tmp2L + pmaddwd mm3, [GOTOFF(ebx,PW_F184_MF076)] ; mm3=tmp2H + + movq mm4, mm1 + movq mm0, mm2 + paddd mm1, mm5 ; mm1=tmp10L + paddd mm2, mm3 ; mm2=tmp10H + psubd mm4, mm5 ; mm4=tmp12L + psubd mm0, mm3 ; mm0=tmp12H + + ; -- Final output stage + + movq mm5, mm1 + movq mm3, mm2 + paddd mm1, mm6 ; mm1=data0L + paddd mm2, mm7 ; mm2=data0H + psubd mm5, mm6 ; mm5=data3L + psubd mm3, mm7 ; mm3=data3H + + movq mm6, [GOTOFF(ebx,PD_DESCALE_P1_4)] ; mm6=[PD_DESCALE_P1_4] + + paddd mm1, mm6 + paddd mm2, mm6 + psrad mm1, DESCALE_P1_4 + psrad mm2, DESCALE_P1_4 + paddd mm5, mm6 + paddd mm3, mm6 + psrad mm5, DESCALE_P1_4 + psrad mm3, DESCALE_P1_4 + + packssdw mm1, mm2 ; mm1=data0=(00 01 02 03) + packssdw mm5, mm3 ; mm5=data3=(30 31 32 33) + + movq mm7, MMWORD [wk(0)] ; mm7=tmp0L + movq mm6, MMWORD [wk(1)] ; mm6=tmp0H + + movq mm2, mm4 + movq mm3, mm0 + paddd mm4, mm7 ; mm4=data1L + paddd mm0, mm6 ; mm0=data1H + psubd mm2, mm7 ; mm2=data2L + psubd mm3, mm6 ; mm3=data2H + + movq mm7, [GOTOFF(ebx,PD_DESCALE_P1_4)] ; mm7=[PD_DESCALE_P1_4] + + paddd mm4, mm7 + paddd mm0, mm7 + psrad mm4, DESCALE_P1_4 + psrad mm0, DESCALE_P1_4 + paddd mm2, mm7 + paddd mm3, mm7 + psrad mm2, DESCALE_P1_4 + psrad mm3, DESCALE_P1_4 + + packssdw mm4, mm0 ; mm4=data1=(10 11 12 13) + packssdw mm2, mm3 ; mm2=data2=(20 21 22 23) + + movq mm6, mm1 ; transpose coefficients(phase 1) + punpcklwd mm1, mm4 ; mm1=(00 10 01 11) + punpckhwd mm6, mm4 ; mm6=(02 12 03 13) + movq mm7, mm2 ; transpose coefficients(phase 1) + punpcklwd mm2, mm5 ; mm2=(20 30 21 31) + punpckhwd mm7, mm5 ; mm7=(22 32 23 33) + + movq mm0, mm1 ; transpose coefficients(phase 2) + punpckldq mm1, mm2 ; mm1=(00 10 20 30) + punpckhdq mm0, mm2 ; mm0=(01 11 21 31) + movq mm3, mm6 ; transpose coefficients(phase 2) + punpckldq mm6, mm7 ; mm6=(02 12 22 32) + punpckhdq mm3, mm7 ; mm3=(03 13 23 33) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_JCOEF)], mm6 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_JCOEF)], mm3 + +.nextcolumn: + add esi, byte 4*SIZEOF_JCOEF ; coef_block + add edx, byte 4*SIZEOF_ISLOW_MULT_TYPE ; quantptr + add edi, byte 4*DCTSIZE*SIZEOF_JCOEF ; wsptr + dec ecx ; ctr + jnz near .columnloop + + ; ---- Pass 2: process rows from work array, store into output array. + + mov eax, [original_ebp] + lea esi, [workspace] ; JCOEF *wsptr + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + + ; -- Odd part + + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + movq mm2, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + + movq mm4, mm0 + movq mm5, mm0 + punpcklwd mm4, mm1 + punpckhwd mm5, mm1 + movq mm0, mm4 + movq mm1, mm5 + pmaddwd mm4, [GOTOFF(ebx,PW_F256_F089)] ; mm4=(tmp2L) + pmaddwd mm5, [GOTOFF(ebx,PW_F256_F089)] ; mm5=(tmp2H) + pmaddwd mm0, [GOTOFF(ebx,PW_F106_MF217)] ; mm0=(tmp0L) + pmaddwd mm1, [GOTOFF(ebx,PW_F106_MF217)] ; mm1=(tmp0H) + + movq mm6, mm2 + movq mm7, mm2 + punpcklwd mm6, mm3 + punpckhwd mm7, mm3 + movq mm2, mm6 + movq mm3, mm7 + pmaddwd mm6, [GOTOFF(ebx,PW_MF060_MF050)] ; mm6=(tmp2L) + pmaddwd mm7, [GOTOFF(ebx,PW_MF060_MF050)] ; mm7=(tmp2H) + pmaddwd mm2, [GOTOFF(ebx,PW_F145_MF021)] ; mm2=(tmp0L) + pmaddwd mm3, [GOTOFF(ebx,PW_F145_MF021)] ; mm3=(tmp0H) + + paddd mm6, mm4 ; mm6=tmp2L + paddd mm7, mm5 ; mm7=tmp2H + paddd mm2, mm0 ; mm2=tmp0L + paddd mm3, mm1 ; mm3=tmp0H + + movq MMWORD [wk(0)], mm2 ; wk(0)=tmp0L + movq MMWORD [wk(1)], mm3 ; wk(1)=tmp0H + + ; -- Even part + + movq mm4, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm5, MMWORD [MMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movq mm0, MMWORD [MMBLOCK(6,0,esi,SIZEOF_JCOEF)] + + pxor mm1, mm1 + pxor mm2, mm2 + punpcklwd mm1, mm4 ; mm1=tmp0L + punpckhwd mm2, mm4 ; mm2=tmp0H + psrad mm1, (16-CONST_BITS-1) ; psrad mm1,16 & pslld mm1,CONST_BITS+1 + psrad mm2, (16-CONST_BITS-1) ; psrad mm2,16 & pslld mm2,CONST_BITS+1 + + movq mm3, mm5 ; mm5=in2=z2 + punpcklwd mm5, mm0 ; mm0=in6=z3 + punpckhwd mm3, mm0 + pmaddwd mm5, [GOTOFF(ebx,PW_F184_MF076)] ; mm5=tmp2L + pmaddwd mm3, [GOTOFF(ebx,PW_F184_MF076)] ; mm3=tmp2H + + movq mm4, mm1 + movq mm0, mm2 + paddd mm1, mm5 ; mm1=tmp10L + paddd mm2, mm3 ; mm2=tmp10H + psubd mm4, mm5 ; mm4=tmp12L + psubd mm0, mm3 ; mm0=tmp12H + + ; -- Final output stage + + movq mm5, mm1 + movq mm3, mm2 + paddd mm1, mm6 ; mm1=data0L + paddd mm2, mm7 ; mm2=data0H + psubd mm5, mm6 ; mm5=data3L + psubd mm3, mm7 ; mm3=data3H + + movq mm6, [GOTOFF(ebx,PD_DESCALE_P2_4)] ; mm6=[PD_DESCALE_P2_4] + + paddd mm1, mm6 + paddd mm2, mm6 + psrad mm1, DESCALE_P2_4 + psrad mm2, DESCALE_P2_4 + paddd mm5, mm6 + paddd mm3, mm6 + psrad mm5, DESCALE_P2_4 + psrad mm3, DESCALE_P2_4 + + packssdw mm1, mm2 ; mm1=data0=(00 10 20 30) + packssdw mm5, mm3 ; mm5=data3=(03 13 23 33) + + movq mm7, MMWORD [wk(0)] ; mm7=tmp0L + movq mm6, MMWORD [wk(1)] ; mm6=tmp0H + + movq mm2, mm4 + movq mm3, mm0 + paddd mm4, mm7 ; mm4=data1L + paddd mm0, mm6 ; mm0=data1H + psubd mm2, mm7 ; mm2=data2L + psubd mm3, mm6 ; mm3=data2H + + movq mm7, [GOTOFF(ebx,PD_DESCALE_P2_4)] ; mm7=[PD_DESCALE_P2_4] + + paddd mm4, mm7 + paddd mm0, mm7 + psrad mm4, DESCALE_P2_4 + psrad mm0, DESCALE_P2_4 + paddd mm2, mm7 + paddd mm3, mm7 + psrad mm2, DESCALE_P2_4 + psrad mm3, DESCALE_P2_4 + + packssdw mm4, mm0 ; mm4=data1=(01 11 21 31) + packssdw mm2, mm3 ; mm2=data2=(02 12 22 32) + + movq mm6, [GOTOFF(ebx,PB_CENTERJSAMP)] ; mm6=[PB_CENTERJSAMP] + + packsswb mm1, mm2 ; mm1=(00 10 20 30 02 12 22 32) + packsswb mm4, mm5 ; mm4=(01 11 21 31 03 13 23 33) + paddb mm1, mm6 + paddb mm4, mm6 + + movq mm7, mm1 ; transpose coefficients(phase 1) + punpcklbw mm1, mm4 ; mm1=(00 01 10 11 20 21 30 31) + punpckhbw mm7, mm4 ; mm7=(02 03 12 13 22 23 32 33) + + movq mm0, mm1 ; transpose coefficients(phase 2) + punpcklwd mm1, mm7 ; mm1=(00 01 02 03 10 11 12 13) + punpckhwd mm0, mm7 ; mm0=(20 21 22 23 30 31 32 33) + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + movd dword [edx+eax*SIZEOF_JSAMPLE], mm1 + movd dword [esi+eax*SIZEOF_JSAMPLE], mm0 + + psrlq mm1, 4*BYTE_BIT + psrlq mm0, 4*BYTE_BIT + + mov edx, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movd dword [edx+eax*SIZEOF_JSAMPLE], mm1 + movd dword [esi+eax*SIZEOF_JSAMPLE], mm0 + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Perform dequantization and inverse DCT on one block of coefficients, +; producing a reduced-size 2x2 output block. +; +; GLOBAL(void) +; jsimd_idct_2x2_mmx(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; void *dct_table +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + + align 32 + GLOBAL_FUNCTION(jsimd_idct_2x2_mmx) + +EXTN(jsimd_idct_2x2_mmx): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input. + + mov edx, POINTER [dct_table(ebp)] ; quantptr + mov esi, JCOEFPTR [coef_block(ebp)] ; inptr + + ; | input: | result: | + ; | 00 01 ** 03 ** 05 ** 07 | | + ; | 10 11 ** 13 ** 15 ** 17 | | + ; | ** ** ** ** ** ** ** ** | | + ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 | + ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 | + ; | 50 51 ** 53 ** 55 ** 57 | | + ; | ** ** ** ** ** ** ** ** | | + ; | 70 71 ** 73 ** 75 ** 77 | | + + ; -- Odd part + + movq mm0, MMWORD [MMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw mm0, MMWORD [MMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm1, MMWORD [MMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movq mm2, MMWORD [MMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movq mm3, MMWORD [MMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw mm2, MMWORD [MMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm3, MMWORD [MMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + ; mm0=(10 11 ** 13), mm1=(30 31 ** 33) + ; mm2=(50 51 ** 53), mm3=(70 71 ** 73) + + pcmpeqd mm7, mm7 + pslld mm7, WORD_BIT ; mm7={0x0000 0xFFFF 0x0000 0xFFFF} + + movq mm4, mm0 ; mm4=(10 11 ** 13) + movq mm5, mm2 ; mm5=(50 51 ** 53) + punpcklwd mm4, mm1 ; mm4=(10 30 11 31) + punpcklwd mm5, mm3 ; mm5=(50 70 51 71) + pmaddwd mm4, [GOTOFF(ebx,PW_F362_MF127)] + pmaddwd mm5, [GOTOFF(ebx,PW_F085_MF072)] + + psrld mm0, WORD_BIT ; mm0=(11 -- 13 --) + pand mm1, mm7 ; mm1=(-- 31 -- 33) + psrld mm2, WORD_BIT ; mm2=(51 -- 53 --) + pand mm3, mm7 ; mm3=(-- 71 -- 73) + por mm0, mm1 ; mm0=(11 31 13 33) + por mm2, mm3 ; mm2=(51 71 53 73) + pmaddwd mm0, [GOTOFF(ebx,PW_F362_MF127)] + pmaddwd mm2, [GOTOFF(ebx,PW_F085_MF072)] + + paddd mm4, mm5 ; mm4=tmp0[col0 col1] + + movq mm6, MMWORD [MMBLOCK(1,1,esi,SIZEOF_JCOEF)] + movq mm1, MMWORD [MMBLOCK(3,1,esi,SIZEOF_JCOEF)] + pmullw mm6, MMWORD [MMBLOCK(1,1,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm1, MMWORD [MMBLOCK(3,1,edx,SIZEOF_ISLOW_MULT_TYPE)] + movq mm3, MMWORD [MMBLOCK(5,1,esi,SIZEOF_JCOEF)] + movq mm5, MMWORD [MMBLOCK(7,1,esi,SIZEOF_JCOEF)] + pmullw mm3, MMWORD [MMBLOCK(5,1,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm5, MMWORD [MMBLOCK(7,1,edx,SIZEOF_ISLOW_MULT_TYPE)] + + ; mm6=(** 15 ** 17), mm1=(** 35 ** 37) + ; mm3=(** 55 ** 57), mm5=(** 75 ** 77) + + psrld mm6, WORD_BIT ; mm6=(15 -- 17 --) + pand mm1, mm7 ; mm1=(-- 35 -- 37) + psrld mm3, WORD_BIT ; mm3=(55 -- 57 --) + pand mm5, mm7 ; mm5=(-- 75 -- 77) + por mm6, mm1 ; mm6=(15 35 17 37) + por mm3, mm5 ; mm3=(55 75 57 77) + pmaddwd mm6, [GOTOFF(ebx,PW_F362_MF127)] + pmaddwd mm3, [GOTOFF(ebx,PW_F085_MF072)] + + paddd mm0, mm2 ; mm0=tmp0[col1 col3] + paddd mm6, mm3 ; mm6=tmp0[col5 col7] + + ; -- Even part + + movq mm1, MMWORD [MMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movq mm5, MMWORD [MMBLOCK(0,1,esi,SIZEOF_JCOEF)] + pmullw mm1, MMWORD [MMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw mm5, MMWORD [MMBLOCK(0,1,edx,SIZEOF_ISLOW_MULT_TYPE)] + + ; mm1=(00 01 ** 03), mm5=(** 05 ** 07) + + movq mm2, mm1 ; mm2=(00 01 ** 03) + pslld mm1, WORD_BIT ; mm1=(-- 00 -- **) + psrad mm1, (WORD_BIT-CONST_BITS-2) ; mm1=tmp10[col0 ****] + + pand mm2, mm7 ; mm2=(-- 01 -- 03) + pand mm5, mm7 ; mm5=(-- 05 -- 07) + psrad mm2, (WORD_BIT-CONST_BITS-2) ; mm2=tmp10[col1 col3] + psrad mm5, (WORD_BIT-CONST_BITS-2) ; mm5=tmp10[col5 col7] + + ; -- Final output stage + + movq mm3, mm1 + paddd mm1, mm4 ; mm1=data0[col0 ****]=(A0 **) + psubd mm3, mm4 ; mm3=data1[col0 ****]=(B0 **) + punpckldq mm1, mm3 ; mm1=(A0 B0) + + movq mm7, [GOTOFF(ebx,PD_DESCALE_P1_2)] ; mm7=[PD_DESCALE_P1_2] + + movq mm4, mm2 + movq mm3, mm5 + paddd mm2, mm0 ; mm2=data0[col1 col3]=(A1 A3) + paddd mm5, mm6 ; mm5=data0[col5 col7]=(A5 A7) + psubd mm4, mm0 ; mm4=data1[col1 col3]=(B1 B3) + psubd mm3, mm6 ; mm3=data1[col5 col7]=(B5 B7) + + paddd mm1, mm7 + psrad mm1, DESCALE_P1_2 + + paddd mm2, mm7 + paddd mm5, mm7 + psrad mm2, DESCALE_P1_2 + psrad mm5, DESCALE_P1_2 + paddd mm4, mm7 + paddd mm3, mm7 + psrad mm4, DESCALE_P1_2 + psrad mm3, DESCALE_P1_2 + + ; ---- Pass 2: process rows, store into output array. + + mov edi, JSAMPARRAY [output_buf(ebp)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(ebp)] + + ; | input:| result:| + ; | A0 B0 | | + ; | A1 B1 | C0 C1 | + ; | A3 B3 | D0 D1 | + ; | A5 B5 | | + ; | A7 B7 | | + + ; -- Odd part + + packssdw mm2, mm4 ; mm2=(A1 A3 B1 B3) + packssdw mm5, mm3 ; mm5=(A5 A7 B5 B7) + pmaddwd mm2, [GOTOFF(ebx,PW_F362_MF127)] + pmaddwd mm5, [GOTOFF(ebx,PW_F085_MF072)] + + paddd mm2, mm5 ; mm2=tmp0[row0 row1] + + ; -- Even part + + pslld mm1, (CONST_BITS+2) ; mm1=tmp10[row0 row1] + + ; -- Final output stage + + movq mm0, [GOTOFF(ebx,PD_DESCALE_P2_2)] ; mm0=[PD_DESCALE_P2_2] + + movq mm6, mm1 + paddd mm1, mm2 ; mm1=data0[row0 row1]=(C0 C1) + psubd mm6, mm2 ; mm6=data1[row0 row1]=(D0 D1) + + paddd mm1, mm0 + paddd mm6, mm0 + psrad mm1, DESCALE_P2_2 + psrad mm6, DESCALE_P2_2 + + movq mm7, mm1 ; transpose coefficients + punpckldq mm1, mm6 ; mm1=(C0 D0) + punpckhdq mm7, mm6 ; mm7=(C1 D1) + + packssdw mm1, mm7 ; mm1=(C0 D0 C1 D1) + packsswb mm1, mm1 ; mm1=(C0 D0 C1 D1 C0 D0 C1 D1) + paddb mm1, [GOTOFF(ebx,PB_CENTERJSAMP)] + + movd ecx, mm1 + movd ebx, mm1 ; ebx=(C0 D0 C1 D1) + shr ecx, 2*BYTE_BIT ; ecx=(C1 D1 -- --) + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + mov word [edx+eax*SIZEOF_JSAMPLE], bx + mov word [esi+eax*SIZEOF_JSAMPLE], cx + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-sse2.asm new file mode 100644 index 0000000..1fe967d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jidctred-sse2.asm @@ -0,0 +1,588 @@ +; +; jidctred.asm - reduced-size IDCT (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains inverse-DCT routines that produce reduced-size +; output: either 4x4 or 2x2 pixels from an 8x8 DCT block. +; The following code is based directly on the IJG's original jidctred.c; +; see the jidctred.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1_4 (CONST_BITS - PASS1_BITS + 1) +%define DESCALE_P2_4 (CONST_BITS + PASS1_BITS + 3 + 1) +%define DESCALE_P1_2 (CONST_BITS - PASS1_BITS + 2) +%define DESCALE_P2_2 (CONST_BITS + PASS1_BITS + 3 + 2) + +%if CONST_BITS == 13 +F_0_211 equ 1730 ; FIX(0.211164243) +F_0_509 equ 4176 ; FIX(0.509795579) +F_0_601 equ 4926 ; FIX(0.601344887) +F_0_720 equ 5906 ; FIX(0.720959822) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_850 equ 6967 ; FIX(0.850430095) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_061 equ 8697 ; FIX(1.061594337) +F_1_272 equ 10426 ; FIX(1.272758580) +F_1_451 equ 11893 ; FIX(1.451774981) +F_1_847 equ 15137 ; FIX(1.847759065) +F_2_172 equ 17799 ; FIX(2.172734803) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_624 equ 29692 ; FIX(3.624509785) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_211 equ DESCALE( 226735879, 30 - CONST_BITS) ; FIX(0.211164243) +F_0_509 equ DESCALE( 547388834, 30 - CONST_BITS) ; FIX(0.509795579) +F_0_601 equ DESCALE( 645689155, 30 - CONST_BITS) ; FIX(0.601344887) +F_0_720 equ DESCALE( 774124714, 30 - CONST_BITS) ; FIX(0.720959822) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_850 equ DESCALE( 913142361, 30 - CONST_BITS) ; FIX(0.850430095) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_061 equ DESCALE(1139878239, 30 - CONST_BITS) ; FIX(1.061594337) +F_1_272 equ DESCALE(1366614119, 30 - CONST_BITS) ; FIX(1.272758580) +F_1_451 equ DESCALE(1558831516, 30 - CONST_BITS) ; FIX(1.451774981) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_2_172 equ DESCALE(2332956230, 30 - CONST_BITS) ; FIX(2.172734803) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_624 equ DESCALE(3891787747, 30 - CONST_BITS) ; FIX(3.624509785) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_red_sse2) + +EXTN(jconst_idct_red_sse2): + +PW_F184_MF076 times 4 dw F_1_847, -F_0_765 +PW_F256_F089 times 4 dw F_2_562, F_0_899 +PW_F106_MF217 times 4 dw F_1_061, -F_2_172 +PW_MF060_MF050 times 4 dw -F_0_601, -F_0_509 +PW_F145_MF021 times 4 dw F_1_451, -F_0_211 +PW_F362_MF127 times 4 dw F_3_624, -F_1_272 +PW_F085_MF072 times 4 dw F_0_850, -F_0_720 +PD_DESCALE_P1_4 times 4 dd 1 << (DESCALE_P1_4 - 1) +PD_DESCALE_P2_4 times 4 dd 1 << (DESCALE_P2_4 - 1) +PD_DESCALE_P1_2 times 4 dd 1 << (DESCALE_P1_2 - 1) +PD_DESCALE_P2_2 times 4 dd 1 << (DESCALE_P2_2 - 1) +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Perform dequantization and inverse DCT on one block of coefficients, +; producing a reduced-size 4x4 output block. +; +; GLOBAL(void) +; jsimd_idct_4x4_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; void *dct_table +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + +%define original_ebp ebp + 0 +%define wk(i) ebp - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_idct_4x4_sse2) + +EXTN(jsimd_idct_4x4_sse2): + push ebp + mov eax, esp ; eax = original ebp + sub esp, byte 4 + and esp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + mov [esp], eax + mov ebp, esp ; ebp = aligned ebp + lea esp, [wk(0)] + PUSHPIC ebx +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input. + +; mov eax, [original_ebp] + mov edx, POINTER [dct_table(eax)] ; quantptr + mov esi, JCOEFPTR [coef_block(eax)] ; inptr + +%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2 + mov eax, dword [DWBLOCK(1,0,esi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,esi,SIZEOF_JCOEF)] + jnz short .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + por xmm0, xmm1 + packsswb xmm0, xmm0 + packsswb xmm0, xmm0 + movd eax, xmm0 + test eax, eax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + psllw xmm0, PASS1_BITS + + movdqa xmm3, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07) + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + punpckhwd xmm3, xmm3 ; xmm3=(04 04 05 05 06 06 07 07) + + pshufd xmm1, xmm0, 0x50 ; xmm1=[col0 col1]=(00 00 00 00 01 01 01 01) + pshufd xmm0, xmm0, 0xFA ; xmm0=[col2 col3]=(02 02 02 02 03 03 03 03) + pshufd xmm6, xmm3, 0x50 ; xmm6=[col4 col5]=(04 04 04 04 05 05 05 05) + pshufd xmm3, xmm3, 0xFA ; xmm3=[col6 col7]=(06 06 06 06 07 07 07 07) + + jmp near .column_end + ALIGNX 16, 7 +%endif +.columnDCT: + + ; -- Odd part + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + movdqa xmm4, xmm0 + movdqa xmm5, xmm0 + punpcklwd xmm4, xmm1 + punpckhwd xmm5, xmm1 + movdqa xmm0, xmm4 + movdqa xmm1, xmm5 + pmaddwd xmm4, [GOTOFF(ebx,PW_F256_F089)] ; xmm4=(tmp2L) + pmaddwd xmm5, [GOTOFF(ebx,PW_F256_F089)] ; xmm5=(tmp2H) + pmaddwd xmm0, [GOTOFF(ebx,PW_F106_MF217)] ; xmm0=(tmp0L) + pmaddwd xmm1, [GOTOFF(ebx,PW_F106_MF217)] ; xmm1=(tmp0H) + + movdqa xmm6, xmm2 + movdqa xmm7, xmm2 + punpcklwd xmm6, xmm3 + punpckhwd xmm7, xmm3 + movdqa xmm2, xmm6 + movdqa xmm3, xmm7 + pmaddwd xmm6, [GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2L) + pmaddwd xmm7, [GOTOFF(ebx,PW_MF060_MF050)] ; xmm7=(tmp2H) + pmaddwd xmm2, [GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0L) + pmaddwd xmm3, [GOTOFF(ebx,PW_F145_MF021)] ; xmm3=(tmp0H) + + paddd xmm6, xmm4 ; xmm6=tmp2L + paddd xmm7, xmm5 ; xmm7=tmp2H + paddd xmm2, xmm0 ; xmm2=tmp0L + paddd xmm3, xmm1 ; xmm3=tmp0H + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp0L + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=tmp0H + + ; -- Even part + + movdqa xmm4, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + movdqa xmm5, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_JCOEF)] + movdqa xmm0, XMMWORD [XMMBLOCK(6,0,esi,SIZEOF_JCOEF)] + pmullw xmm4, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm5, XMMWORD [XMMBLOCK(2,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm0, XMMWORD [XMMBLOCK(6,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + pxor xmm1, xmm1 + pxor xmm2, xmm2 + punpcklwd xmm1, xmm4 ; xmm1=tmp0L + punpckhwd xmm2, xmm4 ; xmm2=tmp0H + psrad xmm1, (16-CONST_BITS-1) ; psrad xmm1,16 & pslld xmm1,CONST_BITS+1 + psrad xmm2, (16-CONST_BITS-1) ; psrad xmm2,16 & pslld xmm2,CONST_BITS+1 + + movdqa xmm3, xmm5 ; xmm5=in2=z2 + punpcklwd xmm5, xmm0 ; xmm0=in6=z3 + punpckhwd xmm3, xmm0 + pmaddwd xmm5, [GOTOFF(ebx,PW_F184_MF076)] ; xmm5=tmp2L + pmaddwd xmm3, [GOTOFF(ebx,PW_F184_MF076)] ; xmm3=tmp2H + + movdqa xmm4, xmm1 + movdqa xmm0, xmm2 + paddd xmm1, xmm5 ; xmm1=tmp10L + paddd xmm2, xmm3 ; xmm2=tmp10H + psubd xmm4, xmm5 ; xmm4=tmp12L + psubd xmm0, xmm3 ; xmm0=tmp12H + + ; -- Final output stage + + movdqa xmm5, xmm1 + movdqa xmm3, xmm2 + paddd xmm1, xmm6 ; xmm1=data0L + paddd xmm2, xmm7 ; xmm2=data0H + psubd xmm5, xmm6 ; xmm5=data3L + psubd xmm3, xmm7 ; xmm3=data3H + + movdqa xmm6, [GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm6=[PD_DESCALE_P1_4] + + paddd xmm1, xmm6 + paddd xmm2, xmm6 + psrad xmm1, DESCALE_P1_4 + psrad xmm2, DESCALE_P1_4 + paddd xmm5, xmm6 + paddd xmm3, xmm6 + psrad xmm5, DESCALE_P1_4 + psrad xmm3, DESCALE_P1_4 + + packssdw xmm1, xmm2 ; xmm1=data0=(00 01 02 03 04 05 06 07) + packssdw xmm5, xmm3 ; xmm5=data3=(30 31 32 33 34 35 36 37) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp0L + movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp0H + + movdqa xmm2, xmm4 + movdqa xmm3, xmm0 + paddd xmm4, xmm7 ; xmm4=data1L + paddd xmm0, xmm6 ; xmm0=data1H + psubd xmm2, xmm7 ; xmm2=data2L + psubd xmm3, xmm6 ; xmm3=data2H + + movdqa xmm7, [GOTOFF(ebx,PD_DESCALE_P1_4)] ; xmm7=[PD_DESCALE_P1_4] + + paddd xmm4, xmm7 + paddd xmm0, xmm7 + psrad xmm4, DESCALE_P1_4 + psrad xmm0, DESCALE_P1_4 + paddd xmm2, xmm7 + paddd xmm3, xmm7 + psrad xmm2, DESCALE_P1_4 + psrad xmm3, DESCALE_P1_4 + + packssdw xmm4, xmm0 ; xmm4=data1=(10 11 12 13 14 15 16 17) + packssdw xmm2, xmm3 ; xmm2=data2=(20 21 22 23 24 25 26 27) + + movdqa xmm6, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm4 ; xmm1=(00 10 01 11 02 12 03 13) + punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17) + movdqa xmm7, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm5 ; xmm2=(20 30 21 31 22 32 23 33) + punpckhwd xmm7, xmm5 ; xmm7=(24 34 25 35 26 36 27 37) + + movdqa xmm0, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm2 ; xmm1=[col0 col1]=(00 10 20 30 01 11 21 31) + punpckhdq xmm0, xmm2 ; xmm0=[col2 col3]=(02 12 22 32 03 13 23 33) + movdqa xmm3, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm7 ; xmm6=[col4 col5]=(04 14 24 34 05 15 25 35) + punpckhdq xmm3, xmm7 ; xmm3=[col6 col7]=(06 16 26 36 07 17 27 37) +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows, store into output array. + + mov eax, [original_ebp] + mov edi, JSAMPARRAY [output_buf(eax)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(eax)] + + ; -- Even part + + pxor xmm4, xmm4 + punpcklwd xmm4, xmm1 ; xmm4=tmp0 + psrad xmm4, (16-CONST_BITS-1) ; psrad xmm4,16 & pslld xmm4,CONST_BITS+1 + + ; -- Odd part + + punpckhwd xmm1, xmm0 + punpckhwd xmm6, xmm3 + movdqa xmm5, xmm1 + movdqa xmm2, xmm6 + pmaddwd xmm1, [GOTOFF(ebx,PW_F256_F089)] ; xmm1=(tmp2) + pmaddwd xmm6, [GOTOFF(ebx,PW_MF060_MF050)] ; xmm6=(tmp2) + pmaddwd xmm5, [GOTOFF(ebx,PW_F106_MF217)] ; xmm5=(tmp0) + pmaddwd xmm2, [GOTOFF(ebx,PW_F145_MF021)] ; xmm2=(tmp0) + + paddd xmm6, xmm1 ; xmm6=tmp2 + paddd xmm2, xmm5 ; xmm2=tmp0 + + ; -- Even part + + punpcklwd xmm0, xmm3 + pmaddwd xmm0, [GOTOFF(ebx,PW_F184_MF076)] ; xmm0=tmp2 + + movdqa xmm7, xmm4 + paddd xmm4, xmm0 ; xmm4=tmp10 + psubd xmm7, xmm0 ; xmm7=tmp12 + + ; -- Final output stage + + movdqa xmm1, [GOTOFF(ebx,PD_DESCALE_P2_4)] ; xmm1=[PD_DESCALE_P2_4] + + movdqa xmm5, xmm4 + movdqa xmm3, xmm7 + paddd xmm4, xmm6 ; xmm4=data0=(00 10 20 30) + paddd xmm7, xmm2 ; xmm7=data1=(01 11 21 31) + psubd xmm5, xmm6 ; xmm5=data3=(03 13 23 33) + psubd xmm3, xmm2 ; xmm3=data2=(02 12 22 32) + + paddd xmm4, xmm1 + paddd xmm7, xmm1 + psrad xmm4, DESCALE_P2_4 + psrad xmm7, DESCALE_P2_4 + paddd xmm5, xmm1 + paddd xmm3, xmm1 + psrad xmm5, DESCALE_P2_4 + psrad xmm3, DESCALE_P2_4 + + packssdw xmm4, xmm3 ; xmm4=(00 10 20 30 02 12 22 32) + packssdw xmm7, xmm5 ; xmm7=(01 11 21 31 03 13 23 33) + + movdqa xmm0, xmm4 ; transpose coefficients(phase 1) + punpcklwd xmm4, xmm7 ; xmm4=(00 01 10 11 20 21 30 31) + punpckhwd xmm0, xmm7 ; xmm0=(02 03 12 13 22 23 32 33) + + movdqa xmm6, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm0 ; xmm4=(00 01 02 03 10 11 12 13) + punpckhdq xmm6, xmm0 ; xmm6=(20 21 22 23 30 31 32 33) + + packsswb xmm4, xmm6 ; xmm4=(00 01 02 03 10 11 12 13 20 ..) + paddb xmm4, [GOTOFF(ebx,PB_CENTERJSAMP)] + + pshufd xmm2, xmm4, 0x39 ; xmm2=(10 11 12 13 20 21 22 23 30 ..) + pshufd xmm1, xmm4, 0x4E ; xmm1=(20 21 22 23 30 31 32 33 00 ..) + pshufd xmm3, xmm4, 0x93 ; xmm3=(30 31 32 33 00 01 02 03 10 ..) + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + movd XMM_DWORD [edx+eax*SIZEOF_JSAMPLE], xmm4 + movd XMM_DWORD [esi+eax*SIZEOF_JSAMPLE], xmm2 + mov edx, JSAMPROW [edi+2*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+3*SIZEOF_JSAMPROW] + movd XMM_DWORD [edx+eax*SIZEOF_JSAMPLE], xmm1 + movd XMM_DWORD [esi+eax*SIZEOF_JSAMPLE], xmm3 + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused + POPPIC ebx + mov esp, ebp ; esp <- aligned ebp + pop esp ; esp <- original ebp + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Perform dequantization and inverse DCT on one block of coefficients, +; producing a reduced-size 2x2 output block. +; +; GLOBAL(void) +; jsimd_idct_2x2_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +%define dct_table(b) (b) + 8 ; void *dct_table +%define coef_block(b) (b) + 12 ; JCOEFPTR coef_block +%define output_buf(b) (b) + 16 ; JSAMPARRAY output_buf +%define output_col(b) (b) + 20 ; JDIMENSION output_col + + align 32 + GLOBAL_FUNCTION(jsimd_idct_2x2_sse2) + +EXTN(jsimd_idct_2x2_sse2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + GET_GOT ebx ; get GOT address + + ; ---- Pass 1: process columns from input. + + mov edx, POINTER [dct_table(ebp)] ; quantptr + mov esi, JCOEFPTR [coef_block(ebp)] ; inptr + + ; | input: | result: | + ; | 00 01 ** 03 ** 05 ** 07 | | + ; | 10 11 ** 13 ** 15 ** 17 | | + ; | ** ** ** ** ** ** ** ** | | + ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 | + ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 | + ; | 50 51 ** 53 ** 55 ** 57 | | + ; | ** ** ** ** ** ** ** ** | | + ; | 70 71 ** 73 ** 75 ** 77 | | + + ; -- Odd part + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(3,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(5,0,esi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,esi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(5,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(7,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + ; xmm0=(10 11 ** 13 ** 15 ** 17), xmm1=(30 31 ** 33 ** 35 ** 37) + ; xmm2=(50 51 ** 53 ** 55 ** 57), xmm3=(70 71 ** 73 ** 75 ** 77) + + pcmpeqd xmm7, xmm7 + pslld xmm7, WORD_BIT ; xmm7={0x0000 0xFFFF 0x0000 0xFFFF ..} + + movdqa xmm4, xmm0 ; xmm4=(10 11 ** 13 ** 15 ** 17) + movdqa xmm5, xmm2 ; xmm5=(50 51 ** 53 ** 55 ** 57) + punpcklwd xmm4, xmm1 ; xmm4=(10 30 11 31 ** ** 13 33) + punpcklwd xmm5, xmm3 ; xmm5=(50 70 51 71 ** ** 53 73) + pmaddwd xmm4, [GOTOFF(ebx,PW_F362_MF127)] + pmaddwd xmm5, [GOTOFF(ebx,PW_F085_MF072)] + + psrld xmm0, WORD_BIT ; xmm0=(11 -- 13 -- 15 -- 17 --) + pand xmm1, xmm7 ; xmm1=(-- 31 -- 33 -- 35 -- 37) + psrld xmm2, WORD_BIT ; xmm2=(51 -- 53 -- 55 -- 57 --) + pand xmm3, xmm7 ; xmm3=(-- 71 -- 73 -- 75 -- 77) + por xmm0, xmm1 ; xmm0=(11 31 13 33 15 35 17 37) + por xmm2, xmm3 ; xmm2=(51 71 53 73 55 75 57 77) + pmaddwd xmm0, [GOTOFF(ebx,PW_F362_MF127)] + pmaddwd xmm2, [GOTOFF(ebx,PW_F085_MF072)] + + paddd xmm4, xmm5 ; xmm4=tmp0[col0 col1 **** col3] + paddd xmm0, xmm2 ; xmm0=tmp0[col1 col3 col5 col7] + + ; -- Even part + + movdqa xmm6, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_JCOEF)] + pmullw xmm6, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_ISLOW_MULT_TYPE)] + + ; xmm6=(00 01 ** 03 ** 05 ** 07) + + movdqa xmm1, xmm6 ; xmm1=(00 01 ** 03 ** 05 ** 07) + pslld xmm6, WORD_BIT ; xmm6=(-- 00 -- ** -- ** -- **) + pand xmm1, xmm7 ; xmm1=(-- 01 -- 03 -- 05 -- 07) + psrad xmm6, (WORD_BIT-CONST_BITS-2) ; xmm6=tmp10[col0 **** **** ****] + psrad xmm1, (WORD_BIT-CONST_BITS-2) ; xmm1=tmp10[col1 col3 col5 col7] + + ; -- Final output stage + + movdqa xmm3, xmm6 + movdqa xmm5, xmm1 + paddd xmm6, xmm4 ; xmm6=data0[col0 **** **** ****]=(A0 ** ** **) + paddd xmm1, xmm0 ; xmm1=data0[col1 col3 col5 col7]=(A1 A3 A5 A7) + psubd xmm3, xmm4 ; xmm3=data1[col0 **** **** ****]=(B0 ** ** **) + psubd xmm5, xmm0 ; xmm5=data1[col1 col3 col5 col7]=(B1 B3 B5 B7) + + movdqa xmm2, [GOTOFF(ebx,PD_DESCALE_P1_2)] ; xmm2=[PD_DESCALE_P1_2] + + punpckldq xmm6, xmm3 ; xmm6=(A0 B0 ** **) + + movdqa xmm7, xmm1 + punpcklqdq xmm1, xmm5 ; xmm1=(A1 A3 B1 B3) + punpckhqdq xmm7, xmm5 ; xmm7=(A5 A7 B5 B7) + + paddd xmm6, xmm2 + psrad xmm6, DESCALE_P1_2 + + paddd xmm1, xmm2 + paddd xmm7, xmm2 + psrad xmm1, DESCALE_P1_2 + psrad xmm7, DESCALE_P1_2 + + ; -- Prefetch the next coefficient block + + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [esi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows, store into output array. + + mov edi, JSAMPARRAY [output_buf(ebp)] ; (JSAMPROW *) + mov eax, JDIMENSION [output_col(ebp)] + + ; | input:| result:| + ; | A0 B0 | | + ; | A1 B1 | C0 C1 | + ; | A3 B3 | D0 D1 | + ; | A5 B5 | | + ; | A7 B7 | | + + ; -- Odd part + + packssdw xmm1, xmm1 ; xmm1=(A1 A3 B1 B3 A1 A3 B1 B3) + packssdw xmm7, xmm7 ; xmm7=(A5 A7 B5 B7 A5 A7 B5 B7) + pmaddwd xmm1, [GOTOFF(ebx,PW_F362_MF127)] + pmaddwd xmm7, [GOTOFF(ebx,PW_F085_MF072)] + + paddd xmm1, xmm7 ; xmm1=tmp0[row0 row1 row0 row1] + + ; -- Even part + + pslld xmm6, (CONST_BITS+2) ; xmm6=tmp10[row0 row1 **** ****] + + ; -- Final output stage + + movdqa xmm4, xmm6 + paddd xmm6, xmm1 ; xmm6=data0[row0 row1 **** ****]=(C0 C1 ** **) + psubd xmm4, xmm1 ; xmm4=data1[row0 row1 **** ****]=(D0 D1 ** **) + + punpckldq xmm6, xmm4 ; xmm6=(C0 D0 C1 D1) + + paddd xmm6, [GOTOFF(ebx,PD_DESCALE_P2_2)] + psrad xmm6, DESCALE_P2_2 + + packssdw xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1) + packsswb xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1 ..) + paddb xmm6, [GOTOFF(ebx,PB_CENTERJSAMP)] + + pextrw ebx, xmm6, 0x00 ; ebx=(C0 D0 -- --) + pextrw ecx, xmm6, 0x01 ; ecx=(C1 D1 -- --) + + mov edx, JSAMPROW [edi+0*SIZEOF_JSAMPROW] + mov esi, JSAMPROW [edi+1*SIZEOF_JSAMPROW] + mov word [edx+eax*SIZEOF_JSAMPLE], bx + mov word [esi+eax*SIZEOF_JSAMPLE], cx + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-3dn.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-3dn.asm new file mode 100644 index 0000000..58e0011 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-3dn.asm @@ -0,0 +1,226 @@ +; +; jquant.asm - sample data conversion and quantization (3DNow! & MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_float_3dnow(JSAMPARRAY sample_data, JDIMENSION start_col, +; FAST_FLOAT *workspace); +; + +%define sample_data ebp + 8 ; JSAMPARRAY sample_data +%define start_col ebp + 12 ; JDIMENSION start_col +%define workspace ebp + 16 ; FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_float_3dnow) + +EXTN(jsimd_convsamp_float_3dnow): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + pcmpeqw mm7, mm7 + psllw mm7, 7 + packsswb mm7, mm7 ; mm7 = PB_CENTERJSAMPLE (0x808080..) + + mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *) + mov eax, JDIMENSION [start_col] + mov edi, POINTER [workspace] ; (DCTELEM *) + mov ecx, DCTSIZE/2 + ALIGNX 16, 7 +.convloop: + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq mm0, MMWORD [ebx+eax*SIZEOF_JSAMPLE] + movq mm1, MMWORD [edx+eax*SIZEOF_JSAMPLE] + + psubb mm0, mm7 ; mm0=(01234567) + psubb mm1, mm7 ; mm1=(89ABCDEF) + + punpcklbw mm2, mm0 ; mm2=(*0*1*2*3) + punpckhbw mm0, mm0 ; mm0=(*4*5*6*7) + punpcklbw mm3, mm1 ; mm3=(*8*9*A*B) + punpckhbw mm1, mm1 ; mm1=(*C*D*E*F) + + punpcklwd mm4, mm2 ; mm4=(***0***1) + punpckhwd mm2, mm2 ; mm2=(***2***3) + punpcklwd mm5, mm0 ; mm5=(***4***5) + punpckhwd mm0, mm0 ; mm0=(***6***7) + + psrad mm4, (DWORD_BIT-BYTE_BIT) ; mm4=(01) + psrad mm2, (DWORD_BIT-BYTE_BIT) ; mm2=(23) + pi2fd mm4, mm4 + pi2fd mm2, mm2 + psrad mm5, (DWORD_BIT-BYTE_BIT) ; mm5=(45) + psrad mm0, (DWORD_BIT-BYTE_BIT) ; mm0=(67) + pi2fd mm5, mm5 + pi2fd mm0, mm0 + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], mm4 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], mm2 + movq MMWORD [MMBLOCK(0,2,edi,SIZEOF_FAST_FLOAT)], mm5 + movq MMWORD [MMBLOCK(0,3,edi,SIZEOF_FAST_FLOAT)], mm0 + + punpcklwd mm6, mm3 ; mm6=(***8***9) + punpckhwd mm3, mm3 ; mm3=(***A***B) + punpcklwd mm4, mm1 ; mm4=(***C***D) + punpckhwd mm1, mm1 ; mm1=(***E***F) + + psrad mm6, (DWORD_BIT-BYTE_BIT) ; mm6=(89) + psrad mm3, (DWORD_BIT-BYTE_BIT) ; mm3=(AB) + pi2fd mm6, mm6 + pi2fd mm3, mm3 + psrad mm4, (DWORD_BIT-BYTE_BIT) ; mm4=(CD) + psrad mm1, (DWORD_BIT-BYTE_BIT) ; mm1=(EF) + pi2fd mm4, mm4 + pi2fd mm1, mm1 + + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], mm6 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], mm3 + movq MMWORD [MMBLOCK(1,2,edi,SIZEOF_FAST_FLOAT)], mm4 + movq MMWORD [MMBLOCK(1,3,edi,SIZEOF_FAST_FLOAT)], mm1 + + add esi, byte 2*SIZEOF_JSAMPROW + add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT + dec ecx + jnz near .convloop + + femms ; empty MMX/3DNow! state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; GLOBAL(void) +; jsimd_quantize_float_3dnow(JCOEFPTR coef_block, FAST_FLOAT *divisors, +; FAST_FLOAT *workspace); +; + +%define coef_block ebp + 8 ; JCOEFPTR coef_block +%define divisors ebp + 12 ; FAST_FLOAT *divisors +%define workspace ebp + 16 ; FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_float_3dnow) + +EXTN(jsimd_quantize_float_3dnow): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + mov eax, 0x4B400000 ; (float)0x00C00000 (rndint_magic) + movd mm7, eax + punpckldq mm7, mm7 ; mm7={12582912.0F 12582912.0F} + + mov esi, POINTER [workspace] + mov edx, POINTER [divisors] + mov edi, JCOEFPTR [coef_block] + mov eax, DCTSIZE2/16 + ALIGNX 16, 7 +.quantloop: + movq mm0, MMWORD [MMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] + movq mm1, MMWORD [MMBLOCK(0,1,esi,SIZEOF_FAST_FLOAT)] + pfmul mm0, MMWORD [MMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] + pfmul mm1, MMWORD [MMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)] + movq mm2, MMWORD [MMBLOCK(0,2,esi,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(0,3,esi,SIZEOF_FAST_FLOAT)] + pfmul mm2, MMWORD [MMBLOCK(0,2,edx,SIZEOF_FAST_FLOAT)] + pfmul mm3, MMWORD [MMBLOCK(0,3,edx,SIZEOF_FAST_FLOAT)] + + pfadd mm0, mm7 ; mm0=(00 ** 01 **) + pfadd mm1, mm7 ; mm1=(02 ** 03 **) + pfadd mm2, mm7 ; mm0=(04 ** 05 **) + pfadd mm3, mm7 ; mm1=(06 ** 07 **) + + movq mm4, mm0 + punpcklwd mm0, mm1 ; mm0=(00 02 ** **) + punpckhwd mm4, mm1 ; mm4=(01 03 ** **) + movq mm5, mm2 + punpcklwd mm2, mm3 ; mm2=(04 06 ** **) + punpckhwd mm5, mm3 ; mm5=(05 07 ** **) + + punpcklwd mm0, mm4 ; mm0=(00 01 02 03) + punpcklwd mm2, mm5 ; mm2=(04 05 06 07) + + movq mm6, MMWORD [MMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] + movq mm1, MMWORD [MMBLOCK(1,1,esi,SIZEOF_FAST_FLOAT)] + pfmul mm6, MMWORD [MMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] + pfmul mm1, MMWORD [MMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)] + movq mm3, MMWORD [MMBLOCK(1,2,esi,SIZEOF_FAST_FLOAT)] + movq mm4, MMWORD [MMBLOCK(1,3,esi,SIZEOF_FAST_FLOAT)] + pfmul mm3, MMWORD [MMBLOCK(1,2,edx,SIZEOF_FAST_FLOAT)] + pfmul mm4, MMWORD [MMBLOCK(1,3,edx,SIZEOF_FAST_FLOAT)] + + pfadd mm6, mm7 ; mm0=(10 ** 11 **) + pfadd mm1, mm7 ; mm4=(12 ** 13 **) + pfadd mm3, mm7 ; mm0=(14 ** 15 **) + pfadd mm4, mm7 ; mm4=(16 ** 17 **) + + movq mm5, mm6 + punpcklwd mm6, mm1 ; mm6=(10 12 ** **) + punpckhwd mm5, mm1 ; mm5=(11 13 ** **) + movq mm1, mm3 + punpcklwd mm3, mm4 ; mm3=(14 16 ** **) + punpckhwd mm1, mm4 ; mm1=(15 17 ** **) + + punpcklwd mm6, mm5 ; mm6=(10 11 12 13) + punpcklwd mm3, mm1 ; mm3=(14 15 16 17) + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm6 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm3 + + add esi, byte 16*SIZEOF_FAST_FLOAT + add edx, byte 16*SIZEOF_FAST_FLOAT + add edi, byte 16*SIZEOF_JCOEF + dec eax + jnz near .quantloop + + femms ; empty MMX/3DNow! state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-mmx.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-mmx.asm new file mode 100644 index 0000000..4eda95c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-mmx.asm @@ -0,0 +1,272 @@ +; +; jquant.asm - sample data conversion and quantization (MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_mmx(JSAMPARRAY sample_data, JDIMENSION start_col, +; DCTELEM *workspace); +; + +%define sample_data ebp + 8 ; JSAMPARRAY sample_data +%define start_col ebp + 12 ; JDIMENSION start_col +%define workspace ebp + 16 ; DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_mmx) + +EXTN(jsimd_convsamp_mmx): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + pxor mm6, mm6 ; mm6=(all 0's) + pcmpeqw mm7, mm7 + psllw mm7, 7 ; mm7={0xFF80 0xFF80 0xFF80 0xFF80} + + mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *) + mov eax, JDIMENSION [start_col] + mov edi, POINTER [workspace] ; (DCTELEM *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.convloop: + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq mm0, MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; mm0=(01234567) + movq mm1, MMWORD [edx+eax*SIZEOF_JSAMPLE] ; mm1=(89ABCDEF) + + mov ebx, JSAMPROW [esi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq mm2, MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; mm2=(GHIJKLMN) + movq mm3, MMWORD [edx+eax*SIZEOF_JSAMPLE] ; mm3=(OPQRSTUV) + + movq mm4, mm0 + punpcklbw mm0, mm6 ; mm0=(0123) + punpckhbw mm4, mm6 ; mm4=(4567) + movq mm5, mm1 + punpcklbw mm1, mm6 ; mm1=(89AB) + punpckhbw mm5, mm6 ; mm5=(CDEF) + + paddw mm0, mm7 + paddw mm4, mm7 + paddw mm1, mm7 + paddw mm5, mm7 + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_DCTELEM)], mm0 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_DCTELEM)], mm4 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_DCTELEM)], mm1 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_DCTELEM)], mm5 + + movq mm0, mm2 + punpcklbw mm2, mm6 ; mm2=(GHIJ) + punpckhbw mm0, mm6 ; mm0=(KLMN) + movq mm4, mm3 + punpcklbw mm3, mm6 ; mm3=(OPQR) + punpckhbw mm4, mm6 ; mm4=(STUV) + + paddw mm2, mm7 + paddw mm0, mm7 + paddw mm3, mm7 + paddw mm4, mm7 + + movq MMWORD [MMBLOCK(2,0,edi,SIZEOF_DCTELEM)], mm2 + movq MMWORD [MMBLOCK(2,1,edi,SIZEOF_DCTELEM)], mm0 + movq MMWORD [MMBLOCK(3,0,edi,SIZEOF_DCTELEM)], mm3 + movq MMWORD [MMBLOCK(3,1,edi,SIZEOF_DCTELEM)], mm4 + + add esi, byte 4*SIZEOF_JSAMPROW + add edi, byte 4*DCTSIZE*SIZEOF_DCTELEM + dec ecx + jnz short .convloop + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; This implementation is based on an algorithm described in +; "Optimizing subroutines in assembly language: +; An optimization guide for x86 platforms" (https://agner.org/optimize). +; +; GLOBAL(void) +; jsimd_quantize_mmx(JCOEFPTR coef_block, DCTELEM *divisors, +; DCTELEM *workspace); +; + +%define RECIPROCAL(m, n, b) \ + MMBLOCK(DCTSIZE * 0 + (m), (n), (b), SIZEOF_DCTELEM) +%define CORRECTION(m, n, b) \ + MMBLOCK(DCTSIZE * 1 + (m), (n), (b), SIZEOF_DCTELEM) +%define SCALE(m, n, b) \ + MMBLOCK(DCTSIZE * 2 + (m), (n), (b), SIZEOF_DCTELEM) +%define SHIFT(m, n, b) \ + MMBLOCK(DCTSIZE * 3 + (m), (n), (b), SIZEOF_DCTELEM) + +%define coef_block ebp + 8 ; JCOEFPTR coef_block +%define divisors ebp + 12 ; DCTELEM *divisors +%define workspace ebp + 16 ; DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_mmx) + +EXTN(jsimd_quantize_mmx): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + mov esi, POINTER [workspace] + mov edx, POINTER [divisors] + mov edi, JCOEFPTR [coef_block] + mov ah, 2 + ALIGNX 16, 7 +.quantloop1: + mov al, DCTSIZE2/8/2 + ALIGNX 16, 7 +.quantloop2: + movq mm2, MMWORD [MMBLOCK(0,0,esi,SIZEOF_DCTELEM)] + movq mm3, MMWORD [MMBLOCK(0,1,esi,SIZEOF_DCTELEM)] + + movq mm0, mm2 + movq mm1, mm3 + + psraw mm2, (WORD_BIT-1) ; -1 if value < 0, 0 otherwise + psraw mm3, (WORD_BIT-1) + + pxor mm0, mm2 ; val = -val + pxor mm1, mm3 + psubw mm0, mm2 + psubw mm1, mm3 + + ; + ; MMX is an annoyingly crappy instruction set. It has two + ; misfeatures that are causing problems here: + ; + ; - All multiplications are signed. + ; + ; - The second operand for the shifts is not treated as packed. + ; + ; + ; We work around the first problem by implementing this algorithm: + ; + ; unsigned long unsigned_multiply(unsigned short x, unsigned short y) + ; { + ; enum { SHORT_BIT = 16 }; + ; signed short sx = (signed short)x; + ; signed short sy = (signed short)y; + ; signed long sz; + ; + ; sz = (long)sx * (long)sy; /* signed multiply */ + ; + ; if (sx < 0) sz += (long)sy << SHORT_BIT; + ; if (sy < 0) sz += (long)sx << SHORT_BIT; + ; + ; return (unsigned long)sz; + ; } + ; + ; (note that a negative sx adds _sy_ and vice versa) + ; + ; For the second problem, we replace the shift by a multiplication. + ; Unfortunately that means we have to deal with the signed issue again. + ; + + paddw mm0, MMWORD [CORRECTION(0,0,edx)] ; correction + roundfactor + paddw mm1, MMWORD [CORRECTION(0,1,edx)] + + movq mm4, mm0 ; store current value for later + movq mm5, mm1 + pmulhw mm0, MMWORD [RECIPROCAL(0,0,edx)] ; reciprocal + pmulhw mm1, MMWORD [RECIPROCAL(0,1,edx)] + paddw mm0, mm4 ; reciprocal is always negative (MSB=1), + paddw mm1, mm5 ; so we always need to add the initial value + ; (input value is never negative as we + ; inverted it at the start of this routine) + + ; here it gets a bit tricky as both scale + ; and mm0/mm1 can be negative + movq mm6, MMWORD [SCALE(0,0,edx)] ; scale + movq mm7, MMWORD [SCALE(0,1,edx)] + movq mm4, mm0 + movq mm5, mm1 + pmulhw mm0, mm6 + pmulhw mm1, mm7 + + psraw mm6, (WORD_BIT-1) ; determine if scale is negative + psraw mm7, (WORD_BIT-1) + + pand mm6, mm4 ; and add input if it is + pand mm7, mm5 + paddw mm0, mm6 + paddw mm1, mm7 + + psraw mm4, (WORD_BIT-1) ; then check if negative input + psraw mm5, (WORD_BIT-1) + + pand mm4, MMWORD [SCALE(0,0,edx)] ; and add scale if it is + pand mm5, MMWORD [SCALE(0,1,edx)] + paddw mm0, mm4 + paddw mm1, mm5 + + pxor mm0, mm2 ; val = -val + pxor mm1, mm3 + psubw mm0, mm2 + psubw mm1, mm3 + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_DCTELEM)], mm0 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_DCTELEM)], mm1 + + add esi, byte 8*SIZEOF_DCTELEM + add edx, byte 8*SIZEOF_DCTELEM + add edi, byte 8*SIZEOF_JCOEF + dec al + jnz near .quantloop2 + dec ah + jnz near .quantloop1 ; to avoid branch misprediction + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-sse.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-sse.asm new file mode 100644 index 0000000..6cb5f79 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jquant-sse.asm @@ -0,0 +1,204 @@ +; +; jquant.asm - sample data conversion and quantization (SSE & MMX) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_float_sse(JSAMPARRAY sample_data, JDIMENSION start_col, +; FAST_FLOAT *workspace); +; + +%define sample_data ebp + 8 ; JSAMPARRAY sample_data +%define start_col ebp + 12 ; JDIMENSION start_col +%define workspace ebp + 16 ; FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_float_sse) + +EXTN(jsimd_convsamp_float_sse): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + pcmpeqw mm7, mm7 + psllw mm7, 7 + packsswb mm7, mm7 ; mm7 = PB_CENTERJSAMPLE (0x808080..) + + mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *) + mov eax, JDIMENSION [start_col] + mov edi, POINTER [workspace] ; (DCTELEM *) + mov ecx, DCTSIZE/2 + ALIGNX 16, 7 +.convloop: + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq mm0, MMWORD [ebx+eax*SIZEOF_JSAMPLE] + movq mm1, MMWORD [edx+eax*SIZEOF_JSAMPLE] + + psubb mm0, mm7 ; mm0=(01234567) + psubb mm1, mm7 ; mm1=(89ABCDEF) + + punpcklbw mm2, mm0 ; mm2=(*0*1*2*3) + punpckhbw mm0, mm0 ; mm0=(*4*5*6*7) + punpcklbw mm3, mm1 ; mm3=(*8*9*A*B) + punpckhbw mm1, mm1 ; mm1=(*C*D*E*F) + + punpcklwd mm4, mm2 ; mm4=(***0***1) + punpckhwd mm2, mm2 ; mm2=(***2***3) + punpcklwd mm5, mm0 ; mm5=(***4***5) + punpckhwd mm0, mm0 ; mm0=(***6***7) + + psrad mm4, (DWORD_BIT-BYTE_BIT) ; mm4=(01) + psrad mm2, (DWORD_BIT-BYTE_BIT) ; mm2=(23) + cvtpi2ps xmm0, mm4 ; xmm0=(01**) + cvtpi2ps xmm1, mm2 ; xmm1=(23**) + psrad mm5, (DWORD_BIT-BYTE_BIT) ; mm5=(45) + psrad mm0, (DWORD_BIT-BYTE_BIT) ; mm0=(67) + cvtpi2ps xmm2, mm5 ; xmm2=(45**) + cvtpi2ps xmm3, mm0 ; xmm3=(67**) + + punpcklwd mm6, mm3 ; mm6=(***8***9) + punpckhwd mm3, mm3 ; mm3=(***A***B) + punpcklwd mm4, mm1 ; mm4=(***C***D) + punpckhwd mm1, mm1 ; mm1=(***E***F) + + psrad mm6, (DWORD_BIT-BYTE_BIT) ; mm6=(89) + psrad mm3, (DWORD_BIT-BYTE_BIT) ; mm3=(AB) + cvtpi2ps xmm4, mm6 ; xmm4=(89**) + cvtpi2ps xmm5, mm3 ; xmm5=(AB**) + psrad mm4, (DWORD_BIT-BYTE_BIT) ; mm4=(CD) + psrad mm1, (DWORD_BIT-BYTE_BIT) ; mm1=(EF) + cvtpi2ps xmm6, mm4 ; xmm6=(CD**) + cvtpi2ps xmm7, mm1 ; xmm7=(EF**) + + movlhps xmm0, xmm1 ; xmm0=(0123) + movlhps xmm2, xmm3 ; xmm2=(4567) + movlhps xmm4, xmm5 ; xmm4=(89AB) + movlhps xmm6, xmm7 ; xmm6=(CDEF) + + movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm4 + movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm6 + + add esi, byte 2*SIZEOF_JSAMPROW + add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT + dec ecx + jnz near .convloop + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; GLOBAL(void) +; jsimd_quantize_float_sse(JCOEFPTR coef_block, FAST_FLOAT *divisors, +; FAST_FLOAT *workspace); +; + +%define coef_block ebp + 8 ; JCOEFPTR coef_block +%define divisors ebp + 12 ; FAST_FLOAT *divisors +%define workspace ebp + 16 ; FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_float_sse) + +EXTN(jsimd_quantize_float_sse): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + mov esi, POINTER [workspace] + mov edx, POINTER [divisors] + mov edi, JCOEFPTR [coef_block] + mov eax, DCTSIZE2/16 + ALIGNX 16, 7 +.quantloop: + movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(0,1,esi,SIZEOF_FAST_FLOAT)] + mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] + mulps xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(1,1,esi,SIZEOF_FAST_FLOAT)] + mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] + mulps xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)] + + movhlps xmm4, xmm0 + movhlps xmm5, xmm1 + + cvtps2pi mm0, xmm0 + cvtps2pi mm1, xmm1 + cvtps2pi mm4, xmm4 + cvtps2pi mm5, xmm5 + + movhlps xmm6, xmm2 + movhlps xmm7, xmm3 + + cvtps2pi mm2, xmm2 + cvtps2pi mm3, xmm3 + cvtps2pi mm6, xmm6 + cvtps2pi mm7, xmm7 + + packssdw mm0, mm4 + packssdw mm1, mm5 + packssdw mm2, mm6 + packssdw mm3, mm7 + + movq MMWORD [MMBLOCK(0,0,edi,SIZEOF_JCOEF)], mm0 + movq MMWORD [MMBLOCK(0,1,edi,SIZEOF_JCOEF)], mm1 + movq MMWORD [MMBLOCK(1,0,edi,SIZEOF_JCOEF)], mm2 + movq MMWORD [MMBLOCK(1,1,edi,SIZEOF_JCOEF)], mm3 + + add esi, byte 16*SIZEOF_FAST_FLOAT + add edx, byte 16*SIZEOF_FAST_FLOAT + add edi, byte 16*SIZEOF_JCOEF + dec eax + jnz short .quantloop + + emms ; empty MMX state + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jquantf-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jquantf-sse2.asm new file mode 100644 index 0000000..5668f8c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jquantf-sse2.asm @@ -0,0 +1,164 @@ +; +; jquantf.asm - sample data conversion and quantization (SSE & SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_float_sse2(JSAMPARRAY sample_data, JDIMENSION start_col, +; FAST_FLOAT *workspace); +; + +%define sample_data ebp + 8 ; JSAMPARRAY sample_data +%define start_col ebp + 12 ; JDIMENSION start_col +%define workspace ebp + 16 ; FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_float_sse2) + +EXTN(jsimd_convsamp_float_sse2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + pcmpeqw xmm7, xmm7 + psllw xmm7, 7 + packsswb xmm7, xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..) + + mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *) + mov eax, JDIMENSION [start_col] + mov edi, POINTER [workspace] ; (DCTELEM *) + mov ecx, DCTSIZE/2 + ALIGNX 16, 7 +.convloop: + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq xmm0, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] + movq xmm1, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] + + psubb xmm0, xmm7 ; xmm0=(01234567) + psubb xmm1, xmm7 ; xmm1=(89ABCDEF) + + punpcklbw xmm0, xmm0 ; xmm0=(*0*1*2*3*4*5*6*7) + punpcklbw xmm1, xmm1 ; xmm1=(*8*9*A*B*C*D*E*F) + + punpcklwd xmm2, xmm0 ; xmm2=(***0***1***2***3) + punpckhwd xmm0, xmm0 ; xmm0=(***4***5***6***7) + punpcklwd xmm3, xmm1 ; xmm3=(***8***9***A***B) + punpckhwd xmm1, xmm1 ; xmm1=(***C***D***E***F) + + psrad xmm2, (DWORD_BIT-BYTE_BIT) ; xmm2=(0123) + psrad xmm0, (DWORD_BIT-BYTE_BIT) ; xmm0=(4567) + cvtdq2ps xmm2, xmm2 ; xmm2=(0123) + cvtdq2ps xmm0, xmm0 ; xmm0=(4567) + psrad xmm3, (DWORD_BIT-BYTE_BIT) ; xmm3=(89AB) + psrad xmm1, (DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF) + cvtdq2ps xmm3, xmm3 ; xmm3=(89AB) + cvtdq2ps xmm1, xmm1 ; xmm1=(CDEF) + + movaps XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(0,1,edi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(1,1,edi,SIZEOF_FAST_FLOAT)], xmm1 + + add esi, byte 2*SIZEOF_JSAMPROW + add edi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT + dec ecx + jnz short .convloop + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; GLOBAL(void) +; jsimd_quantize_float_sse2(JCOEFPTR coef_block, FAST_FLOAT *divisors, +; FAST_FLOAT *workspace); +; + +%define coef_block ebp + 8 ; JCOEFPTR coef_block +%define divisors ebp + 12 ; FAST_FLOAT *divisors +%define workspace ebp + 16 ; FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_float_sse2) + +EXTN(jsimd_quantize_float_sse2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + mov esi, POINTER [workspace] + mov edx, POINTER [divisors] + mov edi, JCOEFPTR [coef_block] + mov eax, DCTSIZE2/16 + ALIGNX 16, 7 +.quantloop: + movaps xmm0, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(0,1,esi,SIZEOF_FAST_FLOAT)] + mulps xmm0, XMMWORD [XMMBLOCK(0,0,edx,SIZEOF_FAST_FLOAT)] + mulps xmm1, XMMWORD [XMMBLOCK(0,1,edx,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(1,1,esi,SIZEOF_FAST_FLOAT)] + mulps xmm2, XMMWORD [XMMBLOCK(1,0,edx,SIZEOF_FAST_FLOAT)] + mulps xmm3, XMMWORD [XMMBLOCK(1,1,edx,SIZEOF_FAST_FLOAT)] + + cvtps2dq xmm0, xmm0 + cvtps2dq xmm1, xmm1 + cvtps2dq xmm2, xmm2 + cvtps2dq xmm3, xmm3 + + packssdw xmm0, xmm1 + packssdw xmm2, xmm3 + + movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_JCOEF)], xmm0 + movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_JCOEF)], xmm2 + + add esi, byte 16*SIZEOF_FAST_FLOAT + add edx, byte 16*SIZEOF_FAST_FLOAT + add edi, byte 16*SIZEOF_JCOEF + dec eax + jnz short .quantloop + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-avx2.asm new file mode 100644 index 0000000..60ae098 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-avx2.asm @@ -0,0 +1,184 @@ +; +; jquanti.asm - sample data conversion and quantization (AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2018, 2024, D. R. Commander. +; Copyright (C) 2016, Matthieu Darbois. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_avx2(JSAMPARRAY sample_data, JDIMENSION start_col, +; DCTELEM *workspace); +; + +%define sample_data ebp + 8 ; JSAMPARRAY sample_data +%define start_col ebp + 12 ; JDIMENSION start_col +%define workspace ebp + 16 ; DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_avx2) + +EXTN(jsimd_convsamp_avx2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *) + mov eax, JDIMENSION [start_col] + mov edi, POINTER [workspace] ; (DCTELEM *) + + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm0, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] + movq xmm1, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] + + mov ebx, JSAMPROW [esi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm2, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] + movq xmm3, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] + + mov ebx, JSAMPROW [esi+4*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+5*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm4, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] + movq xmm5, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] + + mov ebx, JSAMPROW [esi+6*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+7*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm6, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] + movq xmm7, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] + + vinserti128 ymm0, ymm0, xmm1, 1 + vinserti128 ymm2, ymm2, xmm3, 1 + vinserti128 ymm4, ymm4, xmm5, 1 + vinserti128 ymm6, ymm6, xmm7, 1 + + vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's) + vpunpcklbw ymm0, ymm0, ymm1 + vpunpcklbw ymm2, ymm2, ymm1 + vpunpcklbw ymm4, ymm4, ymm1 + vpunpcklbw ymm6, ymm6, ymm1 + + vpcmpeqw ymm7, ymm7, ymm7 + vpsllw ymm7, ymm7, 7 ; ymm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm2, ymm2, ymm7 + vpaddw ymm4, ymm4, ymm7 + vpaddw ymm6, ymm6, ymm7 + + vmovdqu YMMWORD [YMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], ymm0 + vmovdqu YMMWORD [YMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], ymm2 + vmovdqu YMMWORD [YMMBLOCK(4,0,edi,SIZEOF_DCTELEM)], ymm4 + vmovdqu YMMWORD [YMMBLOCK(6,0,edi,SIZEOF_DCTELEM)], ymm6 + + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; This implementation is based on an algorithm described in +; "Optimizing subroutines in assembly language: +; An optimization guide for x86 platforms" (https://agner.org/optimize). +; +; GLOBAL(void) +; jsimd_quantize_avx2(JCOEFPTR coef_block, DCTELEM *divisors, +; DCTELEM *workspace); +; + +%define RECIPROCAL(m, n, b) \ + YMMBLOCK(DCTSIZE * 0 + (m), (n), (b), SIZEOF_DCTELEM) +%define CORRECTION(m, n, b) \ + YMMBLOCK(DCTSIZE * 1 + (m), (n), (b), SIZEOF_DCTELEM) +%define SCALE(m, n, b) \ + YMMBLOCK(DCTSIZE * 2 + (m), (n), (b), SIZEOF_DCTELEM) + +%define coef_block ebp + 8 ; JCOEFPTR coef_block +%define divisors ebp + 12 ; DCTELEM *divisors +%define workspace ebp + 16 ; DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_avx2) + +EXTN(jsimd_quantize_avx2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + mov esi, POINTER [workspace] + mov edx, POINTER [divisors] + mov edi, JCOEFPTR [coef_block] + + vmovdqu ymm4, [YMMBLOCK(0,0,esi,SIZEOF_DCTELEM)] + vmovdqu ymm5, [YMMBLOCK(2,0,esi,SIZEOF_DCTELEM)] + vmovdqu ymm6, [YMMBLOCK(4,0,esi,SIZEOF_DCTELEM)] + vmovdqu ymm7, [YMMBLOCK(6,0,esi,SIZEOF_DCTELEM)] + vpabsw ymm0, ymm4 + vpabsw ymm1, ymm5 + vpabsw ymm2, ymm6 + vpabsw ymm3, ymm7 + + vpaddw ymm0, YMMWORD [CORRECTION(0,0,edx)] ; correction + roundfactor + vpaddw ymm1, YMMWORD [CORRECTION(2,0,edx)] + vpaddw ymm2, YMMWORD [CORRECTION(4,0,edx)] + vpaddw ymm3, YMMWORD [CORRECTION(6,0,edx)] + vpmulhuw ymm0, YMMWORD [RECIPROCAL(0,0,edx)] ; reciprocal + vpmulhuw ymm1, YMMWORD [RECIPROCAL(2,0,edx)] + vpmulhuw ymm2, YMMWORD [RECIPROCAL(4,0,edx)] + vpmulhuw ymm3, YMMWORD [RECIPROCAL(6,0,edx)] + vpmulhuw ymm0, YMMWORD [SCALE(0,0,edx)] ; scale + vpmulhuw ymm1, YMMWORD [SCALE(2,0,edx)] + vpmulhuw ymm2, YMMWORD [SCALE(4,0,edx)] + vpmulhuw ymm3, YMMWORD [SCALE(6,0,edx)] + + vpsignw ymm0, ymm0, ymm4 + vpsignw ymm1, ymm1, ymm5 + vpsignw ymm2, ymm2, ymm6 + vpsignw ymm3, ymm3, ymm7 + + vmovdqu [YMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], ymm0 + vmovdqu [YMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], ymm1 + vmovdqu [YMMBLOCK(4,0,edi,SIZEOF_DCTELEM)], ymm2 + vmovdqu [YMMBLOCK(6,0,edi,SIZEOF_DCTELEM)], ymm3 + + vzeroupper + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-sse2.asm new file mode 100644 index 0000000..c1edde9 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jquanti-sse2.asm @@ -0,0 +1,197 @@ +; +; jquanti.asm - sample data conversion and quantization (SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_sse2(JSAMPARRAY sample_data, JDIMENSION start_col, +; DCTELEM *workspace); +; + +%define sample_data ebp + 8 ; JSAMPARRAY sample_data +%define start_col ebp + 12 ; JDIMENSION start_col +%define workspace ebp + 16 ; DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_sse2) + +EXTN(jsimd_convsamp_sse2): + push ebp + mov ebp, esp + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved + push esi + push edi + + pxor xmm6, xmm6 ; xmm6=(all 0's) + pcmpeqw xmm7, xmm7 + psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + mov esi, JSAMPARRAY [sample_data] ; (JSAMPROW *) + mov eax, JDIMENSION [start_col] + mov edi, POINTER [workspace] ; (DCTELEM *) + mov ecx, DCTSIZE/4 + ALIGNX 16, 7 +.convloop: + mov ebx, JSAMPROW [esi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq xmm0, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; xmm0=(01234567) + movq xmm1, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] ; xmm1=(89ABCDEF) + + mov ebx, JSAMPROW [esi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov edx, JSAMPROW [esi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq xmm2, XMM_MMWORD [ebx+eax*SIZEOF_JSAMPLE] ; xmm2=(GHIJKLMN) + movq xmm3, XMM_MMWORD [edx+eax*SIZEOF_JSAMPLE] ; xmm3=(OPQRSTUV) + + punpcklbw xmm0, xmm6 ; xmm0=(01234567) + punpcklbw xmm1, xmm6 ; xmm1=(89ABCDEF) + paddw xmm0, xmm7 + paddw xmm1, xmm7 + punpcklbw xmm2, xmm6 ; xmm2=(GHIJKLMN) + punpcklbw xmm3, xmm6 ; xmm3=(OPQRSTUV) + paddw xmm2, xmm7 + paddw xmm3, xmm7 + + movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], xmm0 + movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_DCTELEM)], xmm1 + movdqa XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], xmm2 + movdqa XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_DCTELEM)], xmm3 + + add esi, byte 4*SIZEOF_JSAMPROW + add edi, byte 4*DCTSIZE*SIZEOF_DCTELEM + dec ecx + jnz short .convloop + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + pop ebp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; This implementation is based on an algorithm described in +; "Optimizing subroutines in assembly language: +; An optimization guide for x86 platforms" (https://agner.org/optimize). +; +; GLOBAL(void) +; jsimd_quantize_sse2(JCOEFPTR coef_block, DCTELEM *divisors, +; DCTELEM *workspace); +; + +%define RECIPROCAL(m, n, b) \ + XMMBLOCK(DCTSIZE * 0 + (m), (n), (b), SIZEOF_DCTELEM) +%define CORRECTION(m, n, b) \ + XMMBLOCK(DCTSIZE * 1 + (m), (n), (b), SIZEOF_DCTELEM) +%define SCALE(m, n, b) \ + XMMBLOCK(DCTSIZE * 2 + (m), (n), (b), SIZEOF_DCTELEM) + +%define coef_block ebp + 8 ; JCOEFPTR coef_block +%define divisors ebp + 12 ; DCTELEM *divisors +%define workspace ebp + 16 ; DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_sse2) + +EXTN(jsimd_quantize_sse2): + push ebp + mov ebp, esp +; push ebx ; unused +; push ecx ; unused +; push edx ; need not be preserved + push esi + push edi + + mov esi, POINTER [workspace] + mov edx, POINTER [divisors] + mov edi, JCOEFPTR [coef_block] + mov eax, DCTSIZE2/32 + ALIGNX 16, 7 +.quantloop: + movdqa xmm4, XMMWORD [XMMBLOCK(0,0,esi,SIZEOF_DCTELEM)] + movdqa xmm5, XMMWORD [XMMBLOCK(1,0,esi,SIZEOF_DCTELEM)] + movdqa xmm6, XMMWORD [XMMBLOCK(2,0,esi,SIZEOF_DCTELEM)] + movdqa xmm7, XMMWORD [XMMBLOCK(3,0,esi,SIZEOF_DCTELEM)] + movdqa xmm0, xmm4 + movdqa xmm1, xmm5 + movdqa xmm2, xmm6 + movdqa xmm3, xmm7 + psraw xmm4, (WORD_BIT-1) + psraw xmm5, (WORD_BIT-1) + psraw xmm6, (WORD_BIT-1) + psraw xmm7, (WORD_BIT-1) + pxor xmm0, xmm4 + pxor xmm1, xmm5 + pxor xmm2, xmm6 + pxor xmm3, xmm7 + psubw xmm0, xmm4 ; if (xmm0 < 0) xmm0 = -xmm0; + psubw xmm1, xmm5 ; if (xmm1 < 0) xmm1 = -xmm1; + psubw xmm2, xmm6 ; if (xmm2 < 0) xmm2 = -xmm2; + psubw xmm3, xmm7 ; if (xmm3 < 0) xmm3 = -xmm3; + + paddw xmm0, XMMWORD [CORRECTION(0,0,edx)] ; correction + roundfactor + paddw xmm1, XMMWORD [CORRECTION(1,0,edx)] + paddw xmm2, XMMWORD [CORRECTION(2,0,edx)] + paddw xmm3, XMMWORD [CORRECTION(3,0,edx)] + pmulhuw xmm0, XMMWORD [RECIPROCAL(0,0,edx)] ; reciprocal + pmulhuw xmm1, XMMWORD [RECIPROCAL(1,0,edx)] + pmulhuw xmm2, XMMWORD [RECIPROCAL(2,0,edx)] + pmulhuw xmm3, XMMWORD [RECIPROCAL(3,0,edx)] + pmulhuw xmm0, XMMWORD [SCALE(0,0,edx)] ; scale + pmulhuw xmm1, XMMWORD [SCALE(1,0,edx)] + pmulhuw xmm2, XMMWORD [SCALE(2,0,edx)] + pmulhuw xmm3, XMMWORD [SCALE(3,0,edx)] + + pxor xmm0, xmm4 + pxor xmm1, xmm5 + pxor xmm2, xmm6 + pxor xmm3, xmm7 + psubw xmm0, xmm4 + psubw xmm1, xmm5 + psubw xmm2, xmm6 + psubw xmm3, xmm7 + movdqa XMMWORD [XMMBLOCK(0,0,edi,SIZEOF_DCTELEM)], xmm0 + movdqa XMMWORD [XMMBLOCK(1,0,edi,SIZEOF_DCTELEM)], xmm1 + movdqa XMMWORD [XMMBLOCK(2,0,edi,SIZEOF_DCTELEM)], xmm2 + movdqa XMMWORD [XMMBLOCK(3,0,edi,SIZEOF_DCTELEM)], xmm3 + + add esi, byte 32*SIZEOF_DCTELEM + add edx, byte 32*SIZEOF_DCTELEM + add edi, byte 32*SIZEOF_JCOEF + dec eax + jnz near .quantloop + + pop edi + pop esi +; pop edx ; need not be preserved +; pop ecx ; unused +; pop ebx ; unused + pop ebp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jsimd.c b/src/spice2x/external/libjpeg-turbo/simd/i386/jsimd.c new file mode 100644 index 0000000..b429b0a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jsimd.c @@ -0,0 +1,1312 @@ +/* + * jsimd_i386.c + * + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2009-2011, 2013-2014, 2016, 2018, 2022-2023, D. R. Commander. + * Copyright (C) 2015-2016, 2018, 2022, Matthieu Darbois. + * + * Based on the x86 SIMD extension for IJG JPEG library, + * Copyright (C) 1999-2006, MIYASAKA Masaru. + * For conditions of distribution and use, see copyright notice in jsimdext.inc + * + * This file contains the interface between the "normal" portions + * of the library and the SIMD implementations when running on a + * 32-bit x86 architecture. + */ + +#define JPEG_INTERNALS +#include "../../jinclude.h" +#include "../../jpeglib.h" +#include "../../jsimd.h" +#include "../../jdct.h" +#include "../../jsimddct.h" +#include "../jsimd.h" + +/* + * In the PIC cases, we have no guarantee that constants will keep + * their alignment. This macro allows us to verify it at runtime. + */ +#define IS_ALIGNED(ptr, order) (((unsigned)ptr & ((1 << order) - 1)) == 0) + +#define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */ +#define IS_ALIGNED_AVX(ptr) (IS_ALIGNED(ptr, 5)) /* 32 byte alignment */ + +static THREAD_LOCAL unsigned int simd_support = (unsigned int)(~0); +static THREAD_LOCAL unsigned int simd_huffman = 1; + +/* + * Check what SIMD accelerations are supported. + */ +LOCAL(void) +init_simd(void) +{ +#ifndef NO_GETENV + char env[2] = { 0 }; +#endif + + if (simd_support != ~0U) + return; + + simd_support = jpeg_simd_cpu_support(); + +#ifndef NO_GETENV + /* Force different settings through environment variables */ + if (!GETENV_S(env, 2, "JSIMD_FORCEMMX") && !strcmp(env, "1")) + simd_support &= JSIMD_MMX; + if (!GETENV_S(env, 2, "JSIMD_FORCE3DNOW") && !strcmp(env, "1")) + simd_support &= JSIMD_3DNOW | JSIMD_MMX; + if (!GETENV_S(env, 2, "JSIMD_FORCESSE") && !strcmp(env, "1")) + simd_support &= JSIMD_SSE | JSIMD_MMX; + if (!GETENV_S(env, 2, "JSIMD_FORCESSE2") && !strcmp(env, "1")) + simd_support &= JSIMD_SSE2; + if (!GETENV_S(env, 2, "JSIMD_FORCEAVX2") && !strcmp(env, "1")) + simd_support &= JSIMD_AVX2; + if (!GETENV_S(env, 2, "JSIMD_FORCENONE") && !strcmp(env, "1")) + simd_support = 0; + if (!GETENV_S(env, 2, "JSIMD_NOHUFFENC") && !strcmp(env, "1")) + simd_huffman = 0; +#endif +} + +GLOBAL(int) +jsimd_can_rgb_ycc(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_rgb_ycc_convert_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_rgb_gray(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_rgb_gray_convert_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_rgb_gray_convert_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_ycc_rgb(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_ycc_rgb_convert_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_ycc_rgb565(void) +{ + return 0; +} + +GLOBAL(void) +jsimd_rgb_ycc_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows) +{ + void (*avx2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + void (*sse2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + void (*mmxfct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->in_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_extrgb_ycc_convert_avx2; + sse2fct = jsimd_extrgb_ycc_convert_sse2; + mmxfct = jsimd_extrgb_ycc_convert_mmx; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_extrgbx_ycc_convert_avx2; + sse2fct = jsimd_extrgbx_ycc_convert_sse2; + mmxfct = jsimd_extrgbx_ycc_convert_mmx; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_extbgr_ycc_convert_avx2; + sse2fct = jsimd_extbgr_ycc_convert_sse2; + mmxfct = jsimd_extbgr_ycc_convert_mmx; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_extbgrx_ycc_convert_avx2; + sse2fct = jsimd_extbgrx_ycc_convert_sse2; + mmxfct = jsimd_extbgrx_ycc_convert_mmx; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_extxbgr_ycc_convert_avx2; + sse2fct = jsimd_extxbgr_ycc_convert_sse2; + mmxfct = jsimd_extxbgr_ycc_convert_mmx; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_extxrgb_ycc_convert_avx2; + sse2fct = jsimd_extxrgb_ycc_convert_sse2; + mmxfct = jsimd_extxrgb_ycc_convert_mmx; + break; + default: + avx2fct = jsimd_rgb_ycc_convert_avx2; + sse2fct = jsimd_rgb_ycc_convert_sse2; + mmxfct = jsimd_rgb_ycc_convert_mmx; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); + else if (simd_support & JSIMD_SSE2) + sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); + else + mmxfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); +} + +GLOBAL(void) +jsimd_rgb_gray_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows) +{ + void (*avx2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + void (*sse2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + void (*mmxfct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->in_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_extrgb_gray_convert_avx2; + sse2fct = jsimd_extrgb_gray_convert_sse2; + mmxfct = jsimd_extrgb_gray_convert_mmx; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_extrgbx_gray_convert_avx2; + sse2fct = jsimd_extrgbx_gray_convert_sse2; + mmxfct = jsimd_extrgbx_gray_convert_mmx; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_extbgr_gray_convert_avx2; + sse2fct = jsimd_extbgr_gray_convert_sse2; + mmxfct = jsimd_extbgr_gray_convert_mmx; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_extbgrx_gray_convert_avx2; + sse2fct = jsimd_extbgrx_gray_convert_sse2; + mmxfct = jsimd_extbgrx_gray_convert_mmx; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_extxbgr_gray_convert_avx2; + sse2fct = jsimd_extxbgr_gray_convert_sse2; + mmxfct = jsimd_extxbgr_gray_convert_mmx; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_extxrgb_gray_convert_avx2; + sse2fct = jsimd_extxrgb_gray_convert_sse2; + mmxfct = jsimd_extxrgb_gray_convert_mmx; + break; + default: + avx2fct = jsimd_rgb_gray_convert_avx2; + sse2fct = jsimd_rgb_gray_convert_sse2; + mmxfct = jsimd_rgb_gray_convert_mmx; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); + else if (simd_support & JSIMD_SSE2) + sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); + else + mmxfct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); +} + +GLOBAL(void) +jsimd_ycc_rgb_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION input_row, JSAMPARRAY output_buf, + int num_rows) +{ + void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int); + void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int); + void (*mmxfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_ycc_extrgb_convert_avx2; + sse2fct = jsimd_ycc_extrgb_convert_sse2; + mmxfct = jsimd_ycc_extrgb_convert_mmx; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_ycc_extrgbx_convert_avx2; + sse2fct = jsimd_ycc_extrgbx_convert_sse2; + mmxfct = jsimd_ycc_extrgbx_convert_mmx; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_ycc_extbgr_convert_avx2; + sse2fct = jsimd_ycc_extbgr_convert_sse2; + mmxfct = jsimd_ycc_extbgr_convert_mmx; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_ycc_extbgrx_convert_avx2; + sse2fct = jsimd_ycc_extbgrx_convert_sse2; + mmxfct = jsimd_ycc_extbgrx_convert_mmx; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_ycc_extxbgr_convert_avx2; + sse2fct = jsimd_ycc_extxbgr_convert_sse2; + mmxfct = jsimd_ycc_extxbgr_convert_mmx; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_ycc_extxrgb_convert_avx2; + sse2fct = jsimd_ycc_extxrgb_convert_sse2; + mmxfct = jsimd_ycc_extxrgb_convert_mmx; + break; + default: + avx2fct = jsimd_ycc_rgb_convert_avx2; + sse2fct = jsimd_ycc_rgb_convert_sse2; + mmxfct = jsimd_ycc_rgb_convert_mmx; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows); + else if (simd_support & JSIMD_SSE2) + sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows); + else + mmxfct(cinfo->output_width, input_buf, input_row, output_buf, num_rows); +} + +GLOBAL(void) +jsimd_ycc_rgb565_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION input_row, JSAMPARRAY output_buf, + int num_rows) +{ +} + +GLOBAL(int) +jsimd_can_h2v2_downsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_downsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY output_data) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v2_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); + else if (simd_support & JSIMD_SSE2) + jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); + else + jsimd_h2v2_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, compptr->width_in_blocks, + input_data, output_data); +} + +GLOBAL(void) +jsimd_h2v1_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY output_data) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v1_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); + else if (simd_support & JSIMD_SSE2) + jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); + else + jsimd_h2v1_downsample_mmx(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, compptr->width_in_blocks, + input_data, output_data); +} + +GLOBAL(int) +jsimd_can_h2v2_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v2_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); + else if (simd_support & JSIMD_SSE2) + jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); + else + jsimd_h2v2_upsample_mmx(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); +} + +GLOBAL(void) +jsimd_h2v1_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v1_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); + else if (simd_support & JSIMD_SSE2) + jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); + else + jsimd_h2v1_upsample_mmx(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); +} + +GLOBAL(int) +jsimd_can_h2v2_fancy_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_fancy_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_fancy_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_fancy_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v2_fancy_upsample_avx2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); + else if (simd_support & JSIMD_SSE2) + jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); + else + jsimd_h2v2_fancy_upsample_mmx(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); +} + +GLOBAL(void) +jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v1_fancy_upsample_avx2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); + else if (simd_support & JSIMD_SSE2) + jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); + else + jsimd_h2v1_fancy_upsample_mmx(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); +} + +GLOBAL(int) +jsimd_can_h2v2_merged_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_merged_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_merged_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_merged_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf) +{ + void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + void (*mmxfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_h2v2_extrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extrgb_merged_upsample_sse2; + mmxfct = jsimd_h2v2_extrgb_merged_upsample_mmx; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_h2v2_extrgbx_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extrgbx_merged_upsample_sse2; + mmxfct = jsimd_h2v2_extrgbx_merged_upsample_mmx; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_h2v2_extbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extbgr_merged_upsample_sse2; + mmxfct = jsimd_h2v2_extbgr_merged_upsample_mmx; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_h2v2_extbgrx_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extbgrx_merged_upsample_sse2; + mmxfct = jsimd_h2v2_extbgrx_merged_upsample_mmx; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_h2v2_extxbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extxbgr_merged_upsample_sse2; + mmxfct = jsimd_h2v2_extxbgr_merged_upsample_mmx; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_h2v2_extxrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extxrgb_merged_upsample_sse2; + mmxfct = jsimd_h2v2_extxrgb_merged_upsample_mmx; + break; + default: + avx2fct = jsimd_h2v2_merged_upsample_avx2; + sse2fct = jsimd_h2v2_merged_upsample_sse2; + mmxfct = jsimd_h2v2_merged_upsample_mmx; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); + else if (simd_support & JSIMD_SSE2) + sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); + else + mmxfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); +} + +GLOBAL(void) +jsimd_h2v1_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf) +{ + void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + void (*mmxfct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_h2v1_extrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extrgb_merged_upsample_sse2; + mmxfct = jsimd_h2v1_extrgb_merged_upsample_mmx; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_h2v1_extrgbx_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extrgbx_merged_upsample_sse2; + mmxfct = jsimd_h2v1_extrgbx_merged_upsample_mmx; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_h2v1_extbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extbgr_merged_upsample_sse2; + mmxfct = jsimd_h2v1_extbgr_merged_upsample_mmx; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_h2v1_extbgrx_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extbgrx_merged_upsample_sse2; + mmxfct = jsimd_h2v1_extbgrx_merged_upsample_mmx; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_h2v1_extxbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extxbgr_merged_upsample_sse2; + mmxfct = jsimd_h2v1_extxbgr_merged_upsample_mmx; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_h2v1_extxrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extxrgb_merged_upsample_sse2; + mmxfct = jsimd_h2v1_extxrgb_merged_upsample_mmx; + break; + default: + avx2fct = jsimd_h2v1_merged_upsample_avx2; + sse2fct = jsimd_h2v1_merged_upsample_sse2; + mmxfct = jsimd_h2v1_merged_upsample_mmx; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); + else if (simd_support & JSIMD_SSE2) + sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); + else + mmxfct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); +} + +GLOBAL(int) +jsimd_can_convsamp(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_convsamp_float(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_SSE) + return 1; + if (simd_support & JSIMD_3DNOW) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_convsamp(JSAMPARRAY sample_data, JDIMENSION start_col, + DCTELEM *workspace) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_convsamp_avx2(sample_data, start_col, workspace); + else if (simd_support & JSIMD_SSE2) + jsimd_convsamp_sse2(sample_data, start_col, workspace); + else + jsimd_convsamp_mmx(sample_data, start_col, workspace); +} + +GLOBAL(void) +jsimd_convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col, + FAST_FLOAT *workspace) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_SSE2) + jsimd_convsamp_float_sse2(sample_data, start_col, workspace); + else if (simd_support & JSIMD_SSE) + jsimd_convsamp_float_sse(sample_data, start_col, workspace); + else + jsimd_convsamp_float_3dnow(sample_data, start_col, workspace); +} + +GLOBAL(int) +jsimd_can_fdct_islow(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if ((simd_support & JSIMD_AVX2) && IS_ALIGNED_AVX(jconst_fdct_islow_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_fdct_ifast(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_ifast_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_fdct_float(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + + if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse)) + return 1; + if (simd_support & JSIMD_3DNOW) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_fdct_islow(DCTELEM *data) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_fdct_islow_avx2(data); + else if (simd_support & JSIMD_SSE2) + jsimd_fdct_islow_sse2(data); + else + jsimd_fdct_islow_mmx(data); +} + +GLOBAL(void) +jsimd_fdct_ifast(DCTELEM *data) +{ + if (simd_support == ~0U) + init_simd(); + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2)) + jsimd_fdct_ifast_sse2(data); + else + jsimd_fdct_ifast_mmx(data); +} + +GLOBAL(void) +jsimd_fdct_float(FAST_FLOAT *data) +{ + if (simd_support == ~0U) + init_simd(); + + if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse)) + jsimd_fdct_float_sse(data); + else if (simd_support & JSIMD_3DNOW) + jsimd_fdct_float_3dnow(data); +} + +GLOBAL(int) +jsimd_can_quantize(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_quantize_float(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + + if (simd_support & JSIMD_SSE2) + return 1; + if (simd_support & JSIMD_SSE) + return 1; + if (simd_support & JSIMD_3DNOW) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_quantize_avx2(coef_block, divisors, workspace); + else if (simd_support & JSIMD_SSE2) + jsimd_quantize_sse2(coef_block, divisors, workspace); + else + jsimd_quantize_mmx(coef_block, divisors, workspace); +} + +GLOBAL(void) +jsimd_quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors, + FAST_FLOAT *workspace) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_SSE2) + jsimd_quantize_float_sse2(coef_block, divisors, workspace); + else if (simd_support & JSIMD_SSE) + jsimd_quantize_float_sse(coef_block, divisors, workspace); + else + jsimd_quantize_float_3dnow(coef_block, divisors, workspace); +} + +GLOBAL(int) +jsimd_can_idct_2x2(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(ISLOW_MULT_TYPE) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_idct_4x4(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(ISLOW_MULT_TYPE) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_idct_2x2(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + if (simd_support == ~0U) + init_simd(); + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) + jsimd_idct_2x2_sse2(compptr->dct_table, coef_block, output_buf, + output_col); + else + jsimd_idct_2x2_mmx(compptr->dct_table, coef_block, output_buf, output_col); +} + +GLOBAL(void) +jsimd_idct_4x4(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + if (simd_support == ~0U) + init_simd(); + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) + jsimd_idct_4x4_sse2(compptr->dct_table, coef_block, output_buf, + output_col); + else + jsimd_idct_4x4_mmx(compptr->dct_table, coef_block, output_buf, output_col); +} + +GLOBAL(int) +jsimd_can_idct_islow(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(ISLOW_MULT_TYPE) != 2) + return 0; + + if ((simd_support & JSIMD_AVX2) && IS_ALIGNED_AVX(jconst_idct_islow_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_idct_ifast(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(IFAST_MULT_TYPE) != 2) + return 0; + if (IFAST_SCALE_BITS != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2)) + return 1; + if (simd_support & JSIMD_MMX) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_idct_float(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + if (sizeof(FLOAT_MULT_TYPE) != 4) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2)) + return 1; + if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse)) + return 1; + if (simd_support & JSIMD_3DNOW) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_idct_islow(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_idct_islow_avx2(compptr->dct_table, coef_block, output_buf, + output_col); + else if (simd_support & JSIMD_SSE2) + jsimd_idct_islow_sse2(compptr->dct_table, coef_block, output_buf, + output_col); + else + jsimd_idct_islow_mmx(compptr->dct_table, coef_block, output_buf, + output_col); +} + +GLOBAL(void) +jsimd_idct_ifast(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + if (simd_support == ~0U) + init_simd(); + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2)) + jsimd_idct_ifast_sse2(compptr->dct_table, coef_block, output_buf, + output_col); + else + jsimd_idct_ifast_mmx(compptr->dct_table, coef_block, output_buf, + output_col); +} + +GLOBAL(void) +jsimd_idct_float(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + if (simd_support == ~0U) + init_simd(); + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2)) + jsimd_idct_float_sse2(compptr->dct_table, coef_block, output_buf, + output_col); + else if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_idct_float_sse)) + jsimd_idct_float_sse(compptr->dct_table, coef_block, output_buf, + output_col); + else + jsimd_idct_float_3dnow(compptr->dct_table, coef_block, output_buf, + output_col); +} + +GLOBAL(int) +jsimd_can_huff_encode_one_block(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && simd_huffman && + IS_ALIGNED_SSE(jconst_huff_encode_one_block)) + return 1; + + return 0; +} + +GLOBAL(JOCTET *) +jsimd_huff_encode_one_block(void *state, JOCTET *buffer, JCOEFPTR block, + int last_dc_val, c_derived_tbl *dctbl, + c_derived_tbl *actbl) +{ + return jsimd_huff_encode_one_block_sse2(state, buffer, block, last_dc_val, + dctbl, actbl); +} + +GLOBAL(int) +jsimd_can_encode_mcu_AC_first_prepare(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (SIZEOF_SIZE_T != 4) + return 0; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_encode_mcu_AC_first_prepare(const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *values, size_t *zerobits) +{ + jsimd_encode_mcu_AC_first_prepare_sse2(block, jpeg_natural_order_start, + Sl, Al, values, zerobits); +} + +GLOBAL(int) +jsimd_can_encode_mcu_AC_refine_prepare(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (SIZEOF_SIZE_T != 4) + return 0; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_encode_mcu_AC_refine_prepare(const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *absvalues, size_t *bits) +{ + return jsimd_encode_mcu_AC_refine_prepare_sse2(block, + jpeg_natural_order_start, + Sl, Al, absvalues, bits); +} diff --git a/src/spice2x/external/libjpeg-turbo/simd/i386/jsimdcpu.asm b/src/spice2x/external/libjpeg-turbo/simd/i386/jsimdcpu.asm new file mode 100644 index 0000000..df80f17 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/i386/jsimdcpu.asm @@ -0,0 +1,131 @@ +; +; jsimdcpu.asm - SIMD instruction support check +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 32 +; +; Check if the CPU supports SIMD instructions +; +; GLOBAL(unsigned int) +; jpeg_simd_cpu_support(void) +; + + align 32 + GLOBAL_FUNCTION(jpeg_simd_cpu_support) + +EXTN(jpeg_simd_cpu_support): + push ebx +; push ecx ; need not be preserved +; push edx ; need not be preserved +; push esi ; unused + push edi + + xor edi, edi ; simd support flag + + pushfd + pop eax + mov edx, eax + xor eax, 1<<21 ; flip ID bit in EFLAGS + push eax + popfd + pushfd + pop eax + xor eax, edx + jz near .return ; CPUID is not supported + + ; Check whether CPUID leaf 07H is supported + ; (leaf 07H is used to check for AVX2 instruction support) + xor eax, eax + cpuid + test eax, eax + jz near .return + cmp eax, 7 + jl short .no_avx2 ; Maximum leaf < 07H + + ; Check for AVX2 instruction support + mov eax, 7 + xor ecx, ecx + cpuid + mov eax, ebx + test eax, 1<<5 ; bit5:AVX2 + jz short .no_avx2 + + ; Check for AVX2 O/S support + mov eax, 1 + xor ecx, ecx + cpuid + test ecx, 1<<27 + jz short .no_avx2 ; O/S does not support XSAVE + test ecx, 1<<28 + jz short .no_avx2 ; CPU does not support AVX2 + + xor ecx, ecx + xgetbv + and eax, 6 + cmp eax, 6 ; O/S does not manage XMM/YMM state + ; using XSAVE + jnz short .no_avx2 + + or edi, JSIMD_AVX2 +.no_avx2: + + ; Check CPUID leaf 01H for MMX, SSE, and SSE2 support + xor eax, eax + inc eax + cpuid + mov eax, edx ; eax = Standard feature flags + + ; Check for MMX instruction support + test eax, 1<<23 ; bit23:MMX + jz short .no_mmx + or edi, byte JSIMD_MMX +.no_mmx: + test eax, 1<<25 ; bit25:SSE + jz short .no_sse + or edi, byte JSIMD_SSE +.no_sse: + test eax, 1<<26 ; bit26:SSE2 + jz short .no_sse2 + or edi, byte JSIMD_SSE2 +.no_sse2: + + ; Check for 3DNow! instruction support + mov eax, 0x80000000 + cpuid + cmp eax, 0x80000000 + jbe short .return + + mov eax, 0x80000001 + cpuid + mov eax, edx ; eax = Extended feature flags + + test eax, 1<<31 ; bit31:3DNow!(vendor independent) + jz short .no_3dnow + or edi, byte JSIMD_3DNOW +.no_3dnow: + +.return: + mov eax, edi + + pop edi +; pop esi ; unused +; pop edx ; need not be preserved +; pop ecx ; need not be preserved + pop ebx + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/jsimd.h b/src/spice2x/external/libjpeg-turbo/simd/jsimd.h new file mode 100644 index 0000000..a28754a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/jsimd.h @@ -0,0 +1,1258 @@ +/* + * simd/jsimd.h + * + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2011, 2014-2016, 2018, 2020, 2022, D. R. Commander. + * Copyright (C) 2013-2014, MIPS Technologies, Inc., California. + * Copyright (C) 2014, Linaro Limited. + * Copyright (C) 2015-2016, 2018, 2022, Matthieu Darbois. + * Copyright (C) 2016-2018, Loongson Technology Corporation Limited, BeiJing. + * Copyright (C) 2020, Arm Limited. + * + * Based on the x86 SIMD extension for IJG JPEG library, + * Copyright (C) 1999-2006, MIYASAKA Masaru. + * For conditions of distribution and use, see copyright notice in jsimdext.inc + * + */ + +/* Bitmask for supported acceleration methods */ + +#define JSIMD_NONE 0x00 +#define JSIMD_MMX 0x01 +#define JSIMD_3DNOW 0x02 +#define JSIMD_SSE 0x04 +#define JSIMD_SSE2 0x08 +#define JSIMD_NEON 0x10 +#define JSIMD_DSPR2 0x20 +#define JSIMD_ALTIVEC 0x40 +#define JSIMD_AVX2 0x80 +#define JSIMD_MMI 0x100 + +/* SIMD Ext: retrieve SIMD/CPU information */ +EXTERN(unsigned int) jpeg_simd_cpu_support(void); + +/* RGB & extended RGB --> YCC Colorspace Conversion */ +EXTERN(void) jsimd_rgb_ycc_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_ycc_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_ycc_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_ycc_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_ycc_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_ycc_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +extern const int jconst_rgb_ycc_convert_sse2[]; +EXTERN(void) jsimd_rgb_ycc_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_ycc_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_ycc_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_ycc_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_ycc_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_ycc_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +extern const int jconst_rgb_ycc_convert_avx2[]; +EXTERN(void) jsimd_rgb_ycc_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_ycc_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_ycc_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_ycc_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_ycc_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_ycc_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +EXTERN(void) jsimd_rgb_ycc_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_ycc_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_ycc_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_ycc_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_ycc_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_ycc_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +#ifndef NEON_INTRINSICS + +EXTERN(void) jsimd_extrgb_ycc_convert_neon_slowld3 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_neon_slowld3 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +#endif + +EXTERN(void) jsimd_rgb_ycc_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_ycc_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_ycc_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_ycc_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_ycc_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_ycc_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +EXTERN(void) jsimd_rgb_ycc_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_ycc_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_ycc_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_ycc_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_ycc_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_ycc_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +EXTERN(void) jsimd_rgb_ycc_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_ycc_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_ycc_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_ycc_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_ycc_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_ycc_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_ycc_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +/* RGB & extended RGB --> Grayscale Colorspace Conversion */ +EXTERN(void) jsimd_rgb_gray_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_gray_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_gray_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_gray_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_gray_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_gray_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_gray_convert_mmx + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +extern const int jconst_rgb_gray_convert_sse2[]; +EXTERN(void) jsimd_rgb_gray_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_gray_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_gray_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_gray_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_gray_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_gray_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_gray_convert_sse2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +extern const int jconst_rgb_gray_convert_avx2[]; +EXTERN(void) jsimd_rgb_gray_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_gray_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_gray_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_gray_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_gray_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_gray_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_gray_convert_avx2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +EXTERN(void) jsimd_rgb_gray_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_gray_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_gray_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_gray_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_gray_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_gray_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_gray_convert_neon + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +EXTERN(void) jsimd_rgb_gray_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_gray_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_gray_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_gray_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_gray_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_gray_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_gray_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +EXTERN(void) jsimd_rgb_gray_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_gray_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_gray_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_gray_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_gray_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_gray_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_gray_convert_mmi + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +EXTERN(void) jsimd_rgb_gray_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgb_gray_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extrgbx_gray_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgr_gray_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extbgrx_gray_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxbgr_gray_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); +EXTERN(void) jsimd_extxrgb_gray_convert_altivec + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows); + +/* YCC --> RGB & extended RGB Colorspace Conversion */ +EXTERN(void) jsimd_ycc_rgb_convert_mmx + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgb_convert_mmx + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgbx_convert_mmx + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_mmx + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgrx_convert_mmx + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxbgr_convert_mmx + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxrgb_convert_mmx + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +extern const int jconst_ycc_rgb_convert_sse2[]; +EXTERN(void) jsimd_ycc_rgb_convert_sse2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgb_convert_sse2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgbx_convert_sse2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_sse2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgrx_convert_sse2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxbgr_convert_sse2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxrgb_convert_sse2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +extern const int jconst_ycc_rgb_convert_avx2[]; +EXTERN(void) jsimd_ycc_rgb_convert_avx2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgb_convert_avx2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgbx_convert_avx2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_avx2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgrx_convert_avx2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxbgr_convert_avx2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxrgb_convert_avx2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +EXTERN(void) jsimd_ycc_rgb_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgb_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgbx_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgrx_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxbgr_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxrgb_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_rgb565_convert_neon + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +#ifndef NEON_INTRINSICS + +EXTERN(void) jsimd_ycc_extrgb_convert_neon_slowst3 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_neon_slowst3 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +#endif + +EXTERN(void) jsimd_ycc_rgb_convert_dspr2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgb_convert_dspr2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgbx_convert_dspr2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_dspr2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgrx_convert_dspr2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxbgr_convert_dspr2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxrgb_convert_dspr2 + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +EXTERN(void) jsimd_ycc_rgb_convert_mmi + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgb_convert_mmi + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgbx_convert_mmi + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_mmi + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgrx_convert_mmi + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxbgr_convert_mmi + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxrgb_convert_mmi + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +EXTERN(void) jsimd_ycc_rgb_convert_altivec + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgb_convert_altivec + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extrgbx_convert_altivec + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgr_convert_altivec + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extbgrx_convert_altivec + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxbgr_convert_altivec + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); +EXTERN(void) jsimd_ycc_extxrgb_convert_altivec + (JDIMENSION out_width, JSAMPIMAGE input_buf, JDIMENSION input_row, + JSAMPARRAY output_buf, int num_rows); + +/* NULL Colorspace Conversion */ +EXTERN(void) jsimd_c_null_convert_dspr2 + (JDIMENSION img_width, JSAMPARRAY input_buf, JSAMPIMAGE output_buf, + JDIMENSION output_row, int num_rows, int num_components); + +/* h2v1 Downsampling */ +EXTERN(void) jsimd_h2v1_downsample_mmx + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v1_downsample_sse2 + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v1_downsample_avx2 + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v1_downsample_neon + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v1_downsample_dspr2 + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v1_downsample_altivec + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +/* h2v2 Downsampling */ +EXTERN(void) jsimd_h2v2_downsample_mmx + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v2_downsample_sse2 + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v2_downsample_avx2 + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v2_downsample_neon + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v2_downsample_dspr2 + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v2_downsample_mmi + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +EXTERN(void) jsimd_h2v2_downsample_altivec + (JDIMENSION image_width, int max_v_samp_factor, JDIMENSION v_samp_factor, + JDIMENSION width_in_blocks, JSAMPARRAY input_data, JSAMPARRAY output_data); + +/* h2v2 Smooth Downsampling */ +EXTERN(void) jsimd_h2v2_smooth_downsample_dspr2 + (JSAMPARRAY input_data, JSAMPARRAY output_data, JDIMENSION v_samp_factor, + int max_v_samp_factor, int smoothing_factor, JDIMENSION width_in_blocks, + JDIMENSION image_width); + + +/* Upsampling */ +EXTERN(void) jsimd_h2v1_upsample_mmx + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_upsample_mmx + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_upsample_sse2 + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_upsample_sse2 + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_upsample_avx2 + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_upsample_avx2 + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_upsample_neon + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_upsample_neon + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_upsample_dspr2 + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_upsample_dspr2 + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_int_upsample_dspr2 + (UINT8 h_expand, UINT8 v_expand, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr, JDIMENSION output_width, + int max_v_samp_factor); + +EXTERN(void) jsimd_h2v1_upsample_altivec + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_upsample_altivec + (int max_v_samp_factor, JDIMENSION output_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +/* Fancy Upsampling */ +EXTERN(void) jsimd_h2v1_fancy_upsample_mmx + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_fancy_upsample_mmx + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +extern const int jconst_fancy_upsample_sse2[]; +EXTERN(void) jsimd_h2v1_fancy_upsample_sse2 + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_fancy_upsample_sse2 + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +extern const int jconst_fancy_upsample_avx2[]; +EXTERN(void) jsimd_h2v1_fancy_upsample_avx2 + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_fancy_upsample_avx2 + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_fancy_upsample_neon + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_fancy_upsample_neon + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h1v2_fancy_upsample_neon + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_fancy_upsample_dspr2 + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_fancy_upsample_dspr2 + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_fancy_upsample_mmi + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_fancy_upsample_mmi + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +EXTERN(void) jsimd_h2v1_fancy_upsample_altivec + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); +EXTERN(void) jsimd_h2v2_fancy_upsample_altivec + (int max_v_samp_factor, JDIMENSION downsampled_width, JSAMPARRAY input_data, + JSAMPARRAY *output_data_ptr); + +/* Merged Upsampling */ +EXTERN(void) jsimd_h2v1_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v2_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_mmx + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +extern const int jconst_merged_upsample_sse2[]; +EXTERN(void) jsimd_h2v1_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v2_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_sse2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +extern const int jconst_merged_upsample_avx2[]; +EXTERN(void) jsimd_h2v1_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v2_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_avx2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v1_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v2_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_neon + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v1_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); + +EXTERN(void) jsimd_h2v2_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); +EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_dspr2 + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf, JSAMPLE *range); + +EXTERN(void) jsimd_h2v1_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v2_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_mmi + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v1_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgb_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extrgbx_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgr_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extbgrx_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxbgr_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v1_extxrgb_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +EXTERN(void) jsimd_h2v2_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgb_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extrgbx_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgr_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extbgrx_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxbgr_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); +EXTERN(void) jsimd_h2v2_extxrgb_merged_upsample_altivec + (JDIMENSION output_width, JSAMPIMAGE input_buf, JDIMENSION in_row_group_ctr, + JSAMPARRAY output_buf); + +/* Sample Conversion */ +EXTERN(void) jsimd_convsamp_mmx + (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace); + +EXTERN(void) jsimd_convsamp_sse2 + (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace); + +EXTERN(void) jsimd_convsamp_avx2 + (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace); + +EXTERN(void) jsimd_convsamp_neon + (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace); + +EXTERN(void) jsimd_convsamp_dspr2 + (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace); + +EXTERN(void) jsimd_convsamp_altivec + (JSAMPARRAY sample_data, JDIMENSION start_col, DCTELEM *workspace); + +/* Floating Point Sample Conversion */ +EXTERN(void) jsimd_convsamp_float_3dnow + (JSAMPARRAY sample_data, JDIMENSION start_col, FAST_FLOAT *workspace); + +EXTERN(void) jsimd_convsamp_float_sse + (JSAMPARRAY sample_data, JDIMENSION start_col, FAST_FLOAT *workspace); + +EXTERN(void) jsimd_convsamp_float_sse2 + (JSAMPARRAY sample_data, JDIMENSION start_col, FAST_FLOAT *workspace); + +EXTERN(void) jsimd_convsamp_float_dspr2 + (JSAMPARRAY sample_data, JDIMENSION start_col, FAST_FLOAT *workspace); + +/* Accurate Integer Forward DCT */ +EXTERN(void) jsimd_fdct_islow_mmx(DCTELEM *data); + +extern const int jconst_fdct_islow_sse2[]; +EXTERN(void) jsimd_fdct_islow_sse2(DCTELEM *data); + +extern const int jconst_fdct_islow_avx2[]; +EXTERN(void) jsimd_fdct_islow_avx2(DCTELEM *data); + +EXTERN(void) jsimd_fdct_islow_neon(DCTELEM *data); + +EXTERN(void) jsimd_fdct_islow_dspr2(DCTELEM *data); + +EXTERN(void) jsimd_fdct_islow_mmi(DCTELEM *data); + +EXTERN(void) jsimd_fdct_islow_altivec(DCTELEM *data); + +/* Fast Integer Forward DCT */ +EXTERN(void) jsimd_fdct_ifast_mmx(DCTELEM *data); + +extern const int jconst_fdct_ifast_sse2[]; +EXTERN(void) jsimd_fdct_ifast_sse2(DCTELEM *data); + +EXTERN(void) jsimd_fdct_ifast_neon(DCTELEM *data); + +EXTERN(void) jsimd_fdct_ifast_dspr2(DCTELEM *data); + +EXTERN(void) jsimd_fdct_ifast_mmi(DCTELEM *data); + +EXTERN(void) jsimd_fdct_ifast_altivec(DCTELEM *data); + +/* Floating Point Forward DCT */ +EXTERN(void) jsimd_fdct_float_3dnow(FAST_FLOAT *data); + +extern const int jconst_fdct_float_sse[]; +EXTERN(void) jsimd_fdct_float_sse(FAST_FLOAT *data); + +/* Quantization */ +EXTERN(void) jsimd_quantize_mmx + (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace); + +EXTERN(void) jsimd_quantize_sse2 + (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace); + +EXTERN(void) jsimd_quantize_avx2 + (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace); + +EXTERN(void) jsimd_quantize_neon + (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace); + +EXTERN(void) jsimd_quantize_dspr2 + (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace); + +EXTERN(void) jsimd_quantize_mmi + (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace); + +EXTERN(void) jsimd_quantize_altivec + (JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace); + +/* Floating Point Quantization */ +EXTERN(void) jsimd_quantize_float_3dnow + (JCOEFPTR coef_block, FAST_FLOAT *divisors, FAST_FLOAT *workspace); + +EXTERN(void) jsimd_quantize_float_sse + (JCOEFPTR coef_block, FAST_FLOAT *divisors, FAST_FLOAT *workspace); + +EXTERN(void) jsimd_quantize_float_sse2 + (JCOEFPTR coef_block, FAST_FLOAT *divisors, FAST_FLOAT *workspace); + +EXTERN(void) jsimd_quantize_float_dspr2 + (JCOEFPTR coef_block, FAST_FLOAT *divisors, FAST_FLOAT *workspace); + +/* Scaled Inverse DCT */ +EXTERN(void) jsimd_idct_2x2_mmx + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) jsimd_idct_4x4_mmx + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +extern const int jconst_idct_red_sse2[]; +EXTERN(void) jsimd_idct_2x2_sse2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) jsimd_idct_4x4_sse2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_2x2_neon + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) jsimd_idct_4x4_neon + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_2x2_dspr2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) jsimd_idct_4x4_dspr2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col, int *workspace); +EXTERN(void) jsimd_idct_6x6_dspr2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); +EXTERN(void) jsimd_idct_12x12_pass1_dspr2 + (JCOEFPTR coef_block, void *dct_table, int *workspace); +EXTERN(void) jsimd_idct_12x12_pass2_dspr2 + (int *workspace, int *output); + +/* Accurate Integer Inverse DCT */ +EXTERN(void) jsimd_idct_islow_mmx + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +extern const int jconst_idct_islow_sse2[]; +EXTERN(void) jsimd_idct_islow_sse2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +extern const int jconst_idct_islow_avx2[]; +EXTERN(void) jsimd_idct_islow_avx2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_islow_neon + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_islow_dspr2 + (void *dct_table, JCOEFPTR coef_block, int *output_buf, JSAMPLE *output_col); + +EXTERN(void) jsimd_idct_islow_mmi + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_islow_altivec + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +/* Fast Integer Inverse DCT */ +EXTERN(void) jsimd_idct_ifast_mmx + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +extern const int jconst_idct_ifast_sse2[]; +EXTERN(void) jsimd_idct_ifast_sse2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_ifast_neon + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_ifast_cols_dspr2 + (JCOEF *inptr, IFAST_MULT_TYPE *quantptr, DCTELEM *wsptr, + const int *idct_coefs); +EXTERN(void) jsimd_idct_ifast_rows_dspr2 + (DCTELEM *wsptr, JSAMPARRAY output_buf, JDIMENSION output_col, + const int *idct_coefs); + +EXTERN(void) jsimd_idct_ifast_mmi + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +EXTERN(void) jsimd_idct_ifast_altivec + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +/* Floating Point Inverse DCT */ +EXTERN(void) jsimd_idct_float_3dnow + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +extern const int jconst_idct_float_sse[]; +EXTERN(void) jsimd_idct_float_sse + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +extern const int jconst_idct_float_sse2[]; +EXTERN(void) jsimd_idct_float_sse2 + (void *dct_table, JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col); + +/* Huffman coding */ +extern const int jconst_huff_encode_one_block[]; +EXTERN(JOCTET *) jsimd_huff_encode_one_block_sse2 + (void *state, JOCTET *buffer, JCOEFPTR block, int last_dc_val, + c_derived_tbl *dctbl, c_derived_tbl *actbl); + +EXTERN(JOCTET *) jsimd_huff_encode_one_block_neon + (void *state, JOCTET *buffer, JCOEFPTR block, int last_dc_val, + c_derived_tbl *dctbl, c_derived_tbl *actbl); + +#ifndef NEON_INTRINSICS + +EXTERN(JOCTET *) jsimd_huff_encode_one_block_neon_slowtbl + (void *state, JOCTET *buffer, JCOEFPTR block, int last_dc_val, + c_derived_tbl *dctbl, c_derived_tbl *actbl); + +#endif + +/* Progressive Huffman encoding */ +EXTERN(void) jsimd_encode_mcu_AC_first_prepare_sse2 + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *values, size_t *zerobits); + +EXTERN(void) jsimd_encode_mcu_AC_first_prepare_neon + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *values, size_t *zerobits); + +EXTERN(int) jsimd_encode_mcu_AC_refine_prepare_sse2 + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *absvalues, size_t *bits); + +EXTERN(int) jsimd_encode_mcu_AC_refine_prepare_neon + (const JCOEF *block, const int *jpeg_natural_order_start, int Sl, int Al, + UJCOEF *absvalues, size_t *bits); diff --git a/src/spice2x/external/libjpeg-turbo/simd/nasm/jcolsamp.inc b/src/spice2x/external/libjpeg-turbo/simd/nasm/jcolsamp.inc new file mode 100644 index 0000000..6f6d7f2 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/nasm/jcolsamp.inc @@ -0,0 +1,135 @@ +; +; jcolsamp.inc - private declarations for color conversion & up/downsampling +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc + +; -------------------------------------------------------------------------- + +; pseudo-resisters to make ordering of RGB configurable +; +%if RGB_RED == 0 +%define mmA mm0 +%define mmB mm1 +%define xmmA xmm0 +%define xmmB xmm1 +%define ymmA ymm0 +%define ymmB ymm1 +%elif RGB_GREEN == 0 +%define mmA mm2 +%define mmB mm3 +%define xmmA xmm2 +%define xmmB xmm3 +%define ymmA ymm2 +%define ymmB ymm3 +%elif RGB_BLUE == 0 +%define mmA mm4 +%define mmB mm5 +%define xmmA xmm4 +%define xmmB xmm5 +%define ymmA ymm4 +%define ymmB ymm5 +%else +%define mmA mm6 +%define mmB mm7 +%define xmmA xmm6 +%define xmmB xmm7 +%define ymmA ymm6 +%define ymmB ymm7 +%endif + +%if RGB_RED == 1 +%define mmC mm0 +%define mmD mm1 +%define xmmC xmm0 +%define xmmD xmm1 +%define ymmC ymm0 +%define ymmD ymm1 +%elif RGB_GREEN == 1 +%define mmC mm2 +%define mmD mm3 +%define xmmC xmm2 +%define xmmD xmm3 +%define ymmC ymm2 +%define ymmD ymm3 +%elif RGB_BLUE == 1 +%define mmC mm4 +%define mmD mm5 +%define xmmC xmm4 +%define xmmD xmm5 +%define ymmC ymm4 +%define ymmD ymm5 +%else +%define mmC mm6 +%define mmD mm7 +%define xmmC xmm6 +%define xmmD xmm7 +%define ymmC ymm6 +%define ymmD ymm7 +%endif + +%if RGB_RED == 2 +%define mmE mm0 +%define mmF mm1 +%define xmmE xmm0 +%define xmmF xmm1 +%define ymmE ymm0 +%define ymmF ymm1 +%elif RGB_GREEN == 2 +%define mmE mm2 +%define mmF mm3 +%define xmmE xmm2 +%define xmmF xmm3 +%define ymmE ymm2 +%define ymmF ymm3 +%elif RGB_BLUE == 2 +%define mmE mm4 +%define mmF mm5 +%define xmmE xmm4 +%define xmmF xmm5 +%define ymmE ymm4 +%define ymmF ymm5 +%else +%define mmE mm6 +%define mmF mm7 +%define xmmE xmm6 +%define xmmF xmm7 +%define ymmE ymm6 +%define ymmF ymm7 +%endif + +%if RGB_RED == 3 +%define mmG mm0 +%define mmH mm1 +%define xmmG xmm0 +%define xmmH xmm1 +%define ymmG ymm0 +%define ymmH ymm1 +%elif RGB_GREEN == 3 +%define mmG mm2 +%define mmH mm3 +%define xmmG xmm2 +%define xmmH xmm3 +%define ymmG ymm2 +%define ymmH ymm3 +%elif RGB_BLUE == 3 +%define mmG mm4 +%define mmH mm5 +%define xmmG xmm4 +%define xmmH xmm5 +%define ymmG ymm4 +%define ymmH ymm5 +%else +%define mmG mm6 +%define mmH mm7 +%define xmmG xmm6 +%define xmmH xmm7 +%define ymmG ymm6 +%define ymmH ymm7 +%endif + +; -------------------------------------------------------------------------- diff --git a/src/spice2x/external/libjpeg-turbo/simd/nasm/jdct.inc b/src/spice2x/external/libjpeg-turbo/simd/nasm/jdct.inc new file mode 100644 index 0000000..9192f66 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/nasm/jdct.inc @@ -0,0 +1,31 @@ +; +; jdct.inc - private declarations for forward & reverse DCT subsystems +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2018, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc + +; Each IDCT routine is responsible for range-limiting its results and +; converting them to unsigned form (0..MAXJSAMPLE). The raw outputs could +; be quite far out of range if the input data is corrupt, so a bulletproof +; range-limiting step is required. We use a mask-and-table-lookup method +; to do the combined operations quickly. +; +%define RANGE_MASK (MAXJSAMPLE * 4 + 3) ; 2 bits wider than legal samples + +%define ROW(n, b, s) ((b) + (n) * (s)) +%define COL(n, b, s) ((b) + (n) * (s) * DCTSIZE) + +%define DWBLOCK(m, n, b, s) \ + ((b) + (m) * DCTSIZE * (s) + (n) * SIZEOF_DWORD) +%define MMBLOCK(m, n, b, s) \ + ((b) + (m) * DCTSIZE * (s) + (n) * SIZEOF_MMWORD) +%define XMMBLOCK(m, n, b, s) \ + ((b) + (m) * DCTSIZE * (s) + (n) * SIZEOF_XMMWORD) +%define YMMBLOCK(m, n, b, s) \ + ((b) + (m) * DCTSIZE * (s) + (n) * SIZEOF_YMMWORD) + +; -------------------------------------------------------------------------- diff --git a/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc b/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc new file mode 100644 index 0000000..667024a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc @@ -0,0 +1,93 @@ +; +; Automatically generated include file from jsimdcfg.inc.h +; +; +; -- jpeglib.h +; +%define DCTSIZE 8 +%define DCTSIZE2 64 +; +; -- jmorecfg.h +; +%define RGB_RED 0 +%define RGB_GREEN 1 +%define RGB_BLUE 2 +%define RGB_PIXELSIZE 3 +%define EXT_RGB_RED 0 +%define EXT_RGB_GREEN 1 +%define EXT_RGB_BLUE 2 +%define EXT_RGB_PIXELSIZE 3 +%define EXT_RGBX_RED 0 +%define EXT_RGBX_GREEN 1 +%define EXT_RGBX_BLUE 2 +%define EXT_RGBX_PIXELSIZE 4 +%define EXT_BGR_RED 2 +%define EXT_BGR_GREEN 1 +%define EXT_BGR_BLUE 0 +%define EXT_BGR_PIXELSIZE 3 +%define EXT_BGRX_RED 2 +%define EXT_BGRX_GREEN 1 +%define EXT_BGRX_BLUE 0 +%define EXT_BGRX_PIXELSIZE 4 +%define EXT_XBGR_RED 3 +%define EXT_XBGR_GREEN 2 +%define EXT_XBGR_BLUE 1 +%define EXT_XBGR_PIXELSIZE 4 +%define EXT_XRGB_RED 1 +%define EXT_XRGB_GREEN 2 +%define EXT_XRGB_BLUE 3 +%define EXT_XRGB_PIXELSIZE 4 +%define RGBX_FILLER_0XFF 1 +; Representation of a single sample (pixel element value). +; On this SIMD implementation, this must be 'unsigned char'. +; +%define JSAMPLE byte ; unsigned char +%define SIZEOF_JSAMPLE SIZEOF_BYTE ; sizeof(JSAMPLE) +%define CENTERJSAMPLE 128 +; Representation of a DCT frequency coefficient. +; On this SIMD implementation, this must be 'short'. +; +%define JCOEF word ; short +%define SIZEOF_JCOEF SIZEOF_WORD ; sizeof(JCOEF) +; Datatype used for image dimensions. +; On this SIMD implementation, this must be 'unsigned int'. +; +%define JDIMENSION dword ; unsigned int +%define SIZEOF_JDIMENSION SIZEOF_DWORD ; sizeof(JDIMENSION) +%define JSAMPROW POINTER ; JSAMPLE * (jpeglib.h) +%define JSAMPARRAY POINTER ; JSAMPROW * (jpeglib.h) +%define JSAMPIMAGE POINTER ; JSAMPARRAY * (jpeglib.h) +%define JCOEFPTR POINTER ; JCOEF * (jpeglib.h) +%define SIZEOF_JSAMPROW SIZEOF_POINTER ; sizeof(JSAMPROW) +%define SIZEOF_JSAMPARRAY SIZEOF_POINTER ; sizeof(JSAMPARRAY) +%define SIZEOF_JSAMPIMAGE SIZEOF_POINTER ; sizeof(JSAMPIMAGE) +%define SIZEOF_JCOEFPTR SIZEOF_POINTER ; sizeof(JCOEFPTR) +; +; -- jdct.h +; +; A forward DCT routine is given a pointer to a work area of type DCTELEM[]; +; the DCT is to be performed in-place in that buffer. +; To maximize parallelism, Type DCTELEM is changed to short (originally, int). +; +%define DCTELEM word ; short +%define SIZEOF_DCTELEM SIZEOF_WORD ; sizeof(DCTELEM) +%define float FP32 ; float +%define SIZEOF_FAST_FLOAT SIZEOF_FP32 ; sizeof(float) +; To maximize parallelism, Type short is changed to short. +; +%define ISLOW_MULT_TYPE word ; must be short +%define SIZEOF_ISLOW_MULT_TYPE SIZEOF_WORD ; sizeof(ISLOW_MULT_TYPE) +%define IFAST_MULT_TYPE word ; must be short +%define SIZEOF_IFAST_MULT_TYPE SIZEOF_WORD ; sizeof(IFAST_MULT_TYPE) +%define IFAST_SCALE_BITS 2 ; fractional bits in scale factors +%define FLOAT_MULT_TYPE FP32 ; must be float +%define SIZEOF_FLOAT_MULT_TYPE SIZEOF_FP32 ; sizeof(FLOAT_MULT_TYPE) +; +; -- jsimd.h +; +%define JSIMD_NONE 0x00 +%define JSIMD_MMX 0x01 +%define JSIMD_3DNOW 0x02 +%define JSIMD_SSE 0x04 +%define JSIMD_SSE2 0x08 +%define JSIMD_AVX2 0x80 diff --git a/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc.h b/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc.h new file mode 100644 index 0000000..bf2a45a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdcfg.inc.h @@ -0,0 +1,133 @@ +/* + * This file generates the include file for the assembly + * implementations by abusing the C preprocessor. + * + * Note: Some things are manually defined as they need to + * be mapped to NASM types. + */ + +; +; Automatically generated include file from jsimdcfg.inc.h +; + +#define JPEG_INTERNALS + +#include "../jpeglib.h" +#include "../jconfig.h" +#include "../jmorecfg.h" +#include "jsimd.h" + +; +; -- jpeglib.h +; + +%define _cpp_protection_DCTSIZE DCTSIZE +%define _cpp_protection_DCTSIZE2 DCTSIZE2 + +; +; -- jmorecfg.h +; + +%define _cpp_protection_RGB_RED RGB_RED +%define _cpp_protection_RGB_GREEN RGB_GREEN +%define _cpp_protection_RGB_BLUE RGB_BLUE +%define _cpp_protection_RGB_PIXELSIZE RGB_PIXELSIZE + +%define _cpp_protection_EXT_RGB_RED EXT_RGB_RED +%define _cpp_protection_EXT_RGB_GREEN EXT_RGB_GREEN +%define _cpp_protection_EXT_RGB_BLUE EXT_RGB_BLUE +%define _cpp_protection_EXT_RGB_PIXELSIZE EXT_RGB_PIXELSIZE + +%define _cpp_protection_EXT_RGBX_RED EXT_RGBX_RED +%define _cpp_protection_EXT_RGBX_GREEN EXT_RGBX_GREEN +%define _cpp_protection_EXT_RGBX_BLUE EXT_RGBX_BLUE +%define _cpp_protection_EXT_RGBX_PIXELSIZE EXT_RGBX_PIXELSIZE + +%define _cpp_protection_EXT_BGR_RED EXT_BGR_RED +%define _cpp_protection_EXT_BGR_GREEN EXT_BGR_GREEN +%define _cpp_protection_EXT_BGR_BLUE EXT_BGR_BLUE +%define _cpp_protection_EXT_BGR_PIXELSIZE EXT_BGR_PIXELSIZE + +%define _cpp_protection_EXT_BGRX_RED EXT_BGRX_RED +%define _cpp_protection_EXT_BGRX_GREEN EXT_BGRX_GREEN +%define _cpp_protection_EXT_BGRX_BLUE EXT_BGRX_BLUE +%define _cpp_protection_EXT_BGRX_PIXELSIZE EXT_BGRX_PIXELSIZE + +%define _cpp_protection_EXT_XBGR_RED EXT_XBGR_RED +%define _cpp_protection_EXT_XBGR_GREEN EXT_XBGR_GREEN +%define _cpp_protection_EXT_XBGR_BLUE EXT_XBGR_BLUE +%define _cpp_protection_EXT_XBGR_PIXELSIZE EXT_XBGR_PIXELSIZE + +%define _cpp_protection_EXT_XRGB_RED EXT_XRGB_RED +%define _cpp_protection_EXT_XRGB_GREEN EXT_XRGB_GREEN +%define _cpp_protection_EXT_XRGB_BLUE EXT_XRGB_BLUE +%define _cpp_protection_EXT_XRGB_PIXELSIZE EXT_XRGB_PIXELSIZE + +%define RGBX_FILLER_0XFF 1 + +; Representation of a single sample (pixel element value). +; On this SIMD implementation, this must be 'unsigned char'. +; + +%define JSAMPLE byte ; unsigned char +%define SIZEOF_JSAMPLE SIZEOF_BYTE ; sizeof(JSAMPLE) + +%define _cpp_protection_CENTERJSAMPLE CENTERJSAMPLE + +; Representation of a DCT frequency coefficient. +; On this SIMD implementation, this must be 'short'. +; +%define JCOEF word ; short +%define SIZEOF_JCOEF SIZEOF_WORD ; sizeof(JCOEF) + +; Datatype used for image dimensions. +; On this SIMD implementation, this must be 'unsigned int'. +; +%define JDIMENSION dword ; unsigned int +%define SIZEOF_JDIMENSION SIZEOF_DWORD ; sizeof(JDIMENSION) + +%define JSAMPROW POINTER ; JSAMPLE * (jpeglib.h) +%define JSAMPARRAY POINTER ; JSAMPROW * (jpeglib.h) +%define JSAMPIMAGE POINTER ; JSAMPARRAY * (jpeglib.h) +%define JCOEFPTR POINTER ; JCOEF * (jpeglib.h) +%define SIZEOF_JSAMPROW SIZEOF_POINTER ; sizeof(JSAMPROW) +%define SIZEOF_JSAMPARRAY SIZEOF_POINTER ; sizeof(JSAMPARRAY) +%define SIZEOF_JSAMPIMAGE SIZEOF_POINTER ; sizeof(JSAMPIMAGE) +%define SIZEOF_JCOEFPTR SIZEOF_POINTER ; sizeof(JCOEFPTR) + +; +; -- jdct.h +; + +; A forward DCT routine is given a pointer to a work area of type DCTELEM[]; +; the DCT is to be performed in-place in that buffer. +; To maximize parallelism, Type DCTELEM is changed to short (originally, int). +; +%define DCTELEM word ; short +%define SIZEOF_DCTELEM SIZEOF_WORD ; sizeof(DCTELEM) + +%define FAST_FLOAT FP32 ; float +%define SIZEOF_FAST_FLOAT SIZEOF_FP32 ; sizeof(FAST_FLOAT) + +; To maximize parallelism, Type MULTIPLIER is changed to short. +; +%define ISLOW_MULT_TYPE word ; must be short +%define SIZEOF_ISLOW_MULT_TYPE SIZEOF_WORD ; sizeof(ISLOW_MULT_TYPE) + +%define IFAST_MULT_TYPE word ; must be short +%define SIZEOF_IFAST_MULT_TYPE SIZEOF_WORD ; sizeof(IFAST_MULT_TYPE) +%define IFAST_SCALE_BITS 2 ; fractional bits in scale factors + +%define FLOAT_MULT_TYPE FP32 ; must be float +%define SIZEOF_FLOAT_MULT_TYPE SIZEOF_FP32 ; sizeof(FLOAT_MULT_TYPE) + +; +; -- jsimd.h +; + +%define _cpp_protection_JSIMD_NONE JSIMD_NONE +%define _cpp_protection_JSIMD_MMX JSIMD_MMX +%define _cpp_protection_JSIMD_3DNOW JSIMD_3DNOW +%define _cpp_protection_JSIMD_SSE JSIMD_SSE +%define _cpp_protection_JSIMD_SSE2 JSIMD_SSE2 +%define _cpp_protection_JSIMD_AVX2 JSIMD_AVX2 diff --git a/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdext.inc b/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdext.inc new file mode 100644 index 0000000..b5341ed --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/nasm/jsimdext.inc @@ -0,0 +1,542 @@ +; +; jsimdext.inc - common declarations +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2010, 2016, 2018-2019, 2024, D. R. Commander. +; Copyright (C) 2018, Matthieu Darbois. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library - version 1.02 +; +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; +; This software is provided 'as-is', without any express or implied +; warranty. In no event will the authors be held liable for any damages +; arising from the use of this software. +; +; Permission is granted to anyone to use this software for any purpose, +; including commercial applications, and to alter it and redistribute it +; freely, subject to the following restrictions: +; +; 1. The origin of this software must not be misrepresented; you must not +; claim that you wrote the original software. If you use this software +; in a product, an acknowledgment in the product documentation would be +; appreciated but is not required. +; 2. Altered source versions must be plainly marked as such, and must not be +; misrepresented as being the original software. +; 3. This notice may not be removed or altered from any source distribution. + +; ========================================================================== +; System-dependent configurations + +%ifdef WIN32 ; ----(nasm -fwin32 -DWIN32 ...)-------- +; * Microsoft Visual C++ +; * MinGW (Minimalist GNU for Windows) +; * CygWin +; * LCC-Win32 + +; -- segment definition -- +; +%ifdef __YASM_VER__ +%define SEG_TEXT .text align=32 +%define SEG_CONST .rdata align=32 +%else +%define SEG_TEXT .text align=32 public use32 class=CODE +%define SEG_CONST .rdata align=32 public use32 class=CONST +%endif + +%elifdef WIN64 ; ----(nasm -fwin64 -DWIN64 ...)-------- +; * Microsoft Visual C++ + +; -- segment definition -- +; +%ifdef __YASM_VER__ +%define SEG_TEXT .text align=32 +%define SEG_CONST .rdata align=32 +%else +%define SEG_TEXT .text align=32 public use64 class=CODE +%define SEG_CONST .rdata align=32 public use64 class=CONST +%endif +%define EXTN(name) name ; foo() -> foo + +%elifdef OBJ32 ; ----(nasm -fobj -DOBJ32 ...)---------- +; * Borland C++ (Win32) + +; -- segment definition -- +; +%define SEG_TEXT _text align=32 public use32 class=CODE +%define SEG_CONST _data align=32 public use32 class=DATA + +%elifdef ELF ; ----(nasm -felf[64] -DELF ...)------------ +; * Linux +; * *BSD family Unix using elf format +; * Unix System V, including Solaris x86, UnixWare and SCO Unix + +; mark stack as non-executable +section .note.GNU-stack noalloc noexec nowrite progbits + +%ifdef __CET__ +%ifdef __x86_64__ +section .note.gnu.property note alloc noexec align=8 + dd 0x00000004, 0x00000010, 0x00000005, 0x00554e47 + dd 0xc0000002, 0x00000004, 0x00000003, 0x00000000 +%endif +%endif + +; -- segment definition -- +; +%ifdef __x86_64__ +%define SEG_TEXT .text progbits align=32 +%define SEG_CONST .rodata progbits align=32 +%else +%define SEG_TEXT .text progbits alloc exec nowrite align=32 +%define SEG_CONST .rodata progbits alloc noexec nowrite align=32 +%endif + +; To make the code position-independent, append -DPIC to the commandline +; +%define GOT_SYMBOL _GLOBAL_OFFSET_TABLE_ ; ELF supports PIC +%define EXTN(name) name ; foo() -> foo + +%elifdef AOUT ; ----(nasm -faoutb/aout -DAOUT ...)---- +; * Older Linux using a.out format (nasm -f aout -DAOUT ...) +; * *BSD family Unix using a.out format (nasm -f aoutb -DAOUT ...) + +; -- segment definition -- +; +%define SEG_TEXT .text +%define SEG_CONST .data + +; To make the code position-independent, append -DPIC to the commandline +; +%define GOT_SYMBOL __GLOBAL_OFFSET_TABLE_ ; BSD-style a.out supports PIC + +%elifdef MACHO ; ----(nasm -fmacho -DMACHO ...)-------- +; * NeXTstep/OpenStep/Rhapsody/Darwin/MacOS X (Mach-O format) + +; -- segment definition -- +; +%define SEG_TEXT .text ;align=32 ; nasm doesn't accept align=32. why? +%define SEG_CONST .rodata align=32 + +; The generation of position-independent code (PIC) is the default on Darwin. +; +%define PIC +%define GOT_SYMBOL _MACHO_PIC_ ; Mach-O style code-relative addressing + +%else ; ----(Other case)---------------------- + +; -- segment definition -- +; +%define SEG_TEXT .text +%define SEG_CONST .data + +%endif ; ---------------------------------------------- + +; ========================================================================== + +; -------------------------------------------------------------------------- +; Common types +; +%ifdef __x86_64__ +%ifnidn __OUTPUT_FORMAT__, elfx32 +%define POINTER qword ; general pointer type +%define SIZEOF_POINTER SIZEOF_QWORD ; sizeof(POINTER) +%define POINTER_BIT QWORD_BIT ; sizeof(POINTER)*BYTE_BIT +%define resp resq +%define dp dq +%define raxp rax +%define rbxp rbx +%define rcxp rcx +%define rdxp rdx +%define rsip rsi +%define rdip rdi +%define rbpp rbp +%define rspp rsp +%define r8p r8 +%define r9p r9 +%define r10p r10 +%define r11p r11 +%define r12p r12 +%define r13p r13 +%define r14p r14 +%define r15p r15 +%endif +%endif +%ifndef raxp +%define POINTER dword ; general pointer type +%define SIZEOF_POINTER SIZEOF_DWORD ; sizeof(POINTER) +%define POINTER_BIT DWORD_BIT ; sizeof(POINTER)*BYTE_BIT +%define resp resd +%define dp dd +; x86_64 ILP32 ABI (x32) +%define raxp eax +%define rbxp ebx +%define rcxp ecx +%define rdxp edx +%define rsip esi +%define rdip edi +%define rbpp ebp +%define rspp esp +%define r8p r8d +%define r9p r9d +%define r10p r10d +%define r11p r11d +%define r12p r12d +%define r13p r13d +%define r14p r14d +%define r15p r15d +%endif + +%define INT dword ; signed integer type +%define SIZEOF_INT SIZEOF_DWORD ; sizeof(INT) +%define INT_BIT DWORD_BIT ; sizeof(INT)*BYTE_BIT + +%define FP32 dword ; IEEE754 single +%define SIZEOF_FP32 SIZEOF_DWORD ; sizeof(FP32) +%define FP32_BIT DWORD_BIT ; sizeof(FP32)*BYTE_BIT + +%define MMWORD qword ; int64 (MMX register) +%define SIZEOF_MMWORD SIZEOF_QWORD ; sizeof(MMWORD) +%define MMWORD_BIT QWORD_BIT ; sizeof(MMWORD)*BYTE_BIT + +; NASM is buggy and doesn't properly handle operand sizes for SSE +; instructions, so for now we have to define XMMWORD as blank. +%define XMMWORD ; int128 (SSE register) +%define SIZEOF_XMMWORD SIZEOF_OWORD ; sizeof(XMMWORD) +%define XMMWORD_BIT OWORD_BIT ; sizeof(XMMWORD)*BYTE_BIT + +%define YMMWORD ; int256 (AVX register) +%define SIZEOF_YMMWORD SIZEOF_YWORD ; sizeof(YMMWORD) +%define YMMWORD_BIT YWORD_BIT ; sizeof(YMMWORD)*BYTE_BIT + +; Similar hacks for when we load a dword or MMWORD into an xmm# register +%define XMM_DWORD +%define XMM_MMWORD + +%define SIZEOF_BYTE 1 ; sizeof(byte) +%define SIZEOF_WORD 2 ; sizeof(word) +%define SIZEOF_DWORD 4 ; sizeof(dword) +%define SIZEOF_QWORD 8 ; sizeof(qword) +%define SIZEOF_OWORD 16 ; sizeof(oword) +%define SIZEOF_YWORD 32 ; sizeof(yword) + +%define BYTE_BIT 8 ; CHAR_BIT in C +%define WORD_BIT 16 ; sizeof(word)*BYTE_BIT +%define DWORD_BIT 32 ; sizeof(dword)*BYTE_BIT +%define QWORD_BIT 64 ; sizeof(qword)*BYTE_BIT +%define OWORD_BIT 128 ; sizeof(oword)*BYTE_BIT +%define YWORD_BIT 256 ; sizeof(yword)*BYTE_BIT + +; -------------------------------------------------------------------------- +; External Symbol Name +; +%ifndef EXTN +%define EXTN(name) _ %+ name ; foo() -> _foo +%endif + +; -------------------------------------------------------------------------- +; Hidden symbols +; +%ifdef ELF ; ----(nasm -felf[64] -DELF ...)-------- +%define GLOBAL_FUNCTION(name) global EXTN(name):function hidden +%define GLOBAL_DATA(name) global EXTN(name):data hidden +%elifdef MACHO ; ----(nasm -fmacho -DMACHO ...)-------- +%ifdef __YASM_VER__ +%define GLOBAL_FUNCTION(name) global EXTN(name):private_extern +%define GLOBAL_DATA(name) global EXTN(name):private_extern +%else +%if __NASM_VERSION_ID__ >= 0x020E0000 +%define GLOBAL_FUNCTION(name) global EXTN(name):private_extern +%define GLOBAL_DATA(name) global EXTN(name):private_extern +%endif +%endif +%endif + +%ifndef GLOBAL_FUNCTION +%define GLOBAL_FUNCTION(name) global EXTN(name) +%endif +%ifndef GLOBAL_DATA +%define GLOBAL_DATA(name) global EXTN(name) +%endif + +; -------------------------------------------------------------------------- +; Macros for position-independent code (PIC) support +; +%ifndef GOT_SYMBOL +%undef PIC +%endif + +%ifdef PIC ; ------------------------------------------- + +%ifidn GOT_SYMBOL, _MACHO_PIC_ ; -------------------- + +; At present, nasm doesn't seem to support PIC generation for Mach-O. +; The PIC support code below is a little tricky. + + SECTION SEG_CONST +const_base: + +%define GOTOFF(got, sym) (got) + (sym) - const_base + +%imacro GET_GOT 1 + ; NOTE: this macro destroys ecx resister. + call %%geteip + add ecx, byte (%%ref - $) + jmp short %%adjust +%%geteip: + mov ecx, POINTER [esp] + ret +%%adjust: + push ebp + xor ebp, ebp ; ebp = 0 +%ifidni %1, ebx ; (%1 == ebx) + ; db 0x8D,0x9C + jmp near const_base = + ; lea ebx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,9C,E9,(offset32) + db 0x8D, 0x9C ; 8D,9C + jmp near const_base ; E9,(const_base-%%ref) +%%ref: +%else ; (%1 != ebx) + ; db 0x8D,0x8C + jmp near const_base = + ; lea ecx, [ecx+ebp*8+(const_base-%%ref)] ; 8D,8C,E9,(offset32) + db 0x8D, 0x8C ; 8D,8C + jmp near const_base ; E9,(const_base-%%ref) +%%ref: + mov %1, ecx +%endif ; (%1 == ebx) + pop ebp +%endmacro + +%else ; GOT_SYMBOL != _MACHO_PIC_ ---------------- + +%define GOTOFF(got, sym) (got) + (sym) wrt ..gotoff + +%imacro GET_GOT 1 + extern GOT_SYMBOL + call %%geteip + add %1, GOT_SYMBOL + $$ - $ wrt ..gotpc + jmp short %%done +%%geteip: + mov %1, POINTER [esp] + ret +%%done: +%endmacro + +%endif ; GOT_SYMBOL == _MACHO_PIC_ ---------------- + +%imacro PUSHPIC 1.nolist + push %1 +%endmacro +%imacro POPPIC 1.nolist + pop %1 +%endmacro +%imacro MOVPIC 2.nolist + mov %1, %2 +%endmacro + +%else ; !PIC ----------------------------------------- + +%define GOTOFF(got, sym) (sym) + +%imacro GET_GOT 1.nolist +%endmacro +%imacro PUSHPIC 1.nolist +%endmacro +%imacro POPPIC 1.nolist +%endmacro +%imacro MOVPIC 2.nolist +%endmacro + +%endif ; PIC ----------------------------------------- + +; -------------------------------------------------------------------------- +; Align the next instruction on {2,4,8,16,..}-byte boundary. +; ".balign n,,m" in GNU as +; +%define MSKLE(x, y) (~(((y) & 0xFFFF) - ((x) & 0xFFFF)) >> 16) +%define FILLB(b, n) (($$-(b)) & ((n)-1)) + +%imacro ALIGNX 1-2.nolist 0xFFFF +%%bs: \ + times MSKLE(FILLB(%%bs, %1), %2) & MSKLE(16, FILLB($, %1)) & FILLB($, %1) \ + db 0x90 ; nop + times MSKLE(FILLB(%%bs, %1), %2) & FILLB($, %1) / 9 \ + db 0x8D, 0x9C, 0x23, 0x00, 0x00, 0x00, 0x00 ; lea ebx,[ebx+0x00000000] + times MSKLE(FILLB(%%bs, %1), %2) & FILLB($, %1) / 7 \ + db 0x8D, 0xAC, 0x25, 0x00, 0x00, 0x00, 0x00 ; lea ebp,[ebp+0x00000000] + times MSKLE(FILLB(%%bs, %1), %2) & FILLB($, %1) / 6 \ + db 0x8D, 0xAD, 0x00, 0x00, 0x00, 0x00 ; lea ebp,[ebp+0x00000000] + times MSKLE(FILLB(%%bs, %1), %2) & FILLB($, %1) / 4 \ + db 0x8D, 0x6C, 0x25, 0x00 ; lea ebp,[ebp+0x00] + times MSKLE(FILLB(%%bs, %1), %2) & FILLB($, %1) / 3 \ + db 0x8D, 0x6D, 0x00 ; lea ebp,[ebp+0x00] + times MSKLE(FILLB(%%bs, %1), %2) & FILLB($, %1) / 2 \ + db 0x8B, 0xED ; mov ebp,ebp + times MSKLE(FILLB(%%bs, %1), %2) & FILLB($, %1) / 1 \ + db 0x90 ; nop +%endmacro + +; Align the next data on {2,4,8,16,..}-byte boundary. +; +%imacro ALIGNZ 1.nolist + align %1, db 0 ; filling zeros +%endmacro + +%ifdef __x86_64__ + +%ifdef WIN64 + +%imacro COLLECT_ARGS 1 + sub rsp, SIZEOF_XMMWORD + movaps XMMWORD [rsp], xmm6 + sub rsp, SIZEOF_XMMWORD + movaps XMMWORD [rsp], xmm7 + mov r10, rcx +%if %1 > 1 + mov r11, rdx +%endif +%if %1 > 2 + push r12 + mov r12, r8 +%endif +%if %1 > 3 + push r13 + mov r13, r9 +%endif +%if %1 > 4 + push r14 + mov r14, [rbp+48] +%endif +%if %1 > 5 + push r15 + mov r15, [rbp+56] +%endif + push rsi + push rdi +%endmacro + +%imacro UNCOLLECT_ARGS 1 + pop rdi + pop rsi +%if %1 > 5 + pop r15 +%endif +%if %1 > 4 + pop r14 +%endif +%if %1 > 3 + pop r13 +%endif +%if %1 > 2 + pop r12 +%endif + movaps xmm7, XMMWORD [rsp] + add rsp, SIZEOF_XMMWORD + movaps xmm6, XMMWORD [rsp] + add rsp, SIZEOF_XMMWORD +%endmacro + +%imacro PUSH_XMM 1 + sub rsp, %1 * SIZEOF_XMMWORD + movaps XMMWORD [rsp+0*SIZEOF_XMMWORD], xmm8 +%if %1 > 1 + movaps XMMWORD [rsp+1*SIZEOF_XMMWORD], xmm9 +%endif +%if %1 > 2 + movaps XMMWORD [rsp+2*SIZEOF_XMMWORD], xmm10 +%endif +%if %1 > 3 + movaps XMMWORD [rsp+3*SIZEOF_XMMWORD], xmm11 +%endif +%endmacro + +%imacro POP_XMM 1 + movaps xmm8, XMMWORD [rsp+0*SIZEOF_XMMWORD] +%if %1 > 1 + movaps xmm9, XMMWORD [rsp+1*SIZEOF_XMMWORD] +%endif +%if %1 > 2 + movaps xmm10, XMMWORD [rsp+2*SIZEOF_XMMWORD] +%endif +%if %1 > 3 + movaps xmm11, XMMWORD [rsp+3*SIZEOF_XMMWORD] +%endif + add rsp, %1 * SIZEOF_XMMWORD +%endmacro + +%else + +%imacro COLLECT_ARGS 1 + push r10 + mov r10, rdi +%if %1 > 1 + push r11 + mov r11, rsi +%endif +%if %1 > 2 + push r12 + mov r12, rdx +%endif +%if %1 > 3 + push r13 + mov r13, rcx +%endif +%if %1 > 4 + push r14 + mov r14, r8 +%endif +%if %1 > 5 + push r15 + mov r15, r9 +%endif +%endmacro + +%imacro UNCOLLECT_ARGS 1 +%if %1 > 5 + pop r15 +%endif +%if %1 > 4 + pop r14 +%endif +%if %1 > 3 + pop r13 +%endif +%if %1 > 2 + pop r12 +%endif +%if %1 > 1 + pop r11 +%endif + pop r10 +%endmacro + +%imacro PUSH_XMM 1 +%endmacro + +%imacro POP_XMM 1 +%endmacro + +%endif + +%endif + +%ifdef __CET__ + +%imacro ENDBR64 0 + dd 0xfa1e0ff3 +%endmacro + +%else + +%imacro ENDBR64 0 +%endmacro + +%endif + +; -------------------------------------------------------------------------- +; Defines picked up from the C headers +; +%include "jsimdcfg.inc" + +; -------------------------------------------------------------------------- diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-avx2.asm new file mode 100644 index 0000000..aeeda0a --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-avx2.asm @@ -0,0 +1,557 @@ +; +; jccolext.asm - colorspace conversion (64-bit AVX2) +; +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_ycc_convert_avx2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +; r10d = JDIMENSION img_width +; r11 = JSAMPARRAY input_buf +; r12 = JSAMPIMAGE output_buf +; r13d = JDIMENSION output_row +; r14d = int num_rows + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_YMMWORD ; ymmword wk[WK_NUM] +%define WK_NUM 8 + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_ycc_convert_avx2) + +EXTN(jsimd_rgb_ycc_convert_avx2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, (SIZEOF_YMMWORD * WK_NUM) + COLLECT_ARGS 5 + push rbx + + mov ecx, r10d + test rcx, rcx + jz near .return + + push rcx + + mov rsi, r12 + mov ecx, r13d + mov rdip, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rsi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rsi+2*SIZEOF_JSAMPARRAY] + lea rdi, [rdi+rcx*SIZEOF_JSAMPROW] + lea rbx, [rbx+rcx*SIZEOF_JSAMPROW] + lea rdx, [rdx+rcx*SIZEOF_JSAMPROW] + + pop rcx + + mov rsi, r11 + mov eax, r14d + test rax, rax + jle near .return +.rowloop: + push rdx + push rbx + push rdi + push rsi + push rcx ; col + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr0 + mov rbxp, JSAMPROW [rbx] ; outptr1 + mov rdxp, JSAMPROW [rdx] ; outptr2 + + cmp rcx, byte SIZEOF_YMMWORD + jae near .columnloop + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push rax + push rdx + lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub rcx, byte SIZEOF_BYTE + movzx rax, byte [rsi+rcx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub rcx, byte SIZEOF_WORD + movzx rdx, word [rsi+rcx] + shl rax, WORD_BIT + or rax, rdx +.column_ld4: + vmovd xmmA, eax + pop rdx + pop rax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub rcx, byte SIZEOF_DWORD + vmovd xmmF, XMM_DWORD [rsi+rcx] + vpslldq xmmA, xmmA, SIZEOF_DWORD + vpor xmmA, xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub rcx, byte SIZEOF_MMWORD + vmovq xmmB, XMM_MMWORD [rsi+rcx] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + sub rcx, byte SIZEOF_XMMWORD + vmovdqu xmmB, XMM_MMWORD [rsi+rcx] + vperm2i128 ymmA, ymmA, ymmA, 1 + vpor ymmA, ymmB +.column_ld32: + test cl, SIZEOF_YMMWORD + jz short .column_ld64 + sub rcx, byte SIZEOF_YMMWORD + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] +.column_ld64: + test cl, 2*SIZEOF_YMMWORD + mov rcx, SIZEOF_YMMWORD + jz short .rgb_ycc_cnv + vmovdqa ymmB, ymmA + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + jmp short .rgb_ycc_cnv + +.columnloop: + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + vmovdqu ymmB, YMMWORD [rsi+2*SIZEOF_YMMWORD] + +.rgb_ycc_cnv: + ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vmovdqu ymmC, ymmA + vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + + vmovdqa ymmG, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12 + ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I) + vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I + ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A + ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q) + vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27 + ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N) + + vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D + ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T) + vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F + ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V) + + vmovdqa ymmD, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09 + ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P) + vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P + ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T) + vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B + ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R) + + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E + ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F + ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V) + + vmovdqa ymmE, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C + ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S) + vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S + ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D + ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T) + + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F + ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F + ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V) + + vpxor ymmH, ymmH, ymmH + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmB, ymmE + vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + + vmovdqa ymmF, ymmD + vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub rcx, byte SIZEOF_XMMWORD/16 + vmovd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub rcx, byte SIZEOF_XMMWORD/8 + vmovq xmmF, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmF +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub rcx, byte SIZEOF_XMMWORD/4 + vmovdqa xmmF, xmmA + vperm2i128 ymmF, ymmF, ymmF, 1 + vmovdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE] + vpor ymmA, ymmA, ymmF +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + jz short .column_ld16 + sub rcx, byte SIZEOF_XMMWORD/2 + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld16: + test cl, SIZEOF_XMMWORD + mov rcx, SIZEOF_YMMWORD + jz short .rgb_ycc_cnv + vmovdqa ymmE, ymmA + vmovdqa ymmH, ymmF + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + jmp short .rgb_ycc_cnv + +.columnloop: + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + vmovdqu ymmE, YMMWORD [rsi+2*SIZEOF_YMMWORD] + vmovdqu ymmH, YMMWORD [rsi+3*SIZEOF_YMMWORD] + +.rgb_ycc_cnv: + ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmB, ymmA + vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + + vmovdqa ymmB, ymmF + vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmD, ymmA + vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35 + ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L) + vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37 + ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N) + + vmovdqa ymmC, ymmF + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D + ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F + ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V) + + vmovdqa ymmB, ymmA + vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S) + vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D + ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T) + + vmovdqa ymmG, ymmD + vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E + ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U) + vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F + ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V) + + vmovdqa ymmE, ymmA + vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E + ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U) + + vmovdqa ymmH, ymmB + vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F + ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F + ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V) + + vpxor ymmF, ymmF, ymmF + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmD, ymmB + vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + + vmovdqa ymmG, ymmE + vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U) + + vpunpcklbw ymmF, ymmF, ymmH + vpunpckhbw ymmH, ymmH, ymmH + vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE + ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + + vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=RE + vmovdqa YMMWORD [wk(1)], ymm1 ; wk(1)=RO + vmovdqa YMMWORD [wk(2)], ymm4 ; wk(2)=BE + vmovdqa YMMWORD [wk(3)], ymm5 ; wk(3)=BO + + vmovdqa ymm6, ymm1 + vpunpcklwd ymm1, ymm1, ymm3 + vpunpckhwd ymm6, ymm6, ymm3 + vmovdqa ymm7, ymm1 + vmovdqa ymm4, ymm6 + vpmaddwd ymm1, ymm1, [rel PW_F0299_F0337] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337) + vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337) + vpmaddwd ymm7, ymm7, [rel PW_MF016_MF033] ; ymm7=ROL*-FIX(0.168)+GOL*-FIX(0.331) + vpmaddwd ymm4, ymm4, [rel PW_MF016_MF033] ; ymm4=ROH*-FIX(0.168)+GOH*-FIX(0.331) + + vmovdqa YMMWORD [wk(4)], ymm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337) + vmovdqa YMMWORD [wk(5)], ymm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337) + + vpxor ymm1, ymm1, ymm1 + vpxor ymm6, ymm6, ymm6 + vpunpcklwd ymm1, ymm1, ymm5 ; ymm1=BOL + vpunpckhwd ymm6, ymm6, ymm5 ; ymm6=BOH + vpsrld ymm1, ymm1, 1 ; ymm1=BOL*FIX(0.500) + vpsrld ymm6, ymm6, 1 ; ymm6=BOH*FIX(0.500) + + vmovdqa ymm5, [rel PD_ONEHALFM1_CJ] ; ymm5=[PD_ONEHALFM1_CJ] + + vpaddd ymm7, ymm7, ymm1 + vpaddd ymm4, ymm4, ymm6 + vpaddd ymm7, ymm7, ymm5 + vpaddd ymm4, ymm4, ymm5 + vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CbOL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbOH + vpackssdw ymm7, ymm7, ymm4 ; ymm7=CbO + + vmovdqa ymm1, YMMWORD [wk(2)] ; ymm1=BE + + vmovdqa ymm6, ymm0 + vpunpcklwd ymm0, ymm0, ymm2 + vpunpckhwd ymm6, ymm6, ymm2 + vmovdqa ymm5, ymm0 + vmovdqa ymm4, ymm6 + vpmaddwd ymm0, ymm0, [rel PW_F0299_F0337] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337) + vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337) + vpmaddwd ymm5, ymm5, [rel PW_MF016_MF033] ; ymm5=REL*-FIX(0.168)+GEL*-FIX(0.331) + vpmaddwd ymm4, ymm4, [rel PW_MF016_MF033] ; ymm4=REH*-FIX(0.168)+GEH*-FIX(0.331) + + vmovdqa YMMWORD [wk(6)], ymm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337) + vmovdqa YMMWORD [wk(7)], ymm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337) + + vpxor ymm0, ymm0, ymm0 + vpxor ymm6, ymm6, ymm6 + vpunpcklwd ymm0, ymm0, ymm1 ; ymm0=BEL + vpunpckhwd ymm6, ymm6, ymm1 ; ymm6=BEH + vpsrld ymm0, ymm0, 1 ; ymm0=BEL*FIX(0.500) + vpsrld ymm6, ymm6, 1 ; ymm6=BEH*FIX(0.500) + + vmovdqa ymm1, [rel PD_ONEHALFM1_CJ] ; ymm1=[PD_ONEHALFM1_CJ] + + vpaddd ymm5, ymm5, ymm0 + vpaddd ymm4, ymm4, ymm6 + vpaddd ymm5, ymm5, ymm1 + vpaddd ymm4, ymm4, ymm1 + vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CbEL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=CbEH + vpackssdw ymm5, ymm5, ymm4 ; ymm5=CbE + + vpsllw ymm7, ymm7, BYTE_BIT + vpor ymm5, ymm5, ymm7 ; ymm5=Cb + vmovdqu YMMWORD [rbx], ymm5 ; Save Cb + + vmovdqa ymm0, YMMWORD [wk(3)] ; ymm0=BO + vmovdqa ymm6, YMMWORD [wk(2)] ; ymm6=BE + vmovdqa ymm1, YMMWORD [wk(1)] ; ymm1=RO + + vmovdqa ymm4, ymm0 + vpunpcklwd ymm0, ymm0, ymm3 + vpunpckhwd ymm4, ymm4, ymm3 + vmovdqa ymm7, ymm0 + vmovdqa ymm5, ymm4 + vpmaddwd ymm0, ymm0, [rel PW_F0114_F0250] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250) + vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250) + vpmaddwd ymm7, ymm7, [rel PW_MF008_MF041] ; ymm7=BOL*-FIX(0.081)+GOL*-FIX(0.418) + vpmaddwd ymm5, ymm5, [rel PW_MF008_MF041] ; ymm5=BOH*-FIX(0.081)+GOH*-FIX(0.418) + + vmovdqa ymm3, [rel PD_ONEHALF] ; ymm3=[PD_ONEHALF] + + vpaddd ymm0, ymm0, YMMWORD [wk(4)] + vpaddd ymm4, ymm4, YMMWORD [wk(5)] + vpaddd ymm0, ymm0, ymm3 + vpaddd ymm4, ymm4, ymm3 + vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH + vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO + + vpxor ymm3, ymm3, ymm3 + vpxor ymm4, ymm4, ymm4 + vpunpcklwd ymm3, ymm3, ymm1 ; ymm3=ROL + vpunpckhwd ymm4, ymm4, ymm1 ; ymm4=ROH + vpsrld ymm3, ymm3, 1 ; ymm3=ROL*FIX(0.500) + vpsrld ymm4, ymm4, 1 ; ymm4=ROH*FIX(0.500) + + vmovdqa ymm1, [rel PD_ONEHALFM1_CJ] ; ymm1=[PD_ONEHALFM1_CJ] + + vpaddd ymm7, ymm7, ymm3 + vpaddd ymm5, ymm5, ymm4 + vpaddd ymm7, ymm7, ymm1 + vpaddd ymm5, ymm5, ymm1 + vpsrld ymm7, ymm7, SCALEBITS ; ymm7=CrOL + vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrOH + vpackssdw ymm7, ymm7, ymm5 ; ymm7=CrO + + vmovdqa ymm3, YMMWORD [wk(0)] ; ymm3=RE + + vmovdqa ymm4, ymm6 + vpunpcklwd ymm6, ymm6, ymm2 + vpunpckhwd ymm4, ymm4, ymm2 + vmovdqa ymm1, ymm6 + vmovdqa ymm5, ymm4 + vpmaddwd ymm6, ymm6, [rel PW_F0114_F0250] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250) + vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250) + vpmaddwd ymm1, ymm1, [rel PW_MF008_MF041] ; ymm1=BEL*-FIX(0.081)+GEL*-FIX(0.418) + vpmaddwd ymm5, ymm5, [rel PW_MF008_MF041] ; ymm5=BEH*-FIX(0.081)+GEH*-FIX(0.418) + + vmovdqa ymm2, [rel PD_ONEHALF] ; ymm2=[PD_ONEHALF] + + vpaddd ymm6, ymm6, YMMWORD [wk(6)] + vpaddd ymm4, ymm4, YMMWORD [wk(7)] + vpaddd ymm6, ymm6, ymm2 + vpaddd ymm4, ymm4, ymm2 + vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH + vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE + + vpsllw ymm0, ymm0, BYTE_BIT + vpor ymm6, ymm6, ymm0 ; ymm6=Y + vmovdqu YMMWORD [rdi], ymm6 ; Save Y + + vpxor ymm2, ymm2, ymm2 + vpxor ymm4, ymm4, ymm4 + vpunpcklwd ymm2, ymm2, ymm3 ; ymm2=REL + vpunpckhwd ymm4, ymm4, ymm3 ; ymm4=REH + vpsrld ymm2, ymm2, 1 ; ymm2=REL*FIX(0.500) + vpsrld ymm4, ymm4, 1 ; ymm4=REH*FIX(0.500) + + vmovdqa ymm0, [rel PD_ONEHALFM1_CJ] ; ymm0=[PD_ONEHALFM1_CJ] + + vpaddd ymm1, ymm1, ymm2 + vpaddd ymm5, ymm5, ymm4 + vpaddd ymm1, ymm1, ymm0 + vpaddd ymm5, ymm5, ymm0 + vpsrld ymm1, ymm1, SCALEBITS ; ymm1=CrEL + vpsrld ymm5, ymm5, SCALEBITS ; ymm5=CrEH + vpackssdw ymm1, ymm1, ymm5 ; ymm1=CrE + + vpsllw ymm7, ymm7, BYTE_BIT + vpor ymm1, ymm1, ymm7 ; ymm1=Cr + vmovdqu YMMWORD [rdx], ymm1 ; Save Cr + + sub rcx, byte SIZEOF_YMMWORD + add rsi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr + add rdi, byte SIZEOF_YMMWORD ; outptr0 + add rbx, byte SIZEOF_YMMWORD ; outptr1 + add rdx, byte SIZEOF_YMMWORD ; outptr2 + cmp rcx, byte SIZEOF_YMMWORD + jae near .columnloop + test rcx, rcx + jnz near .column_ld1 + + pop rcx ; col + pop rsi + pop rdi + pop rbx + pop rdx + + add rsi, byte SIZEOF_JSAMPROW ; input_buf + add rdi, byte SIZEOF_JSAMPROW + add rbx, byte SIZEOF_JSAMPROW + add rdx, byte SIZEOF_JSAMPROW + dec rax ; num_rows + jg near .rowloop + +.return: + pop rbx + vzeroupper + UNCOLLECT_ARGS 5 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-sse2.asm new file mode 100644 index 0000000..f3a1244 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolext-sse2.asm @@ -0,0 +1,482 @@ +; +; jccolext.asm - colorspace conversion (64-bit SSE2) +; +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_ycc_convert_sse2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +; r10d = JDIMENSION img_width +; r11 = JSAMPARRAY input_buf +; r12 = JSAMPIMAGE output_buf +; r13d = JDIMENSION output_row +; r14d = int num_rows + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 8 + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_ycc_convert_sse2) + +EXTN(jsimd_rgb_ycc_convert_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 5 + push rbx + + mov ecx, r10d + test rcx, rcx + jz near .return + + push rcx + + mov rsi, r12 + mov ecx, r13d + mov rdip, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rsi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rsi+2*SIZEOF_JSAMPARRAY] + lea rdi, [rdi+rcx*SIZEOF_JSAMPROW] + lea rbx, [rbx+rcx*SIZEOF_JSAMPROW] + lea rdx, [rdx+rcx*SIZEOF_JSAMPROW] + + pop rcx + + mov rsi, r11 + mov eax, r14d + test rax, rax + jle near .return +.rowloop: + push rdx + push rbx + push rdi + push rsi + push rcx ; col + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr0 + mov rbxp, JSAMPROW [rbx] ; outptr1 + mov rdxp, JSAMPROW [rdx] ; outptr2 + + cmp rcx, byte SIZEOF_XMMWORD + jae near .columnloop + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push rax + push rdx + lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub rcx, byte SIZEOF_BYTE + movzx rax, byte [rsi+rcx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub rcx, byte SIZEOF_WORD + movzx rdx, word [rsi+rcx] + shl rax, WORD_BIT + or rax, rdx +.column_ld4: + movd xmmA, eax + pop rdx + pop rax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub rcx, byte SIZEOF_DWORD + movd xmmF, XMM_DWORD [rsi+rcx] + pslldq xmmA, SIZEOF_DWORD + por xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub rcx, byte SIZEOF_MMWORD + movq xmmB, XMM_MMWORD [rsi+rcx] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + movdqa xmmF, xmmA + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + mov rcx, SIZEOF_XMMWORD + jmp short .rgb_ycc_cnv +.column_ld32: + test cl, 2*SIZEOF_XMMWORD + mov rcx, SIZEOF_XMMWORD + jz short .rgb_ycc_cnv + movdqa xmmB, xmmA + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD] + jmp short .rgb_ycc_cnv + +.columnloop: + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD] + movdqu xmmB, XMMWORD [rsi+2*SIZEOF_XMMWORD] + +.rgb_ycc_cnv: + ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + movdqa xmmG, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12) + psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A) + pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27) + + punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D) + punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F) + + movdqa xmmD, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09) + psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D) + pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B) + + punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E) + punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F) + + movdqa xmmE, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C) + psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D) + + punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F) + + pxor xmmH, xmmH + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmB, xmmE + punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F) + + movdqa xmmF, xmmD + punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F) + punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub rcx, byte SIZEOF_XMMWORD/16 + movd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub rcx, byte SIZEOF_XMMWORD/8 + movq xmmE, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmE +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub rcx, byte SIZEOF_XMMWORD/4 + movdqa xmmE, xmmA + movdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + mov rcx, SIZEOF_XMMWORD + jz short .rgb_ycc_cnv + movdqa xmmF, xmmA + movdqa xmmH, xmmE + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD] + jmp short .rgb_ycc_cnv + +.columnloop: + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [rsi+2*SIZEOF_XMMWORD] + movdqu xmmH, XMMWORD [rsi+3*SIZEOF_XMMWORD] + +.rgb_ycc_cnv: + ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35) + punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37) + + movdqa xmmC, xmmF + punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D) + punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F) + + movdqa xmmB, xmmA + punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C) + punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D) + + movdqa xmmG, xmmD + punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E) + punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F) + + movdqa xmmE, xmmA + punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E) + + movdqa xmmH, xmmB + punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F) + punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F) + + pxor xmmF, xmmF + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmD, xmmB + punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F) + + movdqa xmmG, xmmE + punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E) + + punpcklbw xmmF, xmmH + punpckhbw xmmH, xmmH + psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F) + psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE + ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + ; Cb = -0.16874 * R - 0.33126 * G + 0.50000 * B + CENTERJSAMPLE + ; Cr = 0.50000 * R - 0.41869 * G - 0.08131 * B + CENTERJSAMPLE + + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=RE + movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=RO + movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=BE + movdqa XMMWORD [wk(3)], xmm5 ; wk(3)=BO + + movdqa xmm6, xmm1 + punpcklwd xmm1, xmm3 + punpckhwd xmm6, xmm3 + movdqa xmm7, xmm1 + movdqa xmm4, xmm6 + pmaddwd xmm1, [rel PW_F0299_F0337] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337) + pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337) + pmaddwd xmm7, [rel PW_MF016_MF033] ; xmm7=ROL*-FIX(0.168)+GOL*-FIX(0.331) + pmaddwd xmm4, [rel PW_MF016_MF033] ; xmm4=ROH*-FIX(0.168)+GOH*-FIX(0.331) + + movdqa XMMWORD [wk(4)], xmm1 ; wk(4)=ROL*FIX(0.299)+GOL*FIX(0.337) + movdqa XMMWORD [wk(5)], xmm6 ; wk(5)=ROH*FIX(0.299)+GOH*FIX(0.337) + + pxor xmm1, xmm1 + pxor xmm6, xmm6 + punpcklwd xmm1, xmm5 ; xmm1=BOL + punpckhwd xmm6, xmm5 ; xmm6=BOH + psrld xmm1, 1 ; xmm1=BOL*FIX(0.500) + psrld xmm6, 1 ; xmm6=BOH*FIX(0.500) + + movdqa xmm5, [rel PD_ONEHALFM1_CJ] ; xmm5=[PD_ONEHALFM1_CJ] + + paddd xmm7, xmm1 + paddd xmm4, xmm6 + paddd xmm7, xmm5 + paddd xmm4, xmm5 + psrld xmm7, SCALEBITS ; xmm7=CbOL + psrld xmm4, SCALEBITS ; xmm4=CbOH + packssdw xmm7, xmm4 ; xmm7=CbO + + movdqa xmm1, XMMWORD [wk(2)] ; xmm1=BE + + movdqa xmm6, xmm0 + punpcklwd xmm0, xmm2 + punpckhwd xmm6, xmm2 + movdqa xmm5, xmm0 + movdqa xmm4, xmm6 + pmaddwd xmm0, [rel PW_F0299_F0337] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337) + pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337) + pmaddwd xmm5, [rel PW_MF016_MF033] ; xmm5=REL*-FIX(0.168)+GEL*-FIX(0.331) + pmaddwd xmm4, [rel PW_MF016_MF033] ; xmm4=REH*-FIX(0.168)+GEH*-FIX(0.331) + + movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=REL*FIX(0.299)+GEL*FIX(0.337) + movdqa XMMWORD [wk(7)], xmm6 ; wk(7)=REH*FIX(0.299)+GEH*FIX(0.337) + + pxor xmm0, xmm0 + pxor xmm6, xmm6 + punpcklwd xmm0, xmm1 ; xmm0=BEL + punpckhwd xmm6, xmm1 ; xmm6=BEH + psrld xmm0, 1 ; xmm0=BEL*FIX(0.500) + psrld xmm6, 1 ; xmm6=BEH*FIX(0.500) + + movdqa xmm1, [rel PD_ONEHALFM1_CJ] ; xmm1=[PD_ONEHALFM1_CJ] + + paddd xmm5, xmm0 + paddd xmm4, xmm6 + paddd xmm5, xmm1 + paddd xmm4, xmm1 + psrld xmm5, SCALEBITS ; xmm5=CbEL + psrld xmm4, SCALEBITS ; xmm4=CbEH + packssdw xmm5, xmm4 ; xmm5=CbE + + psllw xmm7, BYTE_BIT + por xmm5, xmm7 ; xmm5=Cb + movdqa XMMWORD [rbx], xmm5 ; Save Cb + + movdqa xmm0, XMMWORD [wk(3)] ; xmm0=BO + movdqa xmm6, XMMWORD [wk(2)] ; xmm6=BE + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=RO + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm3 + punpckhwd xmm4, xmm3 + movdqa xmm7, xmm0 + movdqa xmm5, xmm4 + pmaddwd xmm0, [rel PW_F0114_F0250] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250) + pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250) + pmaddwd xmm7, [rel PW_MF008_MF041] ; xmm7=BOL*-FIX(0.081)+GOL*-FIX(0.418) + pmaddwd xmm5, [rel PW_MF008_MF041] ; xmm5=BOH*-FIX(0.081)+GOH*-FIX(0.418) + + movdqa xmm3, [rel PD_ONEHALF] ; xmm3=[PD_ONEHALF] + + paddd xmm0, XMMWORD [wk(4)] + paddd xmm4, XMMWORD [wk(5)] + paddd xmm0, xmm3 + paddd xmm4, xmm3 + psrld xmm0, SCALEBITS ; xmm0=YOL + psrld xmm4, SCALEBITS ; xmm4=YOH + packssdw xmm0, xmm4 ; xmm0=YO + + pxor xmm3, xmm3 + pxor xmm4, xmm4 + punpcklwd xmm3, xmm1 ; xmm3=ROL + punpckhwd xmm4, xmm1 ; xmm4=ROH + psrld xmm3, 1 ; xmm3=ROL*FIX(0.500) + psrld xmm4, 1 ; xmm4=ROH*FIX(0.500) + + movdqa xmm1, [rel PD_ONEHALFM1_CJ] ; xmm1=[PD_ONEHALFM1_CJ] + + paddd xmm7, xmm3 + paddd xmm5, xmm4 + paddd xmm7, xmm1 + paddd xmm5, xmm1 + psrld xmm7, SCALEBITS ; xmm7=CrOL + psrld xmm5, SCALEBITS ; xmm5=CrOH + packssdw xmm7, xmm5 ; xmm7=CrO + + movdqa xmm3, XMMWORD [wk(0)] ; xmm3=RE + + movdqa xmm4, xmm6 + punpcklwd xmm6, xmm2 + punpckhwd xmm4, xmm2 + movdqa xmm1, xmm6 + movdqa xmm5, xmm4 + pmaddwd xmm6, [rel PW_F0114_F0250] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250) + pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250) + pmaddwd xmm1, [rel PW_MF008_MF041] ; xmm1=BEL*-FIX(0.081)+GEL*-FIX(0.418) + pmaddwd xmm5, [rel PW_MF008_MF041] ; xmm5=BEH*-FIX(0.081)+GEH*-FIX(0.418) + + movdqa xmm2, [rel PD_ONEHALF] ; xmm2=[PD_ONEHALF] + + paddd xmm6, XMMWORD [wk(6)] + paddd xmm4, XMMWORD [wk(7)] + paddd xmm6, xmm2 + paddd xmm4, xmm2 + psrld xmm6, SCALEBITS ; xmm6=YEL + psrld xmm4, SCALEBITS ; xmm4=YEH + packssdw xmm6, xmm4 ; xmm6=YE + + psllw xmm0, BYTE_BIT + por xmm6, xmm0 ; xmm6=Y + movdqa XMMWORD [rdi], xmm6 ; Save Y + + pxor xmm2, xmm2 + pxor xmm4, xmm4 + punpcklwd xmm2, xmm3 ; xmm2=REL + punpckhwd xmm4, xmm3 ; xmm4=REH + psrld xmm2, 1 ; xmm2=REL*FIX(0.500) + psrld xmm4, 1 ; xmm4=REH*FIX(0.500) + + movdqa xmm0, [rel PD_ONEHALFM1_CJ] ; xmm0=[PD_ONEHALFM1_CJ] + + paddd xmm1, xmm2 + paddd xmm5, xmm4 + paddd xmm1, xmm0 + paddd xmm5, xmm0 + psrld xmm1, SCALEBITS ; xmm1=CrEL + psrld xmm5, SCALEBITS ; xmm5=CrEH + packssdw xmm1, xmm5 ; xmm1=CrE + + psllw xmm7, BYTE_BIT + por xmm1, xmm7 ; xmm1=Cr + movdqa XMMWORD [rdx], xmm1 ; Save Cr + + sub rcx, byte SIZEOF_XMMWORD + add rsi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr + add rdi, byte SIZEOF_XMMWORD ; outptr0 + add rbx, byte SIZEOF_XMMWORD ; outptr1 + add rdx, byte SIZEOF_XMMWORD ; outptr2 + cmp rcx, byte SIZEOF_XMMWORD + jae near .columnloop + test rcx, rcx + jnz near .column_ld1 + + pop rcx ; col + pop rsi + pop rdi + pop rbx + pop rdx + + add rsi, byte SIZEOF_JSAMPROW ; input_buf + add rdi, byte SIZEOF_JSAMPROW + add rbx, byte SIZEOF_JSAMPROW + add rdx, byte SIZEOF_JSAMPROW + dec rax ; num_rows + jg near .rowloop + +.return: + pop rbx + UNCOLLECT_ARGS 5 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-avx2.asm new file mode 100644 index 0000000..e262891 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-avx2.asm @@ -0,0 +1,117 @@ +; +; jccolor.asm - colorspace conversion (64-bit AVX2) +; +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_081 equ 5329 ; FIX(0.08131) +F_0_114 equ 7471 ; FIX(0.11400) +F_0_168 equ 11059 ; FIX(0.16874) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_331 equ 21709 ; FIX(0.33126) +F_0_418 equ 27439 ; FIX(0.41869) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_ycc_convert_avx2) + +EXTN(jconst_rgb_ycc_convert_avx2): + +PW_F0299_F0337 times 8 dw F_0_299, F_0_337 +PW_F0114_F0250 times 8 dw F_0_114, F_0_250 +PW_MF016_MF033 times 8 dw -F_0_168, -F_0_331 +PW_MF008_MF041 times 8 dw -F_0_081, -F_0_418 +PD_ONEHALFM1_CJ times 8 dd (1 << (SCALEBITS - 1)) - 1 + \ + (CENTERJSAMPLE << SCALEBITS) +PD_ONEHALF times 8 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgb_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extrgbx_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgr_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extbgrx_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extxbgr_ycc_convert_avx2 +%include "jccolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_avx2 jsimd_extxrgb_ycc_convert_avx2 +%include "jccolext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-sse2.asm new file mode 100644 index 0000000..cc9edb4 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jccolor-sse2.asm @@ -0,0 +1,116 @@ +; +; jccolor.asm - colorspace conversion (64-bit SSE2) +; +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_081 equ 5329 ; FIX(0.08131) +F_0_114 equ 7471 ; FIX(0.11400) +F_0_168 equ 11059 ; FIX(0.16874) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_331 equ 21709 ; FIX(0.33126) +F_0_418 equ 27439 ; FIX(0.41869) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_ycc_convert_sse2) + +EXTN(jconst_rgb_ycc_convert_sse2): + +PW_F0299_F0337 times 4 dw F_0_299, F_0_337 +PW_F0114_F0250 times 4 dw F_0_114, F_0_250 +PW_MF016_MF033 times 4 dw -F_0_168, -F_0_331 +PW_MF008_MF041 times 4 dw -F_0_081, -F_0_418 +PD_ONEHALFM1_CJ times 4 dd (1 << (SCALEBITS - 1)) - 1 + \ + (CENTERJSAMPLE << SCALEBITS) +PD_ONEHALF times 4 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgb_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extrgbx_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgr_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extbgrx_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extxbgr_ycc_convert_sse2 +%include "jccolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_ycc_convert_sse2 jsimd_extxrgb_ycc_convert_sse2 +%include "jccolext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-avx2.asm new file mode 100644 index 0000000..267ec51 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-avx2.asm @@ -0,0 +1,109 @@ +; +; jcgray.asm - grayscale colorspace conversion (64-bit AVX2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_114 equ 7471 ; FIX(0.11400) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_gray_convert_avx2) + +EXTN(jconst_rgb_gray_convert_avx2): + +PW_F0299_F0337 times 8 dw F_0_299, F_0_337 +PW_F0114_F0250 times 8 dw F_0_114, F_0_250 +PD_ONEHALF times 8 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extrgb_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extrgbx_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extbgr_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extbgrx_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extxbgr_gray_convert_avx2 +%include "jcgryext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_gray_convert_avx2 jsimd_extxrgb_gray_convert_avx2 +%include "jcgryext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-sse2.asm new file mode 100644 index 0000000..4b94d7b --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgray-sse2.asm @@ -0,0 +1,108 @@ +; +; jcgray.asm - grayscale colorspace conversion (64-bit SSE2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_114 equ 7471 ; FIX(0.11400) +F_0_250 equ 16384 ; FIX(0.25000) +F_0_299 equ 19595 ; FIX(0.29900) +F_0_587 equ 38470 ; FIX(0.58700) +F_0_337 equ (F_0_587 - F_0_250) ; FIX(0.58700) - FIX(0.25000) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_rgb_gray_convert_sse2) + +EXTN(jconst_rgb_gray_convert_sse2): + +PW_F0299_F0337 times 4 dw F_0_299, F_0_337 +PW_F0114_F0250 times 4 dw F_0_114, F_0_250 +PD_ONEHALF times 4 dd (1 << (SCALEBITS - 1)) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extrgb_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extrgbx_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extbgr_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extbgrx_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extxbgr_gray_convert_sse2 +%include "jcgryext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_rgb_gray_convert_sse2 jsimd_extxrgb_gray_convert_sse2 +%include "jcgryext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-avx2.asm new file mode 100644 index 0000000..77e85f7 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-avx2.asm @@ -0,0 +1,436 @@ +; +; jcgryext.asm - grayscale colorspace conversion (64-bit AVX2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_gray_convert_avx2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +; r10d = JDIMENSION img_width +; r11 = JSAMPARRAY input_buf +; r12 = JSAMPIMAGE output_buf +; r13d = JDIMENSION output_row +; r14d = int num_rows + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_YMMWORD ; ymmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_gray_convert_avx2) + +EXTN(jsimd_rgb_gray_convert_avx2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_YMMWORD * WK_NUM) + COLLECT_ARGS 5 + push rbx + + mov ecx, r10d + test rcx, rcx + jz near .return + + push rcx + + mov rsi, r12 + mov ecx, r13d + mov rdip, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY] + lea rdi, [rdi+rcx*SIZEOF_JSAMPROW] + + pop rcx + + mov rsi, r11 + mov eax, r14d + test rax, rax + jle near .return +.rowloop: + push rdi + push rsi + push rcx ; col + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr0 + + cmp rcx, byte SIZEOF_YMMWORD + jae near .columnloop + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push rax + push rdx + lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub rcx, byte SIZEOF_BYTE + movzx rax, byte [rsi+rcx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub rcx, byte SIZEOF_WORD + movzx rdx, word [rsi+rcx] + shl rax, WORD_BIT + or rax, rdx +.column_ld4: + vmovd xmmA, eax + pop rdx + pop rax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub rcx, byte SIZEOF_DWORD + vmovd xmmF, XMM_DWORD [rsi+rcx] + vpslldq xmmA, xmmA, SIZEOF_DWORD + vpor xmmA, xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub rcx, byte SIZEOF_MMWORD + vmovq xmmB, XMM_MMWORD [rsi+rcx] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + sub rcx, byte SIZEOF_XMMWORD + vmovdqu xmmB, XMM_MMWORD [rsi+rcx] + vperm2i128 ymmA, ymmA, ymmA, 1 + vpor ymmA, ymmB +.column_ld32: + test cl, SIZEOF_YMMWORD + jz short .column_ld64 + sub rcx, byte SIZEOF_YMMWORD + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] +.column_ld64: + test cl, 2*SIZEOF_YMMWORD + mov rcx, SIZEOF_YMMWORD + jz short .rgb_gray_cnv + vmovdqa ymmB, ymmA + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + jmp short .rgb_gray_cnv + +.columnloop: + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + vmovdqu ymmB, YMMWORD [rsi+2*SIZEOF_YMMWORD] + +.rgb_gray_cnv: + ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + ; ymmB=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vmovdqu ymmC, ymmA + vinserti128 ymmA, ymmF, xmmA, 0 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vinserti128 ymmC, ymmC, xmmB, 0 ; ymmC=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vinserti128 ymmB, ymmB, xmmF, 0 ; ymmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + vperm2i128 ymmF, ymmC, ymmC, 1 ; ymmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + + vmovdqa ymmG, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12 + ; 22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I) + vpsrldq ymmG, ymmG, 8 ; ymmG=(22 03 13 23 04 14 24 05 0G 1G 2G 0H 1H 2H 0I 1I + ; 2I 0J 1J 2J 0K 1K 2K 0L -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmF ; ymmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A + ; 0G 0O 1G 1O 2G 2O 0H 0P 1H 1P 2H 2P 0I 0Q 1I 1Q) + vpslldq ymmF, ymmF, 8 ; ymmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27 + ; 08 18 28 09 19 29 0A 1A 1L 2L 0M 1M 2M 0N 1N 2N) + + vpunpcklbw ymmG, ymmG, ymmB ; ymmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D + ; 2I 2Q 0J 0R 1J 1R 2J 2R 0K 0S 1K 1S 2K 2S 0L 0T) + vpunpckhbw ymmF, ymmF, ymmB ; ymmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F + ; 1L 1T 2L 2T 0M 0U 1M 1U 2M 2U 0N 0V 1N 1V 2N 2V) + + vmovdqa ymmD, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09 + ; 11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P) + vpsrldq ymmD, ymmD, 8 ; ymmD=(11 19 21 29 02 0A 12 1A 0G 0O 1G 1O 2G 2O 0H 0P + ; 1H 1P 2H 2P 0I 0Q 1I 1Q -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmG ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 0H 0L 0P 0T) + vpslldq ymmG, ymmG, 8 ; ymmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B + ; 04 0C 14 1C 24 2C 05 0D 2I 2Q 0J 0R 1J 1R 2J 2R) + + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E + ; 1H 1L 1P 1T 2H 2L 2P 2T 0I 0M 0Q 0U 1I 1M 1Q 1U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F + ; 2I 2M 2Q 2U 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V) + + vmovdqa ymmE, ymmA + vpslldq ymmA, ymmA, 8 ; ymmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C + ; 20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S) + vpsrldq ymmE, ymmE, 8 ; ymmE=(20 24 28 2C 01 05 09 0D 0G 0K 0O 0S 1G 1K 1O 1S + ; 2G 2K 2O 2S 0H 0L 0P 0T -- -- -- -- -- -- -- --) + + vpunpckhbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpslldq ymmD, ymmD, 8 ; ymmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D + ; 02 06 0A 0E 12 16 1A 1E 1H 1L 1P 1T 2H 2L 2P 2T) + + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F + ; 2G 2I 2K 2M 2O 2Q 2S 2U 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmG ; ymmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F + ; 1H 1J 1L 1N 1P 1R 1T 1V 2H 2J 2L 2N 2P 2R 2T 2V) + + vpxor ymmH, ymmH, ymmH + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmH ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmB, ymmE + vpunpcklbw ymmE, ymmE, ymmH ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmB, ymmB, ymmH ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + + vmovdqa ymmF, ymmD + vpunpcklbw ymmD, ymmD, ymmH ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmF, ymmF, ymmH ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub rcx, byte SIZEOF_XMMWORD/16 + vmovd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub rcx, byte SIZEOF_XMMWORD/8 + vmovq xmmF, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE] + vpslldq xmmA, xmmA, SIZEOF_MMWORD + vpor xmmA, xmmA, xmmF +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub rcx, byte SIZEOF_XMMWORD/4 + vmovdqa xmmF, xmmA + vperm2i128 ymmF, ymmF, ymmF, 1 + vmovdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE] + vpor ymmA, ymmA, ymmF +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + jz short .column_ld16 + sub rcx, byte SIZEOF_XMMWORD/2 + vmovdqa ymmF, ymmA + vmovdqu ymmA, YMMWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld16: + test cl, SIZEOF_XMMWORD + mov rcx, SIZEOF_YMMWORD + jz short .rgb_gray_cnv + vmovdqa ymmE, ymmA + vmovdqa ymmH, ymmF + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + jmp short .rgb_gray_cnv + +.columnloop: + vmovdqu ymmA, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymmF, YMMWORD [rsi+1*SIZEOF_YMMWORD] + vmovdqu ymmE, YMMWORD [rsi+2*SIZEOF_YMMWORD] + vmovdqu ymmH, YMMWORD [rsi+3*SIZEOF_YMMWORD] + +.rgb_gray_cnv: + ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + ; ymmE=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmB, ymmA + vinserti128 ymmA, ymmA, xmmE, 1 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vperm2i128 ymmE, ymmB, ymmE, 0x31 ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + + vmovdqa ymmB, ymmF + vinserti128 ymmF, ymmF, xmmH, 1 ; ymmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + vperm2i128 ymmH, ymmB, ymmH, 0x31 ; ymmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + vmovdqa ymmD, ymmA + vpunpcklbw ymmA, ymmA, ymmE ; ymmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35 + ; 0G 0K 1G 1K 2G 2K 3G 3K 0H 0L 1H 1L 2H 2L 3H 3L) + vpunpckhbw ymmD, ymmD, ymmE ; ymmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37 + ; 0I 0M 1I 1M 2I 2M 3I 3M 0J 0N 1J 1N 2J 2N 3J 3N) + + vmovdqa ymmC, ymmF + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D + ; 0O 0S 1O 1S 2O 2S 3O 3S 0P 0T 1P 1T 2P 2T 3P 3T) + vpunpckhbw ymmC, ymmC, ymmH ; ymmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F + ; 0Q 0U 1Q 1U 2Q 2U 3Q 3U 0R 0V 1R 1V 2R 2V 3R 3V) + + vmovdqa ymmB, ymmA + vpunpcklwd ymmA, ymmA, ymmF ; ymmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C + ; 0G 0K 0O 0S 1G 1K 1O 1S 2G 2K 2O 2S 3G 3K 3O 3S) + vpunpckhwd ymmB, ymmB, ymmF ; ymmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D + ; 0H 0L 0P 0T 1H 1L 1P 1T 2H 2L 2P 2T 3H 3L 3P 3T) + + vmovdqa ymmG, ymmD + vpunpcklwd ymmD, ymmD, ymmC ; ymmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E + ; 0I 0M 0Q 0U 1I 1M 1Q 1U 2I 2M 2Q 2U 3I 3M 3Q 3U) + vpunpckhwd ymmG, ymmG, ymmC ; ymmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F + ; 0J 0N 0R 0V 1J 1N 1R 1V 2J 2N 2R 2V 3J 3N 3R 3V) + + vmovdqa ymmE, ymmA + vpunpcklbw ymmA, ymmA, ymmD ; ymmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E + ; 0G 0I 0K 0M 0O 0Q 0S 0U 1G 1I 1K 1M 1O 1Q 1S 1U) + vpunpckhbw ymmE, ymmE, ymmD ; ymmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E + ; 2G 2I 2K 2M 2O 2Q 2S 2U 3G 3I 3K 3M 3O 3Q 3S 3U) + + vmovdqa ymmH, ymmB + vpunpcklbw ymmB, ymmB, ymmG ; ymmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F + ; 0H 0J 0L 0N 0P 0R 0T 0V 1H 1J 1L 1N 1P 1R 1T 1V) + vpunpckhbw ymmH, ymmH, ymmG ; ymmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F + ; 2H 2J 2L 2N 2P 2R 2T 2V 3H 3J 3L 3N 3P 3R 3T 3V) + + vpxor ymmF, ymmF, ymmF + + vmovdqa ymmC, ymmA + vpunpcklbw ymmA, ymmA, ymmF ; ymmA=(00 02 04 06 08 0A 0C 0E 0G 0I 0K 0M 0O 0Q 0S 0U) + vpunpckhbw ymmC, ymmC, ymmF ; ymmC=(10 12 14 16 18 1A 1C 1E 1G 1I 1K 1M 1O 1Q 1S 1U) + + vmovdqa ymmD, ymmB + vpunpcklbw ymmB, ymmB, ymmF ; ymmB=(01 03 05 07 09 0B 0D 0F 0H 0J 0L 0N 0P 0R 0T 0V) + vpunpckhbw ymmD, ymmD, ymmF ; ymmD=(11 13 15 17 19 1B 1D 1F 1H 1J 1L 1N 1P 1R 1T 1V) + + vmovdqa ymmG, ymmE + vpunpcklbw ymmE, ymmE, ymmF ; ymmE=(20 22 24 26 28 2A 2C 2E 2G 2I 2K 2M 2O 2Q 2S 2U) + vpunpckhbw ymmG, ymmG, ymmF ; ymmG=(30 32 34 36 38 3A 3C 3E 3G 3I 3K 3M 3O 3Q 3S 3U) + + vpunpcklbw ymmF, ymmF, ymmH + vpunpckhbw ymmH, ymmH, ymmH + vpsrlw ymmF, ymmF, BYTE_BIT ; ymmF=(21 23 25 27 29 2B 2D 2F 2H 2J 2L 2N 2P 2R 2T 2V) + vpsrlw ymmH, ymmH, BYTE_BIT ; ymmH=(31 33 35 37 39 3B 3D 3F 3H 3J 3L 3N 3P 3R 3T 3V) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; ymm0=R(02468ACEGIKMOQSU)=RE, ymm2=G(02468ACEGIKMOQSU)=GE, ymm4=B(02468ACEGIKMOQSU)=BE + ; ymm1=R(13579BDFHJLNPRTV)=RO, ymm3=G(13579BDFHJLNPRTV)=GO, ymm5=B(13579BDFHJLNPRTV)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + + vmovdqa ymm6, ymm1 + vpunpcklwd ymm1, ymm1, ymm3 + vpunpckhwd ymm6, ymm6, ymm3 + vpmaddwd ymm1, ymm1, [rel PW_F0299_F0337] ; ymm1=ROL*FIX(0.299)+GOL*FIX(0.337) + vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=ROH*FIX(0.299)+GOH*FIX(0.337) + + vmovdqa ymm7, ymm6 ; ymm7=ROH*FIX(0.299)+GOH*FIX(0.337) + + vmovdqa ymm6, ymm0 + vpunpcklwd ymm0, ymm0, ymm2 + vpunpckhwd ymm6, ymm6, ymm2 + vpmaddwd ymm0, ymm0, [rel PW_F0299_F0337] ; ymm0=REL*FIX(0.299)+GEL*FIX(0.337) + vpmaddwd ymm6, ymm6, [rel PW_F0299_F0337] ; ymm6=REH*FIX(0.299)+GEH*FIX(0.337) + + vmovdqa YMMWORD [wk(0)], ymm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337) + vmovdqa YMMWORD [wk(1)], ymm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337) + + vmovdqa ymm0, ymm5 ; ymm0=BO + vmovdqa ymm6, ymm4 ; ymm6=BE + + vmovdqa ymm4, ymm0 + vpunpcklwd ymm0, ymm0, ymm3 + vpunpckhwd ymm4, ymm4, ymm3 + vpmaddwd ymm0, ymm0, [rel PW_F0114_F0250] ; ymm0=BOL*FIX(0.114)+GOL*FIX(0.250) + vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BOH*FIX(0.114)+GOH*FIX(0.250) + + vmovdqa ymm3, [rel PD_ONEHALF] ; ymm3=[PD_ONEHALF] + + vpaddd ymm0, ymm0, ymm1 + vpaddd ymm4, ymm4, ymm7 + vpaddd ymm0, ymm0, ymm3 + vpaddd ymm4, ymm4, ymm3 + vpsrld ymm0, ymm0, SCALEBITS ; ymm0=YOL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YOH + vpackssdw ymm0, ymm0, ymm4 ; ymm0=YO + + vmovdqa ymm4, ymm6 + vpunpcklwd ymm6, ymm6, ymm2 + vpunpckhwd ymm4, ymm4, ymm2 + vpmaddwd ymm6, ymm6, [rel PW_F0114_F0250] ; ymm6=BEL*FIX(0.114)+GEL*FIX(0.250) + vpmaddwd ymm4, ymm4, [rel PW_F0114_F0250] ; ymm4=BEH*FIX(0.114)+GEH*FIX(0.250) + + vmovdqa ymm2, [rel PD_ONEHALF] ; ymm2=[PD_ONEHALF] + + vpaddd ymm6, ymm6, YMMWORD [wk(0)] + vpaddd ymm4, ymm4, YMMWORD [wk(1)] + vpaddd ymm6, ymm6, ymm2 + vpaddd ymm4, ymm4, ymm2 + vpsrld ymm6, ymm6, SCALEBITS ; ymm6=YEL + vpsrld ymm4, ymm4, SCALEBITS ; ymm4=YEH + vpackssdw ymm6, ymm6, ymm4 ; ymm6=YE + + vpsllw ymm0, ymm0, BYTE_BIT + vpor ymm6, ymm6, ymm0 ; ymm6=Y + vmovdqu YMMWORD [rdi], ymm6 ; Save Y + + sub rcx, byte SIZEOF_YMMWORD + add rsi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; inptr + add rdi, byte SIZEOF_YMMWORD ; outptr0 + cmp rcx, byte SIZEOF_YMMWORD + jae near .columnloop + test rcx, rcx + jnz near .column_ld1 + + pop rcx ; col + pop rsi + pop rdi + + add rsi, byte SIZEOF_JSAMPROW ; input_buf + add rdi, byte SIZEOF_JSAMPROW + dec rax ; num_rows + jg near .rowloop + +.return: + pop rbx + vzeroupper + UNCOLLECT_ARGS 5 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-sse2.asm new file mode 100644 index 0000000..3e8087c --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcgryext-sse2.asm @@ -0,0 +1,361 @@ +; +; jcgryext.asm - grayscale colorspace conversion (64-bit SSE2) +; +; Copyright (C) 2011, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_rgb_gray_convert_sse2(JDIMENSION img_width, JSAMPARRAY input_buf, +; JSAMPIMAGE output_buf, JDIMENSION output_row, +; int num_rows); +; + +; r10d = JDIMENSION img_width +; r11 = JSAMPARRAY input_buf +; r12 = JSAMPIMAGE output_buf +; r13d = JDIMENSION output_row +; r14d = int num_rows + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_rgb_gray_convert_sse2) + +EXTN(jsimd_rgb_gray_convert_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 5 + push rbx + + mov ecx, r10d + test rcx, rcx + jz near .return + + push rcx + + mov rsi, r12 + mov ecx, r13d + mov rdip, JSAMPARRAY [rsi+0*SIZEOF_JSAMPARRAY] + lea rdi, [rdi+rcx*SIZEOF_JSAMPROW] + + pop rcx + + mov rsi, r11 + mov eax, r14d + test rax, rax + jle near .return +.rowloop: + push rdi + push rsi + push rcx ; col + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr0 + + cmp rcx, byte SIZEOF_XMMWORD + jae near .columnloop + +%if RGB_PIXELSIZE == 3 ; --------------- + +.column_ld1: + push rax + push rdx + lea rcx, [rcx+rcx*2] ; imul ecx,RGB_PIXELSIZE + test cl, SIZEOF_BYTE + jz short .column_ld2 + sub rcx, byte SIZEOF_BYTE + movzx rax, byte [rsi+rcx] +.column_ld2: + test cl, SIZEOF_WORD + jz short .column_ld4 + sub rcx, byte SIZEOF_WORD + movzx rdx, word [rsi+rcx] + shl rax, WORD_BIT + or rax, rdx +.column_ld4: + movd xmmA, eax + pop rdx + pop rax + test cl, SIZEOF_DWORD + jz short .column_ld8 + sub rcx, byte SIZEOF_DWORD + movd xmmF, XMM_DWORD [rsi+rcx] + pslldq xmmA, SIZEOF_DWORD + por xmmA, xmmF +.column_ld8: + test cl, SIZEOF_MMWORD + jz short .column_ld16 + sub rcx, byte SIZEOF_MMWORD + movq xmmB, XMM_MMWORD [rsi+rcx] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmB +.column_ld16: + test cl, SIZEOF_XMMWORD + jz short .column_ld32 + movdqa xmmF, xmmA + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + mov rcx, SIZEOF_XMMWORD + jmp short .rgb_gray_cnv +.column_ld32: + test cl, 2*SIZEOF_XMMWORD + mov rcx, SIZEOF_XMMWORD + jz short .rgb_gray_cnv + movdqa xmmB, xmmA + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD] + jmp short .rgb_gray_cnv + +.columnloop: + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [rsi+1*SIZEOF_XMMWORD] + movdqu xmmB, XMMWORD [rsi+2*SIZEOF_XMMWORD] + +.rgb_gray_cnv: + ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + ; xmmF=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + ; xmmB=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + movdqa xmmG, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 10 20 01 11 21 02 12) + psrldq xmmG, 8 ; xmmG=(22 03 13 23 04 14 24 05 -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmF ; xmmA=(00 08 10 18 20 28 01 09 11 19 21 29 02 0A 12 1A) + pslldq xmmF, 8 ; xmmF=(-- -- -- -- -- -- -- -- 15 25 06 16 26 07 17 27) + + punpcklbw xmmG, xmmB ; xmmG=(22 2A 03 0B 13 1B 23 2B 04 0C 14 1C 24 2C 05 0D) + punpckhbw xmmF, xmmB ; xmmF=(15 1D 25 2D 06 0E 16 1E 26 2E 07 0F 17 1F 27 2F) + + movdqa xmmD, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 08 10 18 20 28 01 09) + psrldq xmmD, 8 ; xmmD=(11 19 21 29 02 0A 12 1A -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmG ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 01 05 09 0D) + pslldq xmmG, 8 ; xmmG=(-- -- -- -- -- -- -- -- 22 2A 03 0B 13 1B 23 2B) + + punpcklbw xmmD, xmmF ; xmmD=(11 15 19 1D 21 25 29 2D 02 06 0A 0E 12 16 1A 1E) + punpckhbw xmmG, xmmF ; xmmG=(22 26 2A 2E 03 07 0B 0F 13 17 1B 1F 23 27 2B 2F) + + movdqa xmmE, xmmA + pslldq xmmA, 8 ; xmmA=(-- -- -- -- -- -- -- -- 00 04 08 0C 10 14 18 1C) + psrldq xmmE, 8 ; xmmE=(20 24 28 2C 01 05 09 0D -- -- -- -- -- -- -- --) + + punpckhbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + pslldq xmmD, 8 ; xmmD=(-- -- -- -- -- -- -- -- 11 15 19 1D 21 25 29 2D) + + punpcklbw xmmE, xmmG ; xmmE=(20 22 24 26 28 2A 2C 2E 01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmG ; xmmD=(11 13 15 17 19 1B 1D 1F 21 23 25 27 29 2B 2D 2F) + + pxor xmmH, xmmH + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmH ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmH ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmB, xmmE + punpcklbw xmmE, xmmH ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmB, xmmH ; xmmB=(01 03 05 07 09 0B 0D 0F) + + movdqa xmmF, xmmD + punpcklbw xmmD, xmmH ; xmmD=(11 13 15 17 19 1B 1D 1F) + punpckhbw xmmF, xmmH ; xmmF=(21 23 25 27 29 2B 2D 2F) + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +.column_ld1: + test cl, SIZEOF_XMMWORD/16 + jz short .column_ld2 + sub rcx, byte SIZEOF_XMMWORD/16 + movd xmmA, XMM_DWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld2: + test cl, SIZEOF_XMMWORD/8 + jz short .column_ld4 + sub rcx, byte SIZEOF_XMMWORD/8 + movq xmmE, XMM_MMWORD [rsi+rcx*RGB_PIXELSIZE] + pslldq xmmA, SIZEOF_MMWORD + por xmmA, xmmE +.column_ld4: + test cl, SIZEOF_XMMWORD/4 + jz short .column_ld8 + sub rcx, byte SIZEOF_XMMWORD/4 + movdqa xmmE, xmmA + movdqu xmmA, XMMWORD [rsi+rcx*RGB_PIXELSIZE] +.column_ld8: + test cl, SIZEOF_XMMWORD/2 + mov rcx, SIZEOF_XMMWORD + jz short .rgb_gray_cnv + movdqa xmmF, xmmA + movdqa xmmH, xmmE + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD] + jmp short .rgb_gray_cnv + +.columnloop: + movdqu xmmA, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqu xmmE, XMMWORD [rsi+1*SIZEOF_XMMWORD] + movdqu xmmF, XMMWORD [rsi+2*SIZEOF_XMMWORD] + movdqu xmmH, XMMWORD [rsi+3*SIZEOF_XMMWORD] + +.rgb_gray_cnv: + ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + ; xmmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + ; xmmF=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpcklbw xmmA, xmmE ; xmmA=(00 04 10 14 20 24 30 34 01 05 11 15 21 25 31 35) + punpckhbw xmmD, xmmE ; xmmD=(02 06 12 16 22 26 32 36 03 07 13 17 23 27 33 37) + + movdqa xmmC, xmmF + punpcklbw xmmF, xmmH ; xmmF=(08 0C 18 1C 28 2C 38 3C 09 0D 19 1D 29 2D 39 3D) + punpckhbw xmmC, xmmH ; xmmC=(0A 0E 1A 1E 2A 2E 3A 3E 0B 0F 1B 1F 2B 2F 3B 3F) + + movdqa xmmB, xmmA + punpcklwd xmmA, xmmF ; xmmA=(00 04 08 0C 10 14 18 1C 20 24 28 2C 30 34 38 3C) + punpckhwd xmmB, xmmF ; xmmB=(01 05 09 0D 11 15 19 1D 21 25 29 2D 31 35 39 3D) + + movdqa xmmG, xmmD + punpcklwd xmmD, xmmC ; xmmD=(02 06 0A 0E 12 16 1A 1E 22 26 2A 2E 32 36 3A 3E) + punpckhwd xmmG, xmmC ; xmmG=(03 07 0B 0F 13 17 1B 1F 23 27 2B 2F 33 37 3B 3F) + + movdqa xmmE, xmmA + punpcklbw xmmA, xmmD ; xmmA=(00 02 04 06 08 0A 0C 0E 10 12 14 16 18 1A 1C 1E) + punpckhbw xmmE, xmmD ; xmmE=(20 22 24 26 28 2A 2C 2E 30 32 34 36 38 3A 3C 3E) + + movdqa xmmH, xmmB + punpcklbw xmmB, xmmG ; xmmB=(01 03 05 07 09 0B 0D 0F 11 13 15 17 19 1B 1D 1F) + punpckhbw xmmH, xmmG ; xmmH=(21 23 25 27 29 2B 2D 2F 31 33 35 37 39 3B 3D 3F) + + pxor xmmF, xmmF + + movdqa xmmC, xmmA + punpcklbw xmmA, xmmF ; xmmA=(00 02 04 06 08 0A 0C 0E) + punpckhbw xmmC, xmmF ; xmmC=(10 12 14 16 18 1A 1C 1E) + + movdqa xmmD, xmmB + punpcklbw xmmB, xmmF ; xmmB=(01 03 05 07 09 0B 0D 0F) + punpckhbw xmmD, xmmF ; xmmD=(11 13 15 17 19 1B 1D 1F) + + movdqa xmmG, xmmE + punpcklbw xmmE, xmmF ; xmmE=(20 22 24 26 28 2A 2C 2E) + punpckhbw xmmG, xmmF ; xmmG=(30 32 34 36 38 3A 3C 3E) + + punpcklbw xmmF, xmmH + punpckhbw xmmH, xmmH + psrlw xmmF, BYTE_BIT ; xmmF=(21 23 25 27 29 2B 2D 2F) + psrlw xmmH, BYTE_BIT ; xmmH=(31 33 35 37 39 3B 3D 3F) + +%endif ; RGB_PIXELSIZE ; --------------- + + ; xmm0=R(02468ACE)=RE, xmm2=G(02468ACE)=GE, xmm4=B(02468ACE)=BE + ; xmm1=R(13579BDF)=RO, xmm3=G(13579BDF)=GO, xmm5=B(13579BDF)=BO + + ; (Original) + ; Y = 0.29900 * R + 0.58700 * G + 0.11400 * B + ; + ; (This implementation) + ; Y = 0.29900 * R + 0.33700 * G + 0.11400 * B + 0.25000 * G + + movdqa xmm6, xmm1 + punpcklwd xmm1, xmm3 + punpckhwd xmm6, xmm3 + pmaddwd xmm1, [rel PW_F0299_F0337] ; xmm1=ROL*FIX(0.299)+GOL*FIX(0.337) + pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=ROH*FIX(0.299)+GOH*FIX(0.337) + + movdqa xmm7, xmm6 ; xmm7=ROH*FIX(0.299)+GOH*FIX(0.337) + + movdqa xmm6, xmm0 + punpcklwd xmm0, xmm2 + punpckhwd xmm6, xmm2 + pmaddwd xmm0, [rel PW_F0299_F0337] ; xmm0=REL*FIX(0.299)+GEL*FIX(0.337) + pmaddwd xmm6, [rel PW_F0299_F0337] ; xmm6=REH*FIX(0.299)+GEH*FIX(0.337) + + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=REL*FIX(0.299)+GEL*FIX(0.337) + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=REH*FIX(0.299)+GEH*FIX(0.337) + + movdqa xmm0, xmm5 ; xmm0=BO + movdqa xmm6, xmm4 ; xmm6=BE + + movdqa xmm4, xmm0 + punpcklwd xmm0, xmm3 + punpckhwd xmm4, xmm3 + pmaddwd xmm0, [rel PW_F0114_F0250] ; xmm0=BOL*FIX(0.114)+GOL*FIX(0.250) + pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BOH*FIX(0.114)+GOH*FIX(0.250) + + movdqa xmm3, [rel PD_ONEHALF] ; xmm3=[PD_ONEHALF] + + paddd xmm0, xmm1 + paddd xmm4, xmm7 + paddd xmm0, xmm3 + paddd xmm4, xmm3 + psrld xmm0, SCALEBITS ; xmm0=YOL + psrld xmm4, SCALEBITS ; xmm4=YOH + packssdw xmm0, xmm4 ; xmm0=YO + + movdqa xmm4, xmm6 + punpcklwd xmm6, xmm2 + punpckhwd xmm4, xmm2 + pmaddwd xmm6, [rel PW_F0114_F0250] ; xmm6=BEL*FIX(0.114)+GEL*FIX(0.250) + pmaddwd xmm4, [rel PW_F0114_F0250] ; xmm4=BEH*FIX(0.114)+GEH*FIX(0.250) + + movdqa xmm2, [rel PD_ONEHALF] ; xmm2=[PD_ONEHALF] + + paddd xmm6, XMMWORD [wk(0)] + paddd xmm4, XMMWORD [wk(1)] + paddd xmm6, xmm2 + paddd xmm4, xmm2 + psrld xmm6, SCALEBITS ; xmm6=YEL + psrld xmm4, SCALEBITS ; xmm4=YEH + packssdw xmm6, xmm4 ; xmm6=YE + + psllw xmm0, BYTE_BIT + por xmm6, xmm0 ; xmm6=Y + movdqa XMMWORD [rdi], xmm6 ; Save Y + + sub rcx, byte SIZEOF_XMMWORD + add rsi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; inptr + add rdi, byte SIZEOF_XMMWORD ; outptr0 + cmp rcx, byte SIZEOF_XMMWORD + jae near .columnloop + test rcx, rcx + jnz near .column_ld1 + + pop rcx ; col + pop rsi + pop rdi + + add rsi, byte SIZEOF_JSAMPROW ; input_buf + add rdi, byte SIZEOF_JSAMPROW + dec rax ; num_rows + jg near .rowloop + +.return: + pop rbx + UNCOLLECT_ARGS 5 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jchuff-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jchuff-sse2.asm new file mode 100644 index 0000000..b18b7f5 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jchuff-sse2.asm @@ -0,0 +1,581 @@ +; +; jchuff-sse2.asm - Huffman entropy encoding (64-bit SSE2) +; +; Copyright (C) 2009-2011, 2014-2016, 2019, 2021, 2023-2024, D. R. Commander. +; Copyright (C) 2015, Matthieu Darbois. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains an SSE2 implementation for Huffman coding of one block. +; The following code is based on jchuff.c; see jchuff.c for more details. + +%include "jsimdext.inc" + +struc working_state +.next_output_byte: resp 1 ; => next byte to write in buffer +.free_in_buffer: resp 1 ; # of byte spaces remaining in buffer +.cur.put_buffer.simd resq 1 ; current bit accumulation buffer +.cur.free_bits resd 1 ; # of bits available in it +.cur.last_dc_val resd 4 ; last DC coef for each component +.cinfo: resp 1 ; dump_buffer needs access to this +endstruc + +struc c_derived_tbl +.ehufco: resd 256 ; code for each symbol +.ehufsi: resb 256 ; length of code for each symbol +; If no code has been allocated for a symbol S, ehufsi[S] contains 0 +endstruc + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_huff_encode_one_block) + +EXTN(jconst_huff_encode_one_block): + +jpeg_mask_bits dd 0x0000, 0x0001, 0x0003, 0x0007 + dd 0x000f, 0x001f, 0x003f, 0x007f + dd 0x00ff, 0x01ff, 0x03ff, 0x07ff + dd 0x0fff, 0x1fff, 0x3fff, 0x7fff + + ALIGNZ 32 + +times 1 << 14 db 15 +times 1 << 13 db 14 +times 1 << 12 db 13 +times 1 << 11 db 12 +times 1 << 10 db 11 +times 1 << 9 db 10 +times 1 << 8 db 9 +times 1 << 7 db 8 +times 1 << 6 db 7 +times 1 << 5 db 6 +times 1 << 4 db 5 +times 1 << 3 db 4 +times 1 << 2 db 3 +times 1 << 1 db 2 +times 1 << 0 db 1 +times 1 db 0 +GLOBAL_DATA(jpeg_nbits_table) +EXTN(jpeg_nbits_table): +times 1 db 0 +times 1 << 0 db 1 +times 1 << 1 db 2 +times 1 << 2 db 3 +times 1 << 3 db 4 +times 1 << 4 db 5 +times 1 << 5 db 6 +times 1 << 6 db 7 +times 1 << 7 db 8 +times 1 << 8 db 9 +times 1 << 9 db 10 +times 1 << 10 db 11 +times 1 << 11 db 12 +times 1 << 12 db 13 +times 1 << 13 db 14 +times 1 << 14 db 15 +times 1 << 15 db 16 + + ALIGNZ 32 + +%define NBITS(x) nbits_base + x +%define MASK_BITS(x) NBITS((x) * 4) + (jpeg_mask_bits - EXTN(jpeg_nbits_table)) + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +; Shorthand used to describe SIMD operations: +; wN: xmmN treated as eight signed 16-bit values +; wN[i]: perform the same operation on all eight signed 16-bit values, i=0..7 +; bN: xmmN treated as 16 unsigned 8-bit values +; bN[i]: perform the same operation on all 16 unsigned 8-bit values, i=0..15 +; Contents of SIMD registers are shown in memory order. + +; Fill the bit buffer to capacity with the leading bits from code, then output +; the bit buffer and put the remaining bits from code into the bit buffer. +; +; Usage: +; code - contains the bits to shift into the bit buffer (LSB-aligned) +; %1 - the label to which to jump when the macro completes +; %2 (optional) - extra instructions to execute after nbits has been set +; +; Upon completion, free_bits will be set to the number of remaining bits from +; code, and put_buffer will contain those remaining bits. temp and code will +; be clobbered. +; +; This macro encodes any 0xFF bytes as 0xFF 0x00, as does the EMIT_BYTE() +; macro in jchuff.c. + +%macro EMIT_QWORD 1-2 + add nbitsb, free_bitsb ; nbits += free_bits; + neg free_bitsb ; free_bits = -free_bits; + mov tempd, code ; temp = code; + shl put_buffer, nbitsb ; put_buffer <<= nbits; + mov nbitsb, free_bitsb ; nbits = free_bits; + neg free_bitsb ; free_bits = -free_bits; + shr tempd, nbitsb ; temp >>= nbits; + or tempq, put_buffer ; temp |= put_buffer; + movq xmm0, tempq ; xmm0.u64 = { temp, 0 }; + bswap tempq ; temp = htonl(temp); + mov put_buffer, codeq ; put_buffer = code; + pcmpeqb xmm0, xmm1 ; b0[i] = (b0[i] == 0xFF ? 0xFF : 0); + %2 + pmovmskb code, xmm0 ; code = 0; code |= ((b0[i] >> 7) << i); + mov qword [buffer], tempq ; memcpy(buffer, &temp, 8); + ; (speculative; will be overwritten if + ; code contains any 0xFF bytes) + add free_bitsb, 64 ; free_bits += 64; + add bufferp, 8 ; buffer += 8; + test code, code ; if (code == 0) /* No 0xFF bytes */ + jz %1 ; return; + ; Execute the equivalent of the EMIT_BYTE() macro in jchuff.c for all 8 + ; bytes in the qword. + cmp tempb, 0xFF ; Set CF if temp[0] < 0xFF + mov byte [buffer-7], 0 ; buffer[-7] = 0; + sbb bufferp, 6 ; buffer -= (6 + (temp[0] < 0xFF ? 1 : 0)); + mov byte [buffer], temph ; buffer[0] = temp[1]; + cmp temph, 0xFF ; Set CF if temp[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb bufferp, -2 ; buffer -= (-2 + (temp[1] < 0xFF ? 1 : 0)); + shr tempq, 16 ; temp >>= 16; + mov byte [buffer], tempb ; buffer[0] = temp[0]; + cmp tempb, 0xFF ; Set CF if temp[0] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb bufferp, -2 ; buffer -= (-2 + (temp[0] < 0xFF ? 1 : 0)); + mov byte [buffer], temph ; buffer[0] = temp[1]; + cmp temph, 0xFF ; Set CF if temp[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb bufferp, -2 ; buffer -= (-2 + (temp[1] < 0xFF ? 1 : 0)); + shr tempq, 16 ; temp >>= 16; + mov byte [buffer], tempb ; buffer[0] = temp[0]; + cmp tempb, 0xFF ; Set CF if temp[0] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb bufferp, -2 ; buffer -= (-2 + (temp[0] < 0xFF ? 1 : 0)); + mov byte [buffer], temph ; buffer[0] = temp[1]; + cmp temph, 0xFF ; Set CF if temp[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb bufferp, -2 ; buffer -= (-2 + (temp[1] < 0xFF ? 1 : 0)); + shr tempd, 16 ; temp >>= 16; + mov byte [buffer], tempb ; buffer[0] = temp[0]; + cmp tempb, 0xFF ; Set CF if temp[0] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb bufferp, -2 ; buffer -= (-2 + (temp[0] < 0xFF ? 1 : 0)); + mov byte [buffer], temph ; buffer[0] = temp[1]; + cmp temph, 0xFF ; Set CF if temp[1] < 0xFF + mov byte [buffer+1], 0 ; buffer[1] = 0; + sbb bufferp, -2 ; buffer -= (-2 + (temp[1] < 0xFF ? 1 : 0)); + jmp %1 ; return; +%endmacro + +; +; Encode a single block's worth of coefficients. +; +; GLOBAL(JOCTET *) +; jsimd_huff_encode_one_block_sse2(working_state *state, JOCTET *buffer, +; JCOEFPTR block, int last_dc_val, +; c_derived_tbl *dctbl, c_derived_tbl *actbl) +; +; NOTES: +; When shuffling data, we try to avoid pinsrw as much as possible, since it is +; slow on many CPUs. Its reciprocal throughput (issue latency) is 1 even on +; modern CPUs, so chains of pinsrw instructions (even with different outputs) +; can limit performance. pinsrw is a VectorPath instruction on AMD K8 and +; requires 2 µops (with memory operand) on Intel. In either case, only one +; pinsrw instruction can be decoded per cycle (and nothing else if they are +; back-to-back), so out-of-order execution cannot be used to work around long +; pinsrw chains (though for Sandy Bridge and later, this may be less of a +; problem if the code runs from the µop cache.) +; +; We use tzcnt instead of bsf without checking for support. The instruction is +; executed as bsf on CPUs that don't support tzcnt (encoding is equivalent to +; rep bsf.) The destination (first) operand of bsf (and tzcnt on some CPUs) is +; an input dependency (although the behavior is not formally defined, Intel +; CPUs usually leave the destination unmodified if the source is zero.) This +; can prevent out-of-order execution, so we clear the destination before +; invoking tzcnt. +; +; Initial register allocation +; rax - buffer +; rbx - temp +; rcx - nbits +; rdx - code +; rsi - nbits_base +; rdi - t +; r8 - dctbl --> code_temp +; r9 - actbl +; r10 - state +; r11 - index +; r12 - put_buffer +; r15 - block --> free_bits + +%define buffer rax +%ifdef WIN64 +%define bufferp rax +%else +%define bufferp raxp +%endif +%define tempq rbx +%define tempd ebx +%define tempb bl +%define temph bh +%define nbitsq rcx +%define nbits ecx +%define nbitsb cl +%define codeq rdx +%define code edx +%define nbits_base rsi +%define t rdi +%define td edi +%define dctbl r8 +%define actbl r9 +%define state r10 +%define index r11 +%define indexd r11d +%define put_buffer r12 +%define put_bufferd r12d +%define block r15 + +; Step 1: Re-arrange input data according to jpeg_natural_order +; xx 01 02 03 04 05 06 07 xx 01 08 16 09 02 03 10 +; 08 09 10 11 12 13 14 15 17 24 32 25 18 11 04 05 +; 16 17 18 19 20 21 22 23 12 19 26 33 40 48 41 34 +; 24 25 26 27 28 29 30 31 ==> 27 20 13 06 07 14 21 28 +; 32 33 34 35 36 37 38 39 35 42 49 56 57 50 43 36 +; 40 41 42 43 44 45 46 47 29 22 15 23 30 37 44 51 +; 48 49 50 51 52 53 54 55 58 59 52 45 38 31 39 46 +; 56 57 58 59 60 61 62 63 53 60 61 54 47 55 62 63 + + align 32 + GLOBAL_FUNCTION(jsimd_huff_encode_one_block_sse2) + +EXTN(jsimd_huff_encode_one_block_sse2): + ENDBR64 + push rbp + mov rbp, rsp + +%ifdef WIN64 + +; rcx = working_state *state +; rdx = JOCTET *buffer +; r8 = JCOEFPTR block +; r9 = int last_dc_val +; [rbp+48] = c_derived_tbl *dctbl +; [rbp+56] = c_derived_tbl *actbl + + ;X: X = code stream + mov buffer, rdx + push r15 + mov block, r8 + movups xmm3, XMMWORD [block + 0 * SIZEOF_WORD] ;D: w3 = xx 01 02 03 04 05 06 07 + push rbx + movdqa xmm0, xmm3 ;A: w0 = xx 01 02 03 04 05 06 07 + push rsi + push rdi + push r12 + movups xmm1, XMMWORD [block + 8 * SIZEOF_WORD] ;B: w1 = 08 09 10 11 12 13 14 15 + mov state, rcx + movsx code, word [block] ;Z: code = block[0]; + pxor xmm4, xmm4 ;A: w4[i] = 0; + sub code, r9d ;Z: code -= last_dc_val; + mov dctbl, POINTER [rbp+48] + mov actbl, POINTER [rbp+56] + punpckldq xmm0, xmm1 ;A: w0 = xx 01 08 09 02 03 10 11 + lea nbits_base, [rel EXTN(jpeg_nbits_table)] + +%else + +; rdi = working_state *state +; rsi = JOCTET *buffer +; rdx = JCOEFPTR block +; rcx = int last_dc_val +; r8 = c_derived_tbl *dctbl +; r9 = c_derived_tbl *actbl + + ;X: X = code stream + push r15 + mov block, rdx + movups xmm3, XMMWORD [block + 0 * SIZEOF_WORD] ;D: w3 = xx 01 02 03 04 05 06 07 + push rbx + movdqa xmm0, xmm3 ;A: w0 = xx 01 02 03 04 05 06 07 + push r12 + mov state, rdi + mov buffer, rsi + movups xmm1, XMMWORD [block + 8 * SIZEOF_WORD] ;B: w1 = 08 09 10 11 12 13 14 15 + movsx codeq, word [block] ;Z: code = block[0]; + lea nbits_base, [rel EXTN(jpeg_nbits_table)] + pxor xmm4, xmm4 ;A: w4[i] = 0; + sub codeq, rcx ;Z: code -= last_dc_val; + punpckldq xmm0, xmm1 ;A: w0 = xx 01 08 09 02 03 10 11 + +%endif + + ; Allocate stack space for t array, and realign stack. + add rsp, -DCTSIZE2 * SIZEOF_WORD - 8 + mov t, rsp + + pshuflw xmm0, xmm0, 11001001b ;A: w0 = 01 08 xx 09 02 03 10 11 + pinsrw xmm0, word [block + 16 * SIZEOF_WORD], 2 ;A: w0 = 01 08 16 09 02 03 10 11 + punpckhdq xmm3, xmm1 ;D: w3 = 04 05 12 13 06 07 14 15 + punpcklqdq xmm1, xmm3 ;B: w1 = 08 09 10 11 04 05 12 13 + pinsrw xmm0, word [block + 17 * SIZEOF_WORD], 7 ;A: w0 = 01 08 16 09 02 03 10 17 + ;A: (Row 0, offset 1) + pcmpgtw xmm4, xmm0 ;A: w4[i] = (w0[i] < 0 ? -1 : 0); + paddw xmm0, xmm4 ;A: w0[i] += w4[i]; + movaps XMMWORD [t + 0 * SIZEOF_WORD], xmm0 ;A: t[i] = w0[i]; + + movq xmm2, qword [block + 24 * SIZEOF_WORD] ;B: w2 = 24 25 26 27 -- -- -- -- + pshuflw xmm2, xmm2, 11011000b ;B: w2 = 24 26 25 27 -- -- -- -- + pslldq xmm1, 1 * SIZEOF_WORD ;B: w1 = -- 08 09 10 11 04 05 12 + movups xmm5, XMMWORD [block + 48 * SIZEOF_WORD] ;H: w5 = 48 49 50 51 52 53 54 55 + movsd xmm1, xmm2 ;B: w1 = 24 26 25 27 11 04 05 12 + punpcklqdq xmm2, xmm5 ;C: w2 = 24 26 25 27 48 49 50 51 + pinsrw xmm1, word [block + 32 * SIZEOF_WORD], 1 ;B: w1 = 24 32 25 27 11 04 05 12 + pxor xmm4, xmm4 ;A: w4[i] = 0; + psrldq xmm3, 2 * SIZEOF_WORD ;D: w3 = 12 13 06 07 14 15 -- -- + pcmpeqw xmm0, xmm4 ;A: w0[i] = (w0[i] == 0 ? -1 : 0); + pinsrw xmm1, word [block + 18 * SIZEOF_WORD], 3 ;B: w1 = 24 32 25 18 11 04 05 12 + ; (Row 1, offset 1) + pcmpgtw xmm4, xmm1 ;B: w4[i] = (w1[i] < 0 ? -1 : 0); + paddw xmm1, xmm4 ;B: w1[i] += w4[i]; + movaps XMMWORD [t + 8 * SIZEOF_WORD], xmm1 ;B: t[i+8] = w1[i]; + pxor xmm4, xmm4 ;B: w4[i] = 0; + pcmpeqw xmm1, xmm4 ;B: w1[i] = (w1[i] == 0 ? -1 : 0); + + packsswb xmm0, xmm1 ;AB: b0[i] = w0[i], b0[i+8] = w1[i] + ; w/ signed saturation + + pinsrw xmm3, word [block + 20 * SIZEOF_WORD], 0 ;D: w3 = 20 13 06 07 14 15 -- -- + pinsrw xmm3, word [block + 21 * SIZEOF_WORD], 5 ;D: w3 = 20 13 06 07 14 21 -- -- + pinsrw xmm3, word [block + 28 * SIZEOF_WORD], 6 ;D: w3 = 20 13 06 07 14 21 28 -- + pinsrw xmm3, word [block + 35 * SIZEOF_WORD], 7 ;D: w3 = 20 13 06 07 14 21 28 35 + ; (Row 3, offset 1) + pcmpgtw xmm4, xmm3 ;D: w4[i] = (w3[i] < 0 ? -1 : 0); + paddw xmm3, xmm4 ;D: w3[i] += w4[i]; + movaps XMMWORD [t + 24 * SIZEOF_WORD], xmm3 ;D: t[i+24] = w3[i]; + pxor xmm4, xmm4 ;D: w4[i] = 0; + pcmpeqw xmm3, xmm4 ;D: w3[i] = (w3[i] == 0 ? -1 : 0); + + pinsrw xmm2, word [block + 19 * SIZEOF_WORD], 0 ;C: w2 = 19 26 25 27 48 49 50 51 + cmp code, 1 << 31 ;Z: Set CF if code < 0x80000000, + ;Z: i.e. if code is positive + pinsrw xmm2, word [block + 33 * SIZEOF_WORD], 2 ;C: w2 = 19 26 33 27 48 49 50 51 + pinsrw xmm2, word [block + 40 * SIZEOF_WORD], 3 ;C: w2 = 19 26 33 40 48 49 50 51 + adc code, -1 ;Z: code += -1 + (code >= 0 ? 1 : 0); + pinsrw xmm2, word [block + 41 * SIZEOF_WORD], 5 ;C: w2 = 19 26 33 40 48 41 50 51 + pinsrw xmm2, word [block + 34 * SIZEOF_WORD], 6 ;C: w2 = 19 26 33 40 48 41 34 51 + movsxd codeq, code ;Z: sign extend code + pinsrw xmm2, word [block + 27 * SIZEOF_WORD], 7 ;C: w2 = 19 26 33 40 48 41 34 27 + ; (Row 2, offset 1) + pcmpgtw xmm4, xmm2 ;C: w4[i] = (w2[i] < 0 ? -1 : 0); + paddw xmm2, xmm4 ;C: w2[i] += w4[i]; + movaps XMMWORD [t + 16 * SIZEOF_WORD], xmm2 ;C: t[i+16] = w2[i]; + pxor xmm4, xmm4 ;C: w4[i] = 0; + pcmpeqw xmm2, xmm4 ;C: w2[i] = (w2[i] == 0 ? -1 : 0); + + packsswb xmm2, xmm3 ;CD: b2[i] = w2[i], b2[i+8] = w3[i] + ; w/ signed saturation + + movzx nbitsq, byte [NBITS(codeq)] ;Z: nbits = JPEG_NBITS(code); + movdqa xmm3, xmm5 ;H: w3 = 48 49 50 51 52 53 54 55 + pmovmskb tempd, xmm2 ;Z: temp = 0; temp |= ((b2[i] >> 7) << i); + pmovmskb put_bufferd, xmm0 ;Z: put_buffer = 0; put_buffer |= ((b0[i] >> 7) << i); + movups xmm0, XMMWORD [block + 56 * SIZEOF_WORD] ;H: w0 = 56 57 58 59 60 61 62 63 + punpckhdq xmm3, xmm0 ;H: w3 = 52 53 60 61 54 55 62 63 + shl tempd, 16 ;Z: temp <<= 16; + psrldq xmm3, 1 * SIZEOF_WORD ;H: w3 = 53 60 61 54 55 62 63 -- + pxor xmm2, xmm2 ;H: w2[i] = 0; + or put_bufferd, tempd ;Z: put_buffer |= temp; + pshuflw xmm3, xmm3, 00111001b ;H: w3 = 60 61 54 53 55 62 63 -- + movq xmm1, qword [block + 44 * SIZEOF_WORD] ;G: w1 = 44 45 46 47 -- -- -- -- + unpcklps xmm5, xmm0 ;E: w5 = 48 49 56 57 50 51 58 59 + pxor xmm0, xmm0 ;H: w0[i] = 0; + pinsrw xmm3, word [block + 47 * SIZEOF_WORD], 3 ;H: w3 = 60 61 54 47 55 62 63 -- + ; (Row 7, offset 1) + pcmpgtw xmm2, xmm3 ;H: w2[i] = (w3[i] < 0 ? -1 : 0); + paddw xmm3, xmm2 ;H: w3[i] += w2[i]; + movaps XMMWORD [t + 56 * SIZEOF_WORD], xmm3 ;H: t[i+56] = w3[i]; + movq xmm4, qword [block + 36 * SIZEOF_WORD] ;G: w4 = 36 37 38 39 -- -- -- -- + pcmpeqw xmm3, xmm0 ;H: w3[i] = (w3[i] == 0 ? -1 : 0); + punpckldq xmm4, xmm1 ;G: w4 = 36 37 44 45 38 39 46 47 + mov tempd, [dctbl + c_derived_tbl.ehufco + nbitsq * 4] + ;Z: temp = dctbl->ehufco[nbits]; + movdqa xmm1, xmm4 ;F: w1 = 36 37 44 45 38 39 46 47 + psrldq xmm4, 1 * SIZEOF_WORD ;G: w4 = 37 44 45 38 39 46 47 -- + shufpd xmm1, xmm5, 10b ;F: w1 = 36 37 44 45 50 51 58 59 + and code, dword [MASK_BITS(nbitsq)] ;Z: code &= (1 << nbits) - 1; + pshufhw xmm4, xmm4, 11010011b ;G: w4 = 37 44 45 38 -- 39 46 -- + pslldq xmm1, 1 * SIZEOF_WORD ;F: w1 = -- 36 37 44 45 50 51 58 + shl tempq, nbitsb ;Z: temp <<= nbits; + pinsrw xmm4, word [block + 59 * SIZEOF_WORD], 0 ;G: w4 = 59 44 45 38 -- 39 46 -- + pshufd xmm1, xmm1, 11011000b ;F: w1 = -- 36 45 50 37 44 51 58 + pinsrw xmm4, word [block + 52 * SIZEOF_WORD], 1 ;G: w4 = 59 52 45 38 -- 39 46 -- + or code, tempd ;Z: code |= temp; + movlps xmm1, qword [block + 20 * SIZEOF_WORD] ;F: w1 = 20 21 22 23 37 44 51 58 + pinsrw xmm4, word [block + 31 * SIZEOF_WORD], 4 ;G: w4 = 59 52 45 38 31 39 46 -- + pshuflw xmm1, xmm1, 01110010b ;F: w1 = 22 20 23 21 37 44 51 58 + pinsrw xmm4, word [block + 53 * SIZEOF_WORD], 7 ;G: w4 = 59 52 45 38 31 39 46 53 + ; (Row 6, offset 1) + pxor xmm2, xmm2 ;G: w2[i] = 0; + pcmpgtw xmm0, xmm4 ;G: w0[i] = (w4[i] < 0 ? -1 : 0); + pinsrw xmm1, word [block + 15 * SIZEOF_WORD], 1 ;F: w1 = 22 15 23 21 37 44 51 58 + paddw xmm4, xmm0 ;G: w4[i] += w0[i]; + movaps XMMWORD [t + 48 * SIZEOF_WORD], xmm4 ;G: t[48+i] = w4[i]; + pinsrw xmm1, word [block + 30 * SIZEOF_WORD], 3 ;F: w1 = 22 15 23 30 37 44 51 58 + ; (Row 5, offset 1) + pcmpeqw xmm4, xmm2 ;G: w4[i] = (w4[i] == 0 ? -1 : 0); + pinsrw xmm5, word [block + 42 * SIZEOF_WORD], 0 ;E: w5 = 42 49 56 57 50 51 58 59 + + packsswb xmm4, xmm3 ;GH: b4[i] = w4[i], b4[i+8] = w3[i] + ; w/ signed saturation + + pxor xmm0, xmm0 ;F: w0[i] = 0; + pinsrw xmm5, word [block + 43 * SIZEOF_WORD], 5 ;E: w5 = 42 49 56 57 50 43 58 59 + pcmpgtw xmm2, xmm1 ;F: w2[i] = (w1[i] < 0 ? -1 : 0); + pmovmskb tempd, xmm4 ;Z: temp = 0; temp |= ((b4[i] >> 7) << i); + pinsrw xmm5, word [block + 36 * SIZEOF_WORD], 6 ;E: w5 = 42 49 56 57 50 43 36 59 + paddw xmm1, xmm2 ;F: w1[i] += w2[i]; + movaps XMMWORD [t + 40 * SIZEOF_WORD], xmm1 ;F: t[40+i] = w1[i]; + pinsrw xmm5, word [block + 29 * SIZEOF_WORD], 7 ;E: w5 = 42 49 56 57 50 43 36 29 + ; (Row 4, offset 1) +%undef block +%define free_bitsq r15 +%define free_bitsd r15d +%define free_bitsb r15b + pcmpeqw xmm1, xmm0 ;F: w1[i] = (w1[i] == 0 ? -1 : 0); + shl tempq, 48 ;Z: temp <<= 48; + pxor xmm2, xmm2 ;E: w2[i] = 0; + pcmpgtw xmm0, xmm5 ;E: w0[i] = (w5[i] < 0 ? -1 : 0); + paddw xmm5, xmm0 ;E: w5[i] += w0[i]; + or tempq, put_buffer ;Z: temp |= put_buffer; + movaps XMMWORD [t + 32 * SIZEOF_WORD], xmm5 ;E: t[32+i] = w5[i]; + lea t, [dword t - 2] ;Z: t = &t[-1]; + pcmpeqw xmm5, xmm2 ;E: w5[i] = (w5[i] == 0 ? -1 : 0); + + packsswb xmm5, xmm1 ;EF: b5[i] = w5[i], b5[i+8] = w1[i] + ; w/ signed saturation + + add nbitsb, byte [dctbl + c_derived_tbl.ehufsi + nbitsq] + ;Z: nbits += dctbl->ehufsi[nbits]; +%undef dctbl +%define code_temp r8d + pmovmskb indexd, xmm5 ;Z: index = 0; index |= ((b5[i] >> 7) << i); + mov free_bitsd, [state+working_state.cur.free_bits] + ;Z: free_bits = state->cur.free_bits; + pcmpeqw xmm1, xmm1 ;Z: b1[i] = 0xFF; + shl index, 32 ;Z: index <<= 32; + mov put_buffer, [state+working_state.cur.put_buffer.simd] + ;Z: put_buffer = state->cur.put_buffer.simd; + or index, tempq ;Z: index |= temp; + not index ;Z: index = ~index; + sub free_bitsb, nbitsb ;Z: if ((free_bits -= nbits) >= 0) + jnl .ENTRY_SKIP_EMIT_CODE ;Z: goto .ENTRY_SKIP_EMIT_CODE; + align 16 +.EMIT_CODE: ;Z: .EMIT_CODE: + EMIT_QWORD .BLOOP_COND ;Z: insert code, flush buffer, goto .BLOOP_COND + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +.BRLOOP: ; do { + lea code_temp, [nbitsq - 16] ; code_temp = nbits - 16; + movzx nbits, byte [actbl + c_derived_tbl.ehufsi + 0xf0] + ; nbits = actbl->ehufsi[0xf0]; + mov code, [actbl + c_derived_tbl.ehufco + 0xf0 * 4] + ; code = actbl->ehufco[0xf0]; + sub free_bitsb, nbitsb ; if ((free_bits -= nbits) <= 0) + jle .EMIT_BRLOOP_CODE ; goto .EMIT_BRLOOP_CODE; + shl put_buffer, nbitsb ; put_buffer <<= nbits; + mov nbits, code_temp ; nbits = code_temp; + or put_buffer, codeq ; put_buffer |= code; + cmp nbits, 16 ; if (nbits <= 16) + jle .ERLOOP ; break; + jmp .BRLOOP ; } while (1); + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 + times 5 nop +.ENTRY_SKIP_EMIT_CODE: ; .ENTRY_SKIP_EMIT_CODE: + shl put_buffer, nbitsb ; put_buffer <<= nbits; + or put_buffer, codeq ; put_buffer |= code; +.BLOOP_COND: ; .BLOOP_COND: + test index, index ; if (index != 0) + jz .ELOOP ; { +.BLOOP: ; do { + xor nbits, nbits ; nbits = 0; /* kill tzcnt input dependency */ + tzcnt nbitsq, index ; nbits = # of trailing 0 bits in index + inc nbits ; ++nbits; + lea t, [t + nbitsq * 2] ; t = &t[nbits]; + shr index, nbitsb ; index >>= nbits; +.EMIT_BRLOOP_CODE_END: ; .EMIT_BRLOOP_CODE_END: + cmp nbits, 16 ; if (nbits > 16) + jg .BRLOOP ; goto .BRLOOP; +.ERLOOP: ; .ERLOOP: + movsx codeq, word [t] ; code = *t; + lea tempd, [nbitsq * 2] ; temp = nbits * 2; + movzx nbits, byte [NBITS(codeq)] ; nbits = JPEG_NBITS(code); + lea tempd, [nbitsq + tempq * 8] ; temp = temp * 8 + nbits; + mov code_temp, [actbl + c_derived_tbl.ehufco + (tempq - 16) * 4] + ; code_temp = actbl->ehufco[temp-16]; + shl code_temp, nbitsb ; code_temp <<= nbits; + and code, dword [MASK_BITS(nbitsq)] ; code &= (1 << nbits) - 1; + add nbitsb, [actbl + c_derived_tbl.ehufsi + (tempq - 16)] + ; free_bits -= actbl->ehufsi[temp-16]; + or code, code_temp ; code |= code_temp; + sub free_bitsb, nbitsb ; if ((free_bits -= nbits) <= 0) + jle .EMIT_CODE ; goto .EMIT_CODE; + shl put_buffer, nbitsb ; put_buffer <<= nbits; + or put_buffer, codeq ; put_buffer |= code; + test index, index + jnz .BLOOP ; } while (index != 0); +.ELOOP: ; } /* index != 0 */ + sub td, esp ; t -= &t_[0]; + cmp td, (DCTSIZE2 - 2) * SIZEOF_WORD ; if (t != 62) + je .EFN ; { + movzx nbits, byte [actbl + c_derived_tbl.ehufsi + 0] + ; nbits = actbl->ehufsi[0]; + mov code, [actbl + c_derived_tbl.ehufco + 0] ; code = actbl->ehufco[0]; + sub free_bitsb, nbitsb ; if ((free_bits -= nbits) <= 0) + jg .EFN_SKIP_EMIT_CODE ; { + EMIT_QWORD .EFN ; insert code, flush buffer + align 16 +.EFN_SKIP_EMIT_CODE: ; } else { + shl put_buffer, nbitsb ; put_buffer <<= nbits; + or put_buffer, codeq ; put_buffer |= code; +.EFN: ; } } + mov [state + working_state.cur.put_buffer.simd], put_buffer + ; state->cur.put_buffer.simd = put_buffer; + mov byte [state + working_state.cur.free_bits], free_bitsb + ; state->cur.free_bits = free_bits; + sub rsp, -DCTSIZE2 * SIZEOF_WORD - 8 + pop r12 +%ifdef WIN64 + pop rdi + pop rsi + pop rbx +%else + pop rbx +%endif + pop r15 + pop rbp + ret + +; ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ + + align 16 +.EMIT_BRLOOP_CODE: + EMIT_QWORD .EMIT_BRLOOP_CODE_END, { mov nbits, code_temp } + ; insert code, flush buffer, + ; nbits = code_temp, goto .EMIT_BRLOOP_CODE_END + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcphuff-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcphuff-sse2.asm new file mode 100644 index 0000000..c9ac59f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcphuff-sse2.asm @@ -0,0 +1,629 @@ +; +; jcphuff-sse2.asm - prepare data for progressive Huffman encoding +; (64-bit SSE2) +; +; Copyright (C) 2016, 2018, Matthieu Darbois +; Copyright (C) 2023, Aliaksiej Kandracienka. +; Copyright (C) 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains an SSE2 implementation of data preparation for progressive +; Huffman encoding. See jcphuff.c for more details. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +; -------------------------------------------------------------------------- +; Macros to load data for jsimd_encode_mcu_AC_first_prepare_sse2() and +; jsimd_encode_mcu_AC_refine_prepare_sse2() + +%macro LOAD16 0 + pxor N0, N0 + pxor N1, N1 + + mov T0d, INT [LUT + 0*SIZEOF_INT] + mov T1d, INT [LUT + 8*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 0 + pinsrw X1, word [BLOCK + T1 * 2], 0 + + mov T0d, INT [LUT + 1*SIZEOF_INT] + mov T1d, INT [LUT + 9*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 1 + pinsrw X1, word [BLOCK + T1 * 2], 1 + + mov T0d, INT [LUT + 2*SIZEOF_INT] + mov T1d, INT [LUT + 10*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 2 + pinsrw X1, word [BLOCK + T1 * 2], 2 + + mov T0d, INT [LUT + 3*SIZEOF_INT] + mov T1d, INT [LUT + 11*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 3 + pinsrw X1, word [BLOCK + T1 * 2], 3 + + mov T0d, INT [LUT + 4*SIZEOF_INT] + mov T1d, INT [LUT + 12*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 4 + pinsrw X1, word [BLOCK + T1 * 2], 4 + + mov T0d, INT [LUT + 5*SIZEOF_INT] + mov T1d, INT [LUT + 13*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 5 + pinsrw X1, word [BLOCK + T1 * 2], 5 + + mov T0d, INT [LUT + 6*SIZEOF_INT] + mov T1d, INT [LUT + 14*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 6 + pinsrw X1, word [BLOCK + T1 * 2], 6 + + mov T0d, INT [LUT + 7*SIZEOF_INT] + mov T1d, INT [LUT + 15*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 7 + pinsrw X1, word [BLOCK + T1 * 2], 7 +%endmacro + +%macro LOAD15 0 + pxor N0, N0 + pxor N1, N1 + pxor X1, X1 + + mov T0d, INT [LUT + 0*SIZEOF_INT] + mov T1d, INT [LUT + 8*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 0 + pinsrw X1, word [BLOCK + T1 * 2], 0 + + mov T0d, INT [LUT + 1*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 1 + + mov T0d, INT [LUT + 2*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 2 + + mov T0d, INT [LUT + 3*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 3 + + mov T0d, INT [LUT + 4*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 4 + + mov T0d, INT [LUT + 5*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 5 + + mov T0d, INT [LUT + 6*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 6 + + mov T0d, INT [LUT + 7*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 7 + + cmp LENEND, 2 + jl %%.ELOAD15 + mov T1d, INT [LUT + 9*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 1 + + cmp LENEND, 3 + jl %%.ELOAD15 + mov T1d, INT [LUT + 10*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 2 + + cmp LENEND, 4 + jl %%.ELOAD15 + mov T1d, INT [LUT + 11*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 3 + + cmp LENEND, 5 + jl %%.ELOAD15 + mov T1d, INT [LUT + 12*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 4 + + cmp LENEND, 6 + jl %%.ELOAD15 + mov T1d, INT [LUT + 13*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 5 + + cmp LENEND, 7 + jl %%.ELOAD15 + mov T1d, INT [LUT + 14*SIZEOF_INT] + pinsrw X1, word [BLOCK + T1 * 2], 6 +%%.ELOAD15: +%endmacro + +%macro LOAD8 0 + pxor N0, N0 + + mov T0d, INT [LUT + 0*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 0 + + mov T0d, INT [LUT + 1*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 1 + + mov T0d, INT [LUT + 2*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 2 + + mov T0d, INT [LUT + 3*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 3 + + mov T0d, INT [LUT + 4*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 4 + + mov T0d, INT [LUT + 5*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 5 + + mov T0d, INT [LUT + 6*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 6 + + mov T0d, INT [LUT + 7*SIZEOF_INT] + pinsrw X0, word [BLOCK + T0 * 2], 7 +%endmacro + +%macro LOAD7 0 + pxor N0, N0 + pxor X0, X0 + + mov T1d, INT [LUT + 0*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 0 + + cmp LENEND, 2 + jl %%.ELOAD7 + mov T1d, INT [LUT + 1*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 1 + + cmp LENEND, 3 + jl %%.ELOAD7 + mov T1d, INT [LUT + 2*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 2 + + cmp LENEND, 4 + jl %%.ELOAD7 + mov T1d, INT [LUT + 3*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 3 + + cmp LENEND, 5 + jl %%.ELOAD7 + mov T1d, INT [LUT + 4*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 4 + + cmp LENEND, 6 + jl %%.ELOAD7 + mov T1d, INT [LUT + 5*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 5 + + cmp LENEND, 7 + jl %%.ELOAD7 + mov T1d, INT [LUT + 6*SIZEOF_INT] + pinsrw X0, word [BLOCK + T1 * 2], 6 +%%.ELOAD7: +%endmacro + +%macro REDUCE0 0 + movdqa xmm0, XMMWORD [VALUES + ( 0*2)] + movdqa xmm1, XMMWORD [VALUES + ( 8*2)] + movdqa xmm2, XMMWORD [VALUES + (16*2)] + movdqa xmm3, XMMWORD [VALUES + (24*2)] + movdqa xmm4, XMMWORD [VALUES + (32*2)] + movdqa xmm5, XMMWORD [VALUES + (40*2)] + movdqa xmm6, XMMWORD [VALUES + (48*2)] + movdqa xmm7, XMMWORD [VALUES + (56*2)] + + pcmpeqw xmm0, ZERO + pcmpeqw xmm1, ZERO + pcmpeqw xmm2, ZERO + pcmpeqw xmm3, ZERO + pcmpeqw xmm4, ZERO + pcmpeqw xmm5, ZERO + pcmpeqw xmm6, ZERO + pcmpeqw xmm7, ZERO + + packsswb xmm0, xmm1 + packsswb xmm2, xmm3 + packsswb xmm4, xmm5 + packsswb xmm6, xmm7 + + pmovmskb eax, xmm0 + pmovmskb ecx, xmm2 + pmovmskb edx, xmm4 + pmovmskb esi, xmm6 + + shl rcx, 16 + shl rdx, 32 + shl rsi, 48 + + or rax, rcx + or rdx, rsi + or rax, rdx + + not rax + + mov MMWORD [r15], rax +%endmacro + +; +; Prepare data for jsimd_encode_mcu_AC_first(). +; +; GLOBAL(void) +; jsimd_encode_mcu_AC_first_prepare_sse2(const JCOEF *block, +; const int *jpeg_natural_order_start, +; int Sl, int Al, JCOEF *values, +; size_t *zerobits) +; +; r10 = const JCOEF *block +; r11 = const int *jpeg_natural_order_start +; r12 = int Sl +; r13 = int Al +; r14 = JCOEF *values +; r15 = size_t *zerobits + +%define ZERO xmm9 +%define X0 xmm0 +%define X1 xmm1 +%define N0 xmm2 +%define N1 xmm3 +%define AL xmm4 +%define K eax +%define LUT r11 +%define T0 rcx +%define T0d ecx +%define T1 rdx +%define T1d edx +%define BLOCK r10 +%define VALUES r14 +%define LEN r12d +%define LENEND r13d + + align 32 + GLOBAL_FUNCTION(jsimd_encode_mcu_AC_first_prepare_sse2) + +EXTN(jsimd_encode_mcu_AC_first_prepare_sse2): + ENDBR64 + push rbp + mov rbp, rsp + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + sub rsp, SIZEOF_XMMWORD + movdqa XMMWORD [rsp], ZERO + COLLECT_ARGS 6 + + movd AL, r13d + pxor ZERO, ZERO + mov K, LEN + mov LENEND, LEN + and K, -16 + and LENEND, 7 + shr K, 4 + jz .ELOOP16 +.BLOOP16: + LOAD16 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + pxor N0, X0 + pxor N1, X1 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + movdqa XMMWORD [VALUES + (8 + DCTSIZE2) * 2], N1 + add VALUES, 16*2 + add LUT, 16*SIZEOF_INT + dec K + jnz .BLOOP16 + test LEN, 15 + je .PADDING +.ELOOP16: + test LEN, 8 + jz .TRY7 + test LEN, 7 + jz .TRY8 + + LOAD15 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + pxor N0, X0 + pxor N1, X1 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + movdqa XMMWORD [VALUES + (8 + DCTSIZE2) * 2], N1 + add VALUES, 16*2 + jmp .PADDING +.TRY8: + LOAD8 + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + pxor N0, X0 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + add VALUES, 8*2 + jmp .PADDING +.TRY7: + LOAD7 + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + pxor N0, X0 + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (0 + DCTSIZE2) * 2], N0 + add VALUES, 8*2 +.PADDING: + mov K, LEN + add K, 7 + and K, -8 + shr K, 3 + sub K, DCTSIZE2/8 + jz .EPADDING + align 16 +.ZEROLOOP: + movdqa XMMWORD [VALUES + 0], ZERO + add VALUES, 8*2 + inc K + jnz .ZEROLOOP +.EPADDING: + sub VALUES, DCTSIZE2*2 + + REDUCE0 + + UNCOLLECT_ARGS 6 + movdqa ZERO, XMMWORD [rsp] + mov rsp, rbp + pop rbp + ret + +%undef ZERO +%undef X0 +%undef X1 +%undef N0 +%undef N1 +%undef AL +%undef K +%undef LUT +%undef T0 +%undef T0d +%undef T1 +%undef T1d +%undef BLOCK +%undef VALUES +%undef LEN +%undef LENEND + +; +; Prepare data for jsimd_encode_mcu_AC_refine(). +; +; GLOBAL(int) +; jsimd_encode_mcu_AC_refine_prepare_sse2(const JCOEF *block, +; const int *jpeg_natural_order_start, +; int Sl, int Al, JCOEF *absvalues, +; size_t *bits) +; +; r10 = const JCOEF *block +; r11 = const int *jpeg_natural_order_start +; r12 = int Sl +; r13 = int Al +; r14 = JCOEF *values +; r15 = size_t *bits + +%define ZERO xmm9 +%define ONE xmm5 +%define X0 xmm0 +%define X1 xmm1 +%define N0 xmm2 +%define N1 xmm3 +%define AL xmm4 +%define K eax +%define KK r9d +%define EOB r8d +%define SIGN rdi +%define LUT r11 +%define T0 rcx +%define T0d ecx +%define T1 rdx +%define T1d edx +%define BLOCK r10 +%define VALUES r14 +%define LEN r12d +%define LENEND r13d + + align 32 + GLOBAL_FUNCTION(jsimd_encode_mcu_AC_refine_prepare_sse2) + +EXTN(jsimd_encode_mcu_AC_refine_prepare_sse2): + ENDBR64 + push rbp + mov rbp, rsp + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + sub rsp, SIZEOF_XMMWORD + movdqa XMMWORD [rsp], ZERO + COLLECT_ARGS 6 + + xor SIGN, SIGN + xor EOB, EOB + xor KK, KK + movd AL, r13d + pxor ZERO, ZERO + pcmpeqw ONE, ONE + psrlw ONE, 15 + mov K, LEN + mov LENEND, LEN + and K, -16 + and LENEND, 7 + shr K, 4 + jz .ELOOPR16 +.BLOOPR16: + LOAD16 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + pcmpeqw X0, ONE + pcmpeqw X1, ONE + packsswb N0, N1 + packsswb X0, X1 + pmovmskb T0d, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + pmovmskb T1d, X0 ; idx = _mm_movemask_epi8(x1); + shr SIGN, 16 ; make room for sizebits + shl T0, 48 + or SIGN, T0 + bsr T1d, T1d ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER16 ; if (idx) { + mov EOB, KK + add EOB, T1d ; EOB = k + idx; +.CONTINUER16: + add VALUES, 16*2 + add LUT, 16*SIZEOF_INT + add KK, 16 + dec K + jnz .BLOOPR16 + test LEN, 15 + je .PADDINGR +.ELOOPR16: + test LEN, 8 + jz .TRYR7 + test LEN, 7 + jz .TRYR8 + + LOAD15 + pcmpgtw N0, X0 + pcmpgtw N1, X1 + paddw X0, N0 + paddw X1, N1 + pxor X0, N0 + pxor X1, N1 + psrlw X0, AL + psrlw X1, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + movdqa XMMWORD [VALUES + (8) * 2], X1 + pcmpeqw X0, ONE + pcmpeqw X1, ONE + packsswb N0, N1 + packsswb X0, X1 + pmovmskb T0d, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + pmovmskb T1d, X0 ; idx = _mm_movemask_epi8(x1); + shr SIGN, 16 ; make room for sizebits + shl T0, 48 + or SIGN, T0 + bsr T1d, T1d ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER15 ; if (idx) { + mov EOB, KK + add EOB, T1d ; EOB = k + idx; +.CONTINUER15: + add VALUES, 16*2 + jmp .PADDINGR +.TRYR8: + LOAD8 + + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + pcmpeqw X0, ONE + packsswb N0, ZERO + packsswb X0, ZERO + pmovmskb T0d, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + pmovmskb T1d, X0 ; idx = _mm_movemask_epi8(x1); + shr SIGN, 8 ; make room for sizebits + shl T0, 56 + or SIGN, T0 + bsr T1d, T1d ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER8 ; if (idx) { + mov EOB, KK + add EOB, T1d ; EOB = k + idx; +.CONTINUER8: + add VALUES, 8*2 + jmp .PADDINGR +.TRYR7: + LOAD7 + + pcmpgtw N0, X0 + paddw X0, N0 + pxor X0, N0 + psrlw X0, AL + movdqa XMMWORD [VALUES + (0) * 2], X0 + pcmpeqw X0, ONE + packsswb N0, ZERO + packsswb X0, ZERO + pmovmskb T0d, N0 ; lsignbits.val16u[k>>4] = _mm_movemask_epi8(neg); + pmovmskb T1d, X0 ; idx = _mm_movemask_epi8(x1); + shr SIGN, 8 ; make room for sizebits + shl T0, 56 + or SIGN, T0 + bsr T1d, T1d ; idx = 16 - (__builtin_clz(idx)>>1); + jz .CONTINUER7 ; if (idx) { + mov EOB, KK + add EOB, T1d ; EOB = k + idx; +.CONTINUER7: + add VALUES, 8*2 +.PADDINGR: + mov K, LEN + add K, 7 + and K, -8 + shr K, 3 + sub K, DCTSIZE2/8 + jz .EPADDINGR + align 16 +.ZEROLOOPR: + movdqa XMMWORD [VALUES + 0], ZERO + shr SIGN, 8 + add VALUES, 8*2 + inc K + jnz .ZEROLOOPR +.EPADDINGR: + not SIGN + sub VALUES, DCTSIZE2*2 + mov MMWORD [r15+SIZEOF_MMWORD], SIGN + + REDUCE0 + + mov eax, EOB + UNCOLLECT_ARGS 6 + movdqa ZERO, XMMWORD [rsp] + mov rsp, rbp + pop rbp + ret + +%undef ZERO +%undef ONE +%undef X0 +%undef X1 +%undef N0 +%undef N1 +%undef AL +%undef K +%undef KK +%undef EOB +%undef SIGN +%undef LUT +%undef T0 +%undef T0d +%undef T1 +%undef T1d +%undef BLOCK +%undef VALUES +%undef LEN +%undef LENEND + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-avx2.asm new file mode 100644 index 0000000..53afc7d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-avx2.asm @@ -0,0 +1,363 @@ +; +; jcsample.asm - downsampling (64-bit AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2018, Matthias Räncker. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Downsample pixel values of a single component. +; This version handles the common case of 2:1 horizontal and 1:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v1_downsample_avx2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +; r10d = JDIMENSION image_width +; r11 = int max_v_samp_factor +; r12d = JDIMENSION v_samp_factor +; r13d = JDIMENSION width_in_blocks +; r14 = JSAMPARRAY input_data +; r15 = JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_downsample_avx2) + +EXTN(jsimd_h2v1_downsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 6 + + mov ecx, r13d + shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols) + jz near .return + + mov edx, r10d + + ; -- expand_right_edge + + push rcx + shl rcx, 1 ; output_cols * 2 + sub rcx, rdx + jle short .expand_end + + mov rax, r11 + test rax, rax + jle short .expand_end + + cld + mov rsi, r14 ; input_data +.expandloop: + push rax + push rcx + + mov rdip, JSAMPROW [rsi] + add rdi, rdx + mov al, JSAMPLE [rdi-1] + + rep stosb + + pop rcx + pop rax + + add rsi, byte SIZEOF_JSAMPROW + dec rax + jg short .expandloop + +.expand_end: + pop rcx ; output_cols + + ; -- h2v1_downsample + + mov eax, r12d ; rowctr + test eax, eax + jle near .return + + mov rdx, 0x00010000 ; bias pattern + vmovd xmm7, edx + vpshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1} + vperm2i128 ymm7, ymm7, ymm7, 0 ; ymm7={xmm7, xmm7} + vpcmpeqw ymm6, ymm6, ymm6 + vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..} + + mov rsi, r14 ; input_data + mov rdi, r15 ; output_data +.rowloop: + push rcx + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr + + cmp rcx, byte SIZEOF_YMMWORD + jae short .columnloop + +.columnloop_r24: + ; rcx can possibly be 8, 16, 24 + cmp rcx, 24 + jne .columnloop_r16 + vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu xmm1, XMMWORD [rsi+1*SIZEOF_YMMWORD] + mov rcx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r16: + cmp rcx, 16 + jne .columnloop_r8 + vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vpxor ymm1, ymm1, ymm1 + mov rcx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r8: + vmovdqu xmm0, XMMWORD[rsi+0*SIZEOF_YMMWORD] + vpxor ymm1, ymm1, ymm1 + mov rcx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop: + vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [rsi+1*SIZEOF_YMMWORD] + +.downsample: + vpsrlw ymm2, ymm0, BYTE_BIT + vpand ymm0, ymm0, ymm6 + vpsrlw ymm3, ymm1, BYTE_BIT + vpand ymm1, ymm1, ymm6 + + vpaddw ymm0, ymm0, ymm2 + vpaddw ymm1, ymm1, ymm3 + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm1, ymm1, ymm7 + vpsrlw ymm0, ymm0, 1 + vpsrlw ymm1, ymm1, 1 + + vpackuswb ymm0, ymm0, ymm1 + vpermq ymm0, ymm0, 0xd8 + + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0 + + sub rcx, byte SIZEOF_YMMWORD ; outcol + add rsi, byte 2*SIZEOF_YMMWORD ; inptr + add rdi, byte 1*SIZEOF_YMMWORD ; outptr + cmp rcx, byte SIZEOF_YMMWORD + jae short .columnloop + test rcx, rcx + jnz near .columnloop_r24 + + pop rsi + pop rdi + pop rcx + + add rsi, byte SIZEOF_JSAMPROW ; input_data + add rdi, byte SIZEOF_JSAMPROW ; output_data + dec rax ; rowctr + jg near .rowloop + +.return: + vzeroupper + UNCOLLECT_ARGS 6 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Downsample pixel values of a single component. +; This version handles the standard case of 2:1 horizontal and 2:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v2_downsample_avx2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +; r10d = JDIMENSION image_width +; r11 = int max_v_samp_factor +; r12d = JDIMENSION v_samp_factor +; r13d = JDIMENSION width_in_blocks +; r14 = JSAMPARRAY input_data +; r15 = JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_downsample_avx2) + +EXTN(jsimd_h2v2_downsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 6 + + mov ecx, r13d + shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols) + jz near .return + + mov edx, r10d + + ; -- expand_right_edge + + push rcx + shl rcx, 1 ; output_cols * 2 + sub rcx, rdx + jle short .expand_end + + mov rax, r11 + test rax, rax + jle short .expand_end + + cld + mov rsi, r14 ; input_data +.expandloop: + push rax + push rcx + + mov rdip, JSAMPROW [rsi] + add rdi, rdx + mov al, JSAMPLE [rdi-1] + + rep stosb + + pop rcx + pop rax + + add rsi, byte SIZEOF_JSAMPROW + dec rax + jg short .expandloop + +.expand_end: + pop rcx ; output_cols + + ; -- h2v2_downsample + + mov eax, r12d ; rowctr + test rax, rax + jle near .return + + mov rdx, 0x00020001 ; bias pattern + vmovd xmm7, edx + vpcmpeqw ymm6, ymm6, ymm6 + vpshufd xmm7, xmm7, 0x00 ; ymm7={1, 2, 1, 2, 1, 2, 1, 2} + vperm2i128 ymm7, ymm7, ymm7, 0 + vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..} + + mov rsi, r14 ; input_data + mov rdi, r15 ; output_data +.rowloop: + push rcx + push rdi + push rsi + + mov rdxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0 + mov rsip, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1 + mov rdip, JSAMPROW [rdi] ; outptr + + cmp rcx, byte SIZEOF_YMMWORD + jae short .columnloop + +.columnloop_r24: + cmp rcx, 24 + jne .columnloop_r16 + vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu xmm2, XMMWORD [rdx+1*SIZEOF_YMMWORD] + vmovdqu xmm3, XMMWORD [rsi+1*SIZEOF_YMMWORD] + mov rcx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r16: + cmp rcx, 16 + jne .columnloop_r8 + vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vpxor ymm2, ymm2, ymm2 + vpxor ymm3, ymm3, ymm3 + mov rcx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop_r8: + vmovdqu xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD] + vmovdqu xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD] + vpxor ymm2, ymm2, ymm2 + vpxor ymm3, ymm3, ymm3 + mov rcx, SIZEOF_YMMWORD + jmp short .downsample + +.columnloop: + vmovdqu ymm0, YMMWORD [rdx+0*SIZEOF_YMMWORD] + vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD] + vmovdqu ymm2, YMMWORD [rdx+1*SIZEOF_YMMWORD] + vmovdqu ymm3, YMMWORD [rsi+1*SIZEOF_YMMWORD] + +.downsample: + vpand ymm4, ymm0, ymm6 + vpsrlw ymm0, ymm0, BYTE_BIT + vpand ymm5, ymm1, ymm6 + vpsrlw ymm1, ymm1, BYTE_BIT + vpaddw ymm0, ymm0, ymm4 + vpaddw ymm1, ymm1, ymm5 + + vpand ymm4, ymm2, ymm6 + vpsrlw ymm2, ymm2, BYTE_BIT + vpand ymm5, ymm3, ymm6 + vpsrlw ymm3, ymm3, BYTE_BIT + vpaddw ymm2, ymm2, ymm4 + vpaddw ymm3, ymm3, ymm5 + + vpaddw ymm0, ymm0, ymm1 + vpaddw ymm2, ymm2, ymm3 + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm2, ymm2, ymm7 + vpsrlw ymm0, ymm0, 2 + vpsrlw ymm2, ymm2, 2 + + vpackuswb ymm0, ymm0, ymm2 + vpermq ymm0, ymm0, 0xd8 + + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0 + + sub rcx, byte SIZEOF_YMMWORD ; outcol + add rdx, byte 2*SIZEOF_YMMWORD ; inptr0 + add rsi, byte 2*SIZEOF_YMMWORD ; inptr1 + add rdi, byte 1*SIZEOF_YMMWORD ; outptr + cmp rcx, byte SIZEOF_YMMWORD + jae near .columnloop + test rcx, rcx + jnz near .columnloop_r24 + + pop rsi + pop rdi + pop rcx + + add rsi, byte 2*SIZEOF_JSAMPROW ; input_data + add rdi, byte 1*SIZEOF_JSAMPROW ; output_data + dec rax ; rowctr + jg near .rowloop + +.return: + vzeroupper + UNCOLLECT_ARGS 6 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-sse2.asm new file mode 100644 index 0000000..d7ffa93 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jcsample-sse2.asm @@ -0,0 +1,326 @@ +; +; jcsample.asm - downsampling (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Downsample pixel values of a single component. +; This version handles the common case of 2:1 horizontal and 1:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v1_downsample_sse2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +; r10d = JDIMENSION image_width +; r11 = int max_v_samp_factor +; r12d = JDIMENSION v_samp_factor +; r13d = JDIMENSION width_in_blocks +; r14 = JSAMPARRAY input_data +; r15 = JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_downsample_sse2) + +EXTN(jsimd_h2v1_downsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 6 + + mov ecx, r13d + shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols) + jz near .return + + mov edx, r10d + + ; -- expand_right_edge + + push rcx + shl rcx, 1 ; output_cols * 2 + sub rcx, rdx + jle short .expand_end + + mov rax, r11 + test rax, rax + jle short .expand_end + + cld + mov rsi, r14 ; input_data +.expandloop: + push rax + push rcx + + mov rdip, JSAMPROW [rsi] + add rdi, rdx + mov al, JSAMPLE [rdi-1] + + rep stosb + + pop rcx + pop rax + + add rsi, byte SIZEOF_JSAMPROW + dec rax + jg short .expandloop + +.expand_end: + pop rcx ; output_cols + + ; -- h2v1_downsample + + mov eax, r12d ; rowctr + test eax, eax + jle near .return + + mov rdx, 0x00010000 ; bias pattern + movd xmm7, edx + pcmpeqw xmm6, xmm6 + pshufd xmm7, xmm7, 0x00 ; xmm7={0, 1, 0, 1, 0, 1, 0, 1} + psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..} + + mov rsi, r14 ; input_data + mov rdi, r15 ; output_data +.rowloop: + push rcx + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr + + cmp rcx, byte SIZEOF_XMMWORD + jae short .columnloop + +.columnloop_r8: + movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD] + pxor xmm1, xmm1 + mov rcx, SIZEOF_XMMWORD + jmp short .downsample + +.columnloop: + movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqa xmm1, XMMWORD [rsi+1*SIZEOF_XMMWORD] + +.downsample: + movdqa xmm2, xmm0 + movdqa xmm3, xmm1 + + pand xmm0, xmm6 + psrlw xmm2, BYTE_BIT + pand xmm1, xmm6 + psrlw xmm3, BYTE_BIT + + paddw xmm0, xmm2 + paddw xmm1, xmm3 + paddw xmm0, xmm7 + paddw xmm1, xmm7 + psrlw xmm0, 1 + psrlw xmm1, 1 + + packuswb xmm0, xmm1 + + movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0 + + sub rcx, byte SIZEOF_XMMWORD ; outcol + add rsi, byte 2*SIZEOF_XMMWORD ; inptr + add rdi, byte 1*SIZEOF_XMMWORD ; outptr + cmp rcx, byte SIZEOF_XMMWORD + jae short .columnloop + test rcx, rcx + jnz short .columnloop_r8 + + pop rsi + pop rdi + pop rcx + + add rsi, byte SIZEOF_JSAMPROW ; input_data + add rdi, byte SIZEOF_JSAMPROW ; output_data + dec rax ; rowctr + jg near .rowloop + +.return: + UNCOLLECT_ARGS 6 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Downsample pixel values of a single component. +; This version handles the standard case of 2:1 horizontal and 2:1 vertical, +; without smoothing. +; +; GLOBAL(void) +; jsimd_h2v2_downsample_sse2(JDIMENSION image_width, int max_v_samp_factor, +; JDIMENSION v_samp_factor, +; JDIMENSION width_in_blocks, JSAMPARRAY input_data, +; JSAMPARRAY output_data); +; + +; r10d = JDIMENSION image_width +; r11 = int max_v_samp_factor +; r12d = JDIMENSION v_samp_factor +; r13d = JDIMENSION width_in_blocks +; r14 = JSAMPARRAY input_data +; r15 = JSAMPARRAY output_data + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_downsample_sse2) + +EXTN(jsimd_h2v2_downsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 6 + + mov ecx, r13d + shl rcx, 3 ; imul rcx,DCTSIZE (rcx = output_cols) + jz near .return + + mov edx, r10d + + ; -- expand_right_edge + + push rcx + shl rcx, 1 ; output_cols * 2 + sub rcx, rdx + jle short .expand_end + + mov rax, r11 + test rax, rax + jle short .expand_end + + cld + mov rsi, r14 ; input_data +.expandloop: + push rax + push rcx + + mov rdip, JSAMPROW [rsi] + add rdi, rdx + mov al, JSAMPLE [rdi-1] + + rep stosb + + pop rcx + pop rax + + add rsi, byte SIZEOF_JSAMPROW + dec rax + jg short .expandloop + +.expand_end: + pop rcx ; output_cols + + ; -- h2v2_downsample + + mov eax, r12d ; rowctr + test rax, rax + jle near .return + + mov rdx, 0x00020001 ; bias pattern + movd xmm7, edx + pcmpeqw xmm6, xmm6 + pshufd xmm7, xmm7, 0x00 ; xmm7={1, 2, 1, 2, 1, 2, 1, 2} + psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..} + + mov rsi, r14 ; input_data + mov rdi, r15 ; output_data +.rowloop: + push rcx + push rdi + push rsi + + mov rdxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0 + mov rsip, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1 + mov rdip, JSAMPROW [rdi] ; outptr + + cmp rcx, byte SIZEOF_XMMWORD + jae short .columnloop + +.columnloop_r8: + movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD] + movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD] + pxor xmm2, xmm2 + pxor xmm3, xmm3 + mov rcx, SIZEOF_XMMWORD + jmp short .downsample + +.columnloop: + movdqa xmm0, XMMWORD [rdx+0*SIZEOF_XMMWORD] + movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqa xmm2, XMMWORD [rdx+1*SIZEOF_XMMWORD] + movdqa xmm3, XMMWORD [rsi+1*SIZEOF_XMMWORD] + +.downsample: + movdqa xmm4, xmm0 + movdqa xmm5, xmm1 + pand xmm0, xmm6 + psrlw xmm4, BYTE_BIT + pand xmm1, xmm6 + psrlw xmm5, BYTE_BIT + paddw xmm0, xmm4 + paddw xmm1, xmm5 + + movdqa xmm4, xmm2 + movdqa xmm5, xmm3 + pand xmm2, xmm6 + psrlw xmm4, BYTE_BIT + pand xmm3, xmm6 + psrlw xmm5, BYTE_BIT + paddw xmm2, xmm4 + paddw xmm3, xmm5 + + paddw xmm0, xmm1 + paddw xmm2, xmm3 + paddw xmm0, xmm7 + paddw xmm2, xmm7 + psrlw xmm0, 2 + psrlw xmm2, 2 + + packuswb xmm0, xmm2 + + movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0 + + sub rcx, byte SIZEOF_XMMWORD ; outcol + add rdx, byte 2*SIZEOF_XMMWORD ; inptr0 + add rsi, byte 2*SIZEOF_XMMWORD ; inptr1 + add rdi, byte 1*SIZEOF_XMMWORD ; outptr + cmp rcx, byte SIZEOF_XMMWORD + jae near .columnloop + test rcx, rcx + jnz near .columnloop_r8 + + pop rsi + pop rdi + pop rcx + + add rsi, byte 2*SIZEOF_JSAMPROW ; input_data + add rdi, byte 1*SIZEOF_JSAMPROW ; output_data + dec rax ; rowctr + jg near .rowloop + +.return: + UNCOLLECT_ARGS 6 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-avx2.asm new file mode 100644 index 0000000..7b8a084 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-avx2.asm @@ -0,0 +1,494 @@ +; +; jdcolext.asm - colorspace conversion (64-bit AVX2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2012, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_ycc_rgb_convert_avx2(JDIMENSION out_width, JSAMPIMAGE input_buf, +; JDIMENSION input_row, JSAMPARRAY output_buf, +; int num_rows) +; + +; r10d = JDIMENSION out_width +; r11 = JSAMPIMAGE input_buf +; r12d = JDIMENSION input_row +; r13 = JSAMPARRAY output_buf +; r14d = int num_rows + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_YMMWORD ; ymmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_ycc_rgb_convert_avx2) + +EXTN(jsimd_ycc_rgb_convert_avx2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (WK_NUM * SIZEOF_YMMWORD) + COLLECT_ARGS 5 + push rbx + + mov ecx, r10d ; num_cols + test rcx, rcx + jz near .return + + push rcx + + mov rdi, r11 + mov ecx, r12d + mov rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] + lea rsi, [rsi+rcx*SIZEOF_JSAMPROW] + lea rbx, [rbx+rcx*SIZEOF_JSAMPROW] + lea rdx, [rdx+rcx*SIZEOF_JSAMPROW] + + pop rcx + + mov rdi, r13 + mov eax, r14d + test rax, rax + jle near .return +.rowloop: + push rax + push rdi + push rdx + push rbx + push rsi + push rcx ; col + + mov rsip, JSAMPROW [rsi] ; inptr0 + mov rbxp, JSAMPROW [rbx] ; inptr1 + mov rdxp, JSAMPROW [rdx] ; inptr2 + mov rdip, JSAMPROW [rdi] ; outptr +.columnloop: + + vmovdqu ymm5, YMMWORD [rbx] ; ymm5=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV) + vmovdqu ymm1, YMMWORD [rdx] ; ymm1=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpcmpeqw ymm0, ymm0, ymm0 + vpcmpeqw ymm7, ymm7, ymm7 + vpsrlw ymm0, ymm0, BYTE_BIT ; ymm0={0xFF 0x00 0xFF 0x00 ..} + vpsllw ymm7, ymm7, 7 ; ymm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + vpand ymm4, ymm0, ymm5 ; ymm4=Cb(02468ACEGIKMOQSU)=CbE + vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Cb(13579BDFHJLNPRTV)=CbO + vpand ymm0, ymm0, ymm1 ; ymm0=Cr(02468ACEGIKMOQSU)=CrE + vpsrlw ymm1, ymm1, BYTE_BIT ; ymm1=Cr(13579BDFHJLNPRTV)=CrO + + vpaddw ymm2, ymm4, ymm7 + vpaddw ymm3, ymm5, ymm7 + vpaddw ymm6, ymm0, ymm7 + vpaddw ymm7, ymm1, ymm7 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbE + vpaddw ymm5, ymm3, ymm3 ; ymm5=2*CbO + vpaddw ymm0, ymm6, ymm6 ; ymm0=2*CrE + vpaddw ymm1, ymm7, ymm7 ; ymm1=2*CrO + + vpmulhw ymm4, ymm4, [rel PW_MF0228] ; ymm4=(2*CbE * -FIX(0.22800)) + vpmulhw ymm5, ymm5, [rel PW_MF0228] ; ymm5=(2*CbO * -FIX(0.22800)) + vpmulhw ymm0, ymm0, [rel PW_F0402] ; ymm0=(2*CrE * FIX(0.40200)) + vpmulhw ymm1, ymm1, [rel PW_F0402] ; ymm1=(2*CrO * FIX(0.40200)) + + vpaddw ymm4, ymm4, [rel PW_ONE] + vpaddw ymm5, ymm5, [rel PW_ONE] + vpsraw ymm4, ymm4, 1 ; ymm4=(CbE * -FIX(0.22800)) + vpsraw ymm5, ymm5, 1 ; ymm5=(CbO * -FIX(0.22800)) + vpaddw ymm0, ymm0, [rel PW_ONE] + vpaddw ymm1, ymm1, [rel PW_ONE] + vpsraw ymm0, ymm0, 1 ; ymm0=(CrE * FIX(0.40200)) + vpsraw ymm1, ymm1, 1 ; ymm1=(CrO * FIX(0.40200)) + + vpaddw ymm4, ymm4, ymm2 + vpaddw ymm5, ymm5, ymm3 + vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbE * FIX(1.77200))=(B-Y)E + vpaddw ymm5, ymm5, ymm3 ; ymm5=(CbO * FIX(1.77200))=(B-Y)O + vpaddw ymm0, ymm0, ymm6 ; ymm0=(CrE * FIX(1.40200))=(R-Y)E + vpaddw ymm1, ymm1, ymm7 ; ymm1=(CrO * FIX(1.40200))=(R-Y)O + + vmovdqa YMMWORD [wk(0)], ymm4 ; wk(0)=(B-Y)E + vmovdqa YMMWORD [wk(1)], ymm5 ; wk(1)=(B-Y)O + + vpunpckhwd ymm4, ymm2, ymm6 + vpunpcklwd ymm2, ymm2, ymm6 + vpmaddwd ymm2, ymm2, [rel PW_MF0344_F0285] + vpmaddwd ymm4, ymm4, [rel PW_MF0344_F0285] + vpunpckhwd ymm5, ymm3, ymm7 + vpunpcklwd ymm3, ymm3, ymm7 + vpmaddwd ymm3, ymm3, [rel PW_MF0344_F0285] + vpmaddwd ymm5, ymm5, [rel PW_MF0344_F0285] + + vpaddd ymm2, ymm2, [rel PD_ONEHALF] + vpaddd ymm4, ymm4, [rel PD_ONEHALF] + vpsrad ymm2, ymm2, SCALEBITS + vpsrad ymm4, ymm4, SCALEBITS + vpaddd ymm3, ymm3, [rel PD_ONEHALF] + vpaddd ymm5, ymm5, [rel PD_ONEHALF] + vpsrad ymm3, ymm3, SCALEBITS + vpsrad ymm5, ymm5, SCALEBITS + + vpackssdw ymm2, ymm2, ymm4 ; ymm2=CbE*-FIX(0.344)+CrE*FIX(0.285) + vpackssdw ymm3, ymm3, ymm5 ; ymm3=CbO*-FIX(0.344)+CrO*FIX(0.285) + vpsubw ymm2, ymm2, ymm6 ; ymm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E + vpsubw ymm3, ymm3, ymm7 ; ymm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O + + vmovdqu ymm5, YMMWORD [rsi] ; ymm5=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpcmpeqw ymm4, ymm4, ymm4 + vpsrlw ymm4, ymm4, BYTE_BIT ; ymm4={0xFF 0x00 0xFF 0x00 ..} + vpand ymm4, ymm4, ymm5 ; ymm4=Y(02468ACEGIKMOQSU)=YE + vpsrlw ymm5, ymm5, BYTE_BIT ; ymm5=Y(13579BDFHJLNPRTV)=YO + + vpaddw ymm0, ymm0, ymm4 ; ymm0=((R-Y)E+YE)=RE=R(02468ACEGIKMOQSU) + vpaddw ymm1, ymm1, ymm5 ; ymm1=((R-Y)O+YO)=RO=R(13579BDFHJLNPRTV) + vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********) + vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********) + + vpaddw ymm2, ymm2, ymm4 ; ymm2=((G-Y)E+YE)=GE=G(02468ACEGIKMOQSU) + vpaddw ymm3, ymm3, ymm5 ; ymm3=((G-Y)O+YO)=GO=G(13579BDFHJLNPRTV) + vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********) + vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********) + + vpaddw ymm4, ymm4, YMMWORD [wk(0)] ; ymm4=(YE+(B-Y)E)=BE=B(02468ACEGIKMOQSU) + vpaddw ymm5, ymm5, YMMWORD [wk(1)] ; ymm5=(YO+(B-Y)O)=BO=B(13579BDFHJLNPRTV) + vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********) + vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F + ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V) + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F + ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V) + + vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G + ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --) + vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F + ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07 + ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N) + + vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H + ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --) + + vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H + ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --) + vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G + ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --) + vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18 + ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O) + + vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H + ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --) + vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29 + ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P) + + vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03 + ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J) + vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14 + ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K) + vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29 + ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P) + vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07 + ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N) + + vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B + ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R) + vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C + ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S) + vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H + ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --) + vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V) + + vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + cmp rcx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test rdi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF +.out0: + add rdi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub rcx, byte SIZEOF_YMMWORD + jz near .nextrow + + add rsi, byte SIZEOF_YMMWORD ; inptr0 + add rbx, byte SIZEOF_YMMWORD ; inptr1 + add rdx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + +.column_st64: + lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE + cmp rcx, byte 2*SIZEOF_YMMWORD + jb short .column_st32 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + add rdi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmF + sub rcx, byte 2*SIZEOF_YMMWORD + jmp short .column_st31 +.column_st32: + cmp rcx, byte SIZEOF_YMMWORD + jb short .column_st31 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + add rdi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub rcx, byte SIZEOF_YMMWORD + jmp short .column_st31 +.column_st31: + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st15 + vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + add rdi, byte SIZEOF_XMMWORD ; outptr + vperm2i128 ymmA, ymmA, ymmA, 1 + sub rcx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_MMWORD + jb short .column_st7 + vmovq XMM_MMWORD [rdi], xmmA + add rdi, byte SIZEOF_MMWORD + sub rcx, byte SIZEOF_MMWORD + vpsrldq xmmA, xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_DWORD + jb short .column_st3 + vmovd XMM_DWORD [rdi], xmmA + add rdi, byte SIZEOF_DWORD + sub rcx, byte SIZEOF_DWORD + vpsrldq xmmA, xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of rax to the output when it has enough + ; space. + vmovd eax, xmmA + cmp rcx, byte SIZEOF_WORD + jb short .column_st1 + mov word [rdi], ax + add rdi, byte SIZEOF_WORD + sub rcx, byte SIZEOF_WORD + shr rax, 16 +.column_st1: + ; Store the lower 1 byte of rax to the output when it has enough + ; space. + test rcx, rcx + jz short .nextrow + mov byte [rdi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%else + vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%endif + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **) + ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E + ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U) + vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F + ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V) + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F + ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V) + + vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E + ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36 + ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M) + vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F + ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V) + vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37 + ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N) + + vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + + vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + cmp rcx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test rdi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC + vmovntdq YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC + vmovdqu YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH +.out0: + add rdi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub rcx, byte SIZEOF_YMMWORD + jz near .nextrow + + add rsi, byte SIZEOF_YMMWORD ; inptr0 + add rbx, byte SIZEOF_YMMWORD ; inptr1 + add rdx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + +.column_st64: + cmp rcx, byte SIZEOF_YMMWORD/2 + jb short .column_st32 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + add rdi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmC + vmovdqa ymmD, ymmH + sub rcx, byte SIZEOF_YMMWORD/2 +.column_st32: + cmp rcx, byte SIZEOF_YMMWORD/4 + jb short .column_st16 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + add rdi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub rcx, byte SIZEOF_YMMWORD/4 +.column_st16: + cmp rcx, byte SIZEOF_YMMWORD/8 + jb short .column_st15 + vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + vperm2i128 ymmA, ymmA, ymmA, 1 + add rdi, byte SIZEOF_XMMWORD ; outptr + sub rcx, byte SIZEOF_YMMWORD/8 +.column_st15: + ; Store two pixels (8 bytes) of ymmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_YMMWORD/16 + jb short .column_st7 + vmovq MMWORD [rdi], xmmA + add rdi, byte SIZEOF_YMMWORD/16*4 + sub rcx, byte SIZEOF_YMMWORD/16 + vpsrldq xmmA, SIZEOF_YMMWORD/16*4 +.column_st7: + ; Store one pixel (4 bytes) of ymmA to the output when it has enough + ; space. + test rcx, rcx + jz short .nextrow + vmovd XMM_DWORD [rdi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + +.nextrow: + pop rcx + pop rsi + pop rbx + pop rdx + pop rdi + pop rax + + add rsi, byte SIZEOF_JSAMPROW + add rbx, byte SIZEOF_JSAMPROW + add rdx, byte SIZEOF_JSAMPROW + add rdi, byte SIZEOF_JSAMPROW ; output_buf + dec rax ; num_rows + jg near .rowloop + + sfence ; flush the write buffer + +.return: + pop rbx + vzeroupper + UNCOLLECT_ARGS 5 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-sse2.asm new file mode 100644 index 0000000..261f74d --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolext-sse2.asm @@ -0,0 +1,437 @@ +; +; jdcolext.asm - colorspace conversion (64-bit SSE2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2012, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Convert some rows of samples to the output colorspace. +; +; GLOBAL(void) +; jsimd_ycc_rgb_convert_sse2(JDIMENSION out_width, JSAMPIMAGE input_buf, +; JDIMENSION input_row, JSAMPARRAY output_buf, +; int num_rows) +; + +; r10d = JDIMENSION out_width +; r11 = JSAMPIMAGE input_buf +; r12d = JDIMENSION input_row +; r13 = JSAMPARRAY output_buf +; r14d = int num_rows + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_ycc_rgb_convert_sse2) + +EXTN(jsimd_ycc_rgb_convert_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 5 + push rbx + + mov ecx, r10d ; num_cols + test rcx, rcx + jz near .return + + push rcx + + mov rdi, r11 + mov ecx, r12d + mov rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] + lea rsi, [rsi+rcx*SIZEOF_JSAMPROW] + lea rbx, [rbx+rcx*SIZEOF_JSAMPROW] + lea rdx, [rdx+rcx*SIZEOF_JSAMPROW] + + pop rcx + + mov rdi, r13 + mov eax, r14d + test rax, rax + jle near .return +.rowloop: + push rax + push rdi + push rdx + push rbx + push rsi + push rcx ; col + + mov rsip, JSAMPROW [rsi] ; inptr0 + mov rbxp, JSAMPROW [rbx] ; inptr1 + mov rdxp, JSAMPROW [rdx] ; inptr2 + mov rdip, JSAMPROW [rdi] ; outptr +.columnloop: + + movdqa xmm5, XMMWORD [rbx] ; xmm5=Cb(0123456789ABCDEF) + movdqa xmm1, XMMWORD [rdx] ; xmm1=Cr(0123456789ABCDEF) + + pcmpeqw xmm4, xmm4 + pcmpeqw xmm7, xmm7 + psrlw xmm4, BYTE_BIT + psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + movdqa xmm0, xmm4 ; xmm0=xmm4={0xFF 0x00 0xFF 0x00 ..} + + pand xmm4, xmm5 ; xmm4=Cb(02468ACE)=CbE + psrlw xmm5, BYTE_BIT ; xmm5=Cb(13579BDF)=CbO + pand xmm0, xmm1 ; xmm0=Cr(02468ACE)=CrE + psrlw xmm1, BYTE_BIT ; xmm1=Cr(13579BDF)=CrO + + paddw xmm4, xmm7 + paddw xmm5, xmm7 + paddw xmm0, xmm7 + paddw xmm1, xmm7 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + movdqa xmm2, xmm4 ; xmm2=CbE + movdqa xmm3, xmm5 ; xmm3=CbO + paddw xmm4, xmm4 ; xmm4=2*CbE + paddw xmm5, xmm5 ; xmm5=2*CbO + movdqa xmm6, xmm0 ; xmm6=CrE + movdqa xmm7, xmm1 ; xmm7=CrO + paddw xmm0, xmm0 ; xmm0=2*CrE + paddw xmm1, xmm1 ; xmm1=2*CrO + + pmulhw xmm4, [rel PW_MF0228] ; xmm4=(2*CbE * -FIX(0.22800)) + pmulhw xmm5, [rel PW_MF0228] ; xmm5=(2*CbO * -FIX(0.22800)) + pmulhw xmm0, [rel PW_F0402] ; xmm0=(2*CrE * FIX(0.40200)) + pmulhw xmm1, [rel PW_F0402] ; xmm1=(2*CrO * FIX(0.40200)) + + paddw xmm4, [rel PW_ONE] + paddw xmm5, [rel PW_ONE] + psraw xmm4, 1 ; xmm4=(CbE * -FIX(0.22800)) + psraw xmm5, 1 ; xmm5=(CbO * -FIX(0.22800)) + paddw xmm0, [rel PW_ONE] + paddw xmm1, [rel PW_ONE] + psraw xmm0, 1 ; xmm0=(CrE * FIX(0.40200)) + psraw xmm1, 1 ; xmm1=(CrO * FIX(0.40200)) + + paddw xmm4, xmm2 + paddw xmm5, xmm3 + paddw xmm4, xmm2 ; xmm4=(CbE * FIX(1.77200))=(B-Y)E + paddw xmm5, xmm3 ; xmm5=(CbO * FIX(1.77200))=(B-Y)O + paddw xmm0, xmm6 ; xmm0=(CrE * FIX(1.40200))=(R-Y)E + paddw xmm1, xmm7 ; xmm1=(CrO * FIX(1.40200))=(R-Y)O + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=(B-Y)E + movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(B-Y)O + + movdqa xmm4, xmm2 + movdqa xmm5, xmm3 + punpcklwd xmm2, xmm6 + punpckhwd xmm4, xmm6 + pmaddwd xmm2, [rel PW_MF0344_F0285] + pmaddwd xmm4, [rel PW_MF0344_F0285] + punpcklwd xmm3, xmm7 + punpckhwd xmm5, xmm7 + pmaddwd xmm3, [rel PW_MF0344_F0285] + pmaddwd xmm5, [rel PW_MF0344_F0285] + + paddd xmm2, [rel PD_ONEHALF] + paddd xmm4, [rel PD_ONEHALF] + psrad xmm2, SCALEBITS + psrad xmm4, SCALEBITS + paddd xmm3, [rel PD_ONEHALF] + paddd xmm5, [rel PD_ONEHALF] + psrad xmm3, SCALEBITS + psrad xmm5, SCALEBITS + + packssdw xmm2, xmm4 ; xmm2=CbE*-FIX(0.344)+CrE*FIX(0.285) + packssdw xmm3, xmm5 ; xmm3=CbO*-FIX(0.344)+CrO*FIX(0.285) + psubw xmm2, xmm6 ; xmm2=CbE*-FIX(0.344)+CrE*-FIX(0.714)=(G-Y)E + psubw xmm3, xmm7 ; xmm3=CbO*-FIX(0.344)+CrO*-FIX(0.714)=(G-Y)O + + movdqa xmm5, XMMWORD [rsi] ; xmm5=Y(0123456789ABCDEF) + + pcmpeqw xmm4, xmm4 + psrlw xmm4, BYTE_BIT ; xmm4={0xFF 0x00 0xFF 0x00 ..} + pand xmm4, xmm5 ; xmm4=Y(02468ACE)=YE + psrlw xmm5, BYTE_BIT ; xmm5=Y(13579BDF)=YO + + paddw xmm0, xmm4 ; xmm0=((R-Y)E+YE)=RE=R(02468ACE) + paddw xmm1, xmm5 ; xmm1=((R-Y)O+YO)=RO=R(13579BDF) + packuswb xmm0, xmm0 ; xmm0=R(02468ACE********) + packuswb xmm1, xmm1 ; xmm1=R(13579BDF********) + + paddw xmm2, xmm4 ; xmm2=((G-Y)E+YE)=GE=G(02468ACE) + paddw xmm3, xmm5 ; xmm3=((G-Y)O+YO)=GO=G(13579BDF) + packuswb xmm2, xmm2 ; xmm2=G(02468ACE********) + packuswb xmm3, xmm3 ; xmm3=G(13579BDF********) + + paddw xmm4, XMMWORD [wk(0)] ; xmm4=(YE+(B-Y)E)=BE=B(02468ACE) + paddw xmm5, XMMWORD [wk(1)] ; xmm5=(YO+(B-Y)O)=BO=B(13579BDF) + packuswb xmm4, xmm4 ; xmm4=B(02468ACE********) + packuswb xmm5, xmm5 ; xmm5=B(13579BDF********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F) + punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F) + + movdqa xmmG, xmmA + movdqa xmmH, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07) + punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F) + + psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --) + psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --) + + movdqa xmmC, xmmD + movdqa xmmB, xmmD + punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18) + punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --) + + psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --) + + movdqa xmmF, xmmE + punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29) + punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --) + + pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03) + movdqa xmmB, xmmE + punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14) + punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07) + punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29) + + pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B) + movdqa xmmB, xmmF + punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C) + punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F) + punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --) + + punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test rdi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF +.out0: + add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub rcx, byte SIZEOF_XMMWORD + jz near .nextrow + + add rsi, byte SIZEOF_XMMWORD ; inptr0 + add rbx, byte SIZEOF_XMMWORD ; inptr1 + add rdx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + +.column_st32: + lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE + cmp rcx, byte 2*SIZEOF_XMMWORD + jb short .column_st16 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + add rdi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmF + sub rcx, byte 2*SIZEOF_XMMWORD + jmp short .column_st15 +.column_st16: + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st15 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + add rdi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub rcx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_MMWORD + jb short .column_st7 + movq XMM_MMWORD [rdi], xmmA + add rdi, byte SIZEOF_MMWORD + sub rcx, byte SIZEOF_MMWORD + psrldq xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_DWORD + jb short .column_st3 + movd XMM_DWORD [rdi], xmmA + add rdi, byte SIZEOF_DWORD + sub rcx, byte SIZEOF_DWORD + psrldq xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of rax to the output when it has enough + ; space. + movd eax, xmmA + cmp rcx, byte SIZEOF_WORD + jb short .column_st1 + mov word [rdi], ax + add rdi, byte SIZEOF_WORD + sub rcx, byte SIZEOF_WORD + shr rax, 16 +.column_st1: + ; Store the lower 1 byte of rax to the output when it has enough + ; space. + test rcx, rcx + jz short .nextrow + mov byte [rdi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%else + pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%endif + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E) + punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F) + punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F) + + movdqa xmmC, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36) + punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E) + movdqa xmmG, xmmB + punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37) + punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + movdqa xmmH, xmmC + punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test rdi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC + movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC + movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH +.out0: + add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub rcx, byte SIZEOF_XMMWORD + jz near .nextrow + + add rsi, byte SIZEOF_XMMWORD ; inptr0 + add rbx, byte SIZEOF_XMMWORD ; inptr1 + add rdx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + +.column_st32: + cmp rcx, byte SIZEOF_XMMWORD/2 + jb short .column_st16 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + add rdi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmC + movdqa xmmD, xmmH + sub rcx, byte SIZEOF_XMMWORD/2 +.column_st16: + cmp rcx, byte SIZEOF_XMMWORD/4 + jb short .column_st15 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + add rdi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub rcx, byte SIZEOF_XMMWORD/4 +.column_st15: + ; Store two pixels (8 bytes) of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_XMMWORD/8 + jb short .column_st7 + movq MMWORD [rdi], xmmA + add rdi, byte SIZEOF_XMMWORD/8*4 + sub rcx, byte SIZEOF_XMMWORD/8 + psrldq xmmA, SIZEOF_XMMWORD/8*4 +.column_st7: + ; Store one pixel (4 bytes) of xmmA to the output when it has enough + ; space. + test rcx, rcx + jz short .nextrow + movd XMM_DWORD [rdi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + +.nextrow: + pop rcx + pop rsi + pop rbx + pop rdx + pop rdi + pop rax + + add rsi, byte SIZEOF_JSAMPROW + add rbx, byte SIZEOF_JSAMPROW + add rdx, byte SIZEOF_JSAMPROW + add rdi, byte SIZEOF_JSAMPROW ; output_buf + dec rax ; num_rows + jg near .rowloop + + sfence ; flush the write buffer + +.return: + pop rbx + UNCOLLECT_ARGS 5 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-avx2.asm new file mode 100644 index 0000000..bd5aa00 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-avx2.asm @@ -0,0 +1,114 @@ +; +; jdcolor.asm - colorspace conversion (64-bit AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_ycc_rgb_convert_avx2) + +EXTN(jconst_ycc_rgb_convert_avx2): + +PW_F0402 times 16 dw F_0_402 +PW_MF0228 times 16 dw -F_0_228 +PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285 +PW_ONE times 16 dw 1 +PD_ONEHALF times 8 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgb_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extrgbx_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgr_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extbgrx_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxbgr_convert_avx2 +%include "jdcolext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_avx2 jsimd_ycc_extxrgb_convert_avx2 +%include "jdcolext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-sse2.asm new file mode 100644 index 0000000..40343fe --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdcolor-sse2.asm @@ -0,0 +1,113 @@ +; +; jdcolor.asm - colorspace conversion (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_ycc_rgb_convert_sse2) + +EXTN(jconst_ycc_rgb_convert_sse2): + +PW_F0402 times 8 dw F_0_402 +PW_MF0228 times 8 dw -F_0_228 +PW_MF0344_F0285 times 4 dw -F_0_344, F_0_285 +PW_ONE times 8 dw 1 +PD_ONEHALF times 4 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgb_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extrgbx_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgr_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extbgrx_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxbgr_convert_sse2 +%include "jdcolext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_ycc_rgb_convert_sse2 jsimd_ycc_extxrgb_convert_sse2 +%include "jdcolext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-avx2.asm new file mode 100644 index 0000000..6a5f1da --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-avx2.asm @@ -0,0 +1,132 @@ +; +; jdmerge.asm - merged upsampling/color conversion (64-bit AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_merged_upsample_avx2) + +EXTN(jconst_merged_upsample_avx2): + +PW_F0402 times 16 dw F_0_402 +PW_MF0228 times 16 dw -F_0_228 +PW_MF0344_F0285 times 8 dw -F_0_344, F_0_285 +PW_ONE times 16 dw 1 +PD_ONEHALF times 8 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extrgb_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extrgb_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extrgbx_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extrgbx_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extbgr_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extbgr_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extbgrx_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extbgrx_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extxbgr_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extxbgr_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_avx2 \ + jsimd_h2v1_extxrgb_merged_upsample_avx2 +%define jsimd_h2v2_merged_upsample_avx2 \ + jsimd_h2v2_extxrgb_merged_upsample_avx2 +%include "jdmrgext-avx2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-sse2.asm new file mode 100644 index 0000000..8c269b8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmerge-sse2.asm @@ -0,0 +1,131 @@ +; +; jdmerge.asm - merged upsampling/color conversion (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + +%define SCALEBITS 16 + +F_0_344 equ 22554 ; FIX(0.34414) +F_0_714 equ 46802 ; FIX(0.71414) +F_1_402 equ 91881 ; FIX(1.40200) +F_1_772 equ 116130 ; FIX(1.77200) +F_0_402 equ (F_1_402 - 65536) ; FIX(1.40200) - FIX(1) +F_0_285 equ ( 65536 - F_0_714) ; FIX(1) - FIX(0.71414) +F_0_228 equ (131072 - F_1_772) ; FIX(2) - FIX(1.77200) + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_merged_upsample_sse2) + +EXTN(jconst_merged_upsample_sse2): + +PW_F0402 times 8 dw F_0_402 +PW_MF0228 times 8 dw -F_0_228 +PW_MF0344_F0285 times 4 dw -F_0_344, F_0_285 +PW_ONE times 8 dw 1 +PD_ONEHALF times 4 dd 1 << (SCALEBITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 + +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGB_RED +%define RGB_GREEN EXT_RGB_GREEN +%define RGB_BLUE EXT_RGB_BLUE +%define RGB_PIXELSIZE EXT_RGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extrgb_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extrgb_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_RGBX_RED +%define RGB_GREEN EXT_RGBX_GREEN +%define RGB_BLUE EXT_RGBX_BLUE +%define RGB_PIXELSIZE EXT_RGBX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extrgbx_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extrgbx_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGR_RED +%define RGB_GREEN EXT_BGR_GREEN +%define RGB_BLUE EXT_BGR_BLUE +%define RGB_PIXELSIZE EXT_BGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extbgr_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extbgr_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_BGRX_RED +%define RGB_GREEN EXT_BGRX_GREEN +%define RGB_BLUE EXT_BGRX_BLUE +%define RGB_PIXELSIZE EXT_BGRX_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extbgrx_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extbgrx_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XBGR_RED +%define RGB_GREEN EXT_XBGR_GREEN +%define RGB_BLUE EXT_XBGR_BLUE +%define RGB_PIXELSIZE EXT_XBGR_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extxbgr_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extxbgr_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" + +%undef RGB_RED +%undef RGB_GREEN +%undef RGB_BLUE +%undef RGB_PIXELSIZE +%define RGB_RED EXT_XRGB_RED +%define RGB_GREEN EXT_XRGB_GREEN +%define RGB_BLUE EXT_XRGB_BLUE +%define RGB_PIXELSIZE EXT_XRGB_PIXELSIZE +%define jsimd_h2v1_merged_upsample_sse2 \ + jsimd_h2v1_extxrgb_merged_upsample_sse2 +%define jsimd_h2v2_merged_upsample_sse2 \ + jsimd_h2v2_extxrgb_merged_upsample_sse2 +%include "jdmrgext-sse2.asm" diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-avx2.asm new file mode 100644 index 0000000..01826fb --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-avx2.asm @@ -0,0 +1,594 @@ +; +; jdmrgext.asm - merged upsampling/color conversion (64-bit AVX2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2012, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v1_merged_upsample_avx2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +; r10d = JDIMENSION output_width +; r11 = JSAMPIMAGE input_buf +; r12d = JDIMENSION in_row_group_ctr +; r13 = JSAMPARRAY output_buf + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_YMMWORD ; ymmword wk[WK_NUM] +%define WK_NUM 3 + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_merged_upsample_avx2) + +EXTN(jsimd_h2v1_merged_upsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_YMMWORD) ; align to 256 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, SIZEOF_YMMWORD * WK_NUM + COLLECT_ARGS 4 + push rbx + + mov ecx, r10d ; col + test rcx, rcx + jz near .return + + push rcx + + mov rdi, r11 + mov ecx, r12d + mov rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] + mov rdi, r13 + mov rsip, JSAMPROW [rsi+rcx*SIZEOF_JSAMPROW] ; inptr0 + mov rbxp, JSAMPROW [rbx+rcx*SIZEOF_JSAMPROW] ; inptr1 + mov rdxp, JSAMPROW [rdx+rcx*SIZEOF_JSAMPROW] ; inptr2 + mov rdip, JSAMPROW [rdi] ; outptr + + pop rcx ; col + +.columnloop: + + vmovdqu ymm6, YMMWORD [rbx] ; ymm6=Cb(0123456789ABCDEFGHIJKLMNOPQRSTUV) + vmovdqu ymm7, YMMWORD [rdx] ; ymm7=Cr(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpxor ymm1, ymm1, ymm1 ; ymm1=(all 0's) + vpcmpeqw ymm3, ymm3, ymm3 + vpsllw ymm3, ymm3, 7 ; ymm3={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + vpermq ymm6, ymm6, 0xd8 ; ymm6=Cb(01234567GHIJKLMN89ABCDEFOPQRSTUV) + vpermq ymm7, ymm7, 0xd8 ; ymm7=Cr(01234567GHIJKLMN89ABCDEFOPQRSTUV) + vpunpcklbw ymm4, ymm6, ymm1 ; ymm4=Cb(0123456789ABCDEF)=CbL + vpunpckhbw ymm6, ymm6, ymm1 ; ymm6=Cb(GHIJKLMNOPQRSTUV)=CbH + vpunpcklbw ymm0, ymm7, ymm1 ; ymm0=Cr(0123456789ABCDEF)=CrL + vpunpckhbw ymm7, ymm7, ymm1 ; ymm7=Cr(GHIJKLMNOPQRSTUV)=CrH + + vpaddw ymm5, ymm6, ymm3 + vpaddw ymm2, ymm4, ymm3 + vpaddw ymm1, ymm7, ymm3 + vpaddw ymm3, ymm0, ymm3 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + vpaddw ymm6, ymm5, ymm5 ; ymm6=2*CbH + vpaddw ymm4, ymm2, ymm2 ; ymm4=2*CbL + vpaddw ymm7, ymm1, ymm1 ; ymm7=2*CrH + vpaddw ymm0, ymm3, ymm3 ; ymm0=2*CrL + + vpmulhw ymm6, ymm6, [rel PW_MF0228] ; ymm6=(2*CbH * -FIX(0.22800)) + vpmulhw ymm4, ymm4, [rel PW_MF0228] ; ymm4=(2*CbL * -FIX(0.22800)) + vpmulhw ymm7, ymm7, [rel PW_F0402] ; ymm7=(2*CrH * FIX(0.40200)) + vpmulhw ymm0, ymm0, [rel PW_F0402] ; ymm0=(2*CrL * FIX(0.40200)) + + vpaddw ymm6, ymm6, [rel PW_ONE] + vpaddw ymm4, ymm4, [rel PW_ONE] + vpsraw ymm6, ymm6, 1 ; ymm6=(CbH * -FIX(0.22800)) + vpsraw ymm4, ymm4, 1 ; ymm4=(CbL * -FIX(0.22800)) + vpaddw ymm7, ymm7, [rel PW_ONE] + vpaddw ymm0, ymm0, [rel PW_ONE] + vpsraw ymm7, ymm7, 1 ; ymm7=(CrH * FIX(0.40200)) + vpsraw ymm0, ymm0, 1 ; ymm0=(CrL * FIX(0.40200)) + + vpaddw ymm6, ymm6, ymm5 + vpaddw ymm4, ymm4, ymm2 + vpaddw ymm6, ymm6, ymm5 ; ymm6=(CbH * FIX(1.77200))=(B-Y)H + vpaddw ymm4, ymm4, ymm2 ; ymm4=(CbL * FIX(1.77200))=(B-Y)L + vpaddw ymm7, ymm7, ymm1 ; ymm7=(CrH * FIX(1.40200))=(R-Y)H + vpaddw ymm0, ymm0, ymm3 ; ymm0=(CrL * FIX(1.40200))=(R-Y)L + + vmovdqa YMMWORD [wk(0)], ymm6 ; wk(0)=(B-Y)H + vmovdqa YMMWORD [wk(1)], ymm7 ; wk(1)=(R-Y)H + + vpunpckhwd ymm6, ymm5, ymm1 + vpunpcklwd ymm5, ymm5, ymm1 + vpmaddwd ymm5, ymm5, [rel PW_MF0344_F0285] + vpmaddwd ymm6, ymm6, [rel PW_MF0344_F0285] + vpunpckhwd ymm7, ymm2, ymm3 + vpunpcklwd ymm2, ymm2, ymm3 + vpmaddwd ymm2, ymm2, [rel PW_MF0344_F0285] + vpmaddwd ymm7, ymm7, [rel PW_MF0344_F0285] + + vpaddd ymm5, ymm5, [rel PD_ONEHALF] + vpaddd ymm6, ymm6, [rel PD_ONEHALF] + vpsrad ymm5, ymm5, SCALEBITS + vpsrad ymm6, ymm6, SCALEBITS + vpaddd ymm2, ymm2, [rel PD_ONEHALF] + vpaddd ymm7, ymm7, [rel PD_ONEHALF] + vpsrad ymm2, ymm2, SCALEBITS + vpsrad ymm7, ymm7, SCALEBITS + + vpackssdw ymm5, ymm5, ymm6 ; ymm5=CbH*-FIX(0.344)+CrH*FIX(0.285) + vpackssdw ymm2, ymm2, ymm7 ; ymm2=CbL*-FIX(0.344)+CrL*FIX(0.285) + vpsubw ymm5, ymm5, ymm1 ; ymm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H + vpsubw ymm2, ymm2, ymm3 ; ymm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L + + vmovdqa YMMWORD [wk(2)], ymm5 ; wk(2)=(G-Y)H + + mov al, 2 ; Yctr + jmp short .Yloop_1st + +.Yloop_2nd: + vmovdqa ymm0, YMMWORD [wk(1)] ; ymm0=(R-Y)H + vmovdqa ymm2, YMMWORD [wk(2)] ; ymm2=(G-Y)H + vmovdqa ymm4, YMMWORD [wk(0)] ; ymm4=(B-Y)H + +.Yloop_1st: + vmovdqu ymm7, YMMWORD [rsi] ; ymm7=Y(0123456789ABCDEFGHIJKLMNOPQRSTUV) + + vpcmpeqw ymm6, ymm6, ymm6 + vpsrlw ymm6, ymm6, BYTE_BIT ; ymm6={0xFF 0x00 0xFF 0x00 ..} + vpand ymm6, ymm6, ymm7 ; ymm6=Y(02468ACEGIKMOQSU)=YE + vpsrlw ymm7, ymm7, BYTE_BIT ; ymm7=Y(13579BDFHJLNPRTV)=YO + + vmovdqa ymm1, ymm0 ; ymm1=ymm0=(R-Y)(L/H) + vmovdqa ymm3, ymm2 ; ymm3=ymm2=(G-Y)(L/H) + vmovdqa ymm5, ymm4 ; ymm5=ymm4=(B-Y)(L/H) + + vpaddw ymm0, ymm0, ymm6 ; ymm0=((R-Y)+YE)=RE=R(02468ACEGIKMOQSU) + vpaddw ymm1, ymm1, ymm7 ; ymm1=((R-Y)+YO)=RO=R(13579BDFHJLNPRTV) + vpackuswb ymm0, ymm0, ymm0 ; ymm0=R(02468ACE********GIKMOQSU********) + vpackuswb ymm1, ymm1, ymm1 ; ymm1=R(13579BDF********HJLNPRTV********) + + vpaddw ymm2, ymm2, ymm6 ; ymm2=((G-Y)+YE)=GE=G(02468ACEGIKMOQSU) + vpaddw ymm3, ymm3, ymm7 ; ymm3=((G-Y)+YO)=GO=G(13579BDFHJLNPRTV) + vpackuswb ymm2, ymm2, ymm2 ; ymm2=G(02468ACE********GIKMOQSU********) + vpackuswb ymm3, ymm3, ymm3 ; ymm3=G(13579BDF********HJLNPRTV********) + + vpaddw ymm4, ymm4, ymm6 ; ymm4=((B-Y)+YE)=BE=B(02468ACEGIKMOQSU) + vpaddw ymm5, ymm5, ymm7 ; ymm5=((B-Y)+YO)=BO=B(13579BDFHJLNPRTV) + vpackuswb ymm4, ymm4, ymm4 ; ymm4=B(02468ACE********GIKMOQSU********) + vpackuswb ymm5, ymm5, ymm5 ; ymm5=B(13579BDF********HJLNPRTV********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + ; ymmH=(** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** ** **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmB ; ymmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F + ; 2G 0H 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V) + vpunpcklbw ymmD, ymmD, ymmF ; ymmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F + ; 1H 2H 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V) + + vpsrldq ymmH, ymmA, 2 ; ymmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E 0G 1G + ; 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U -- --) + vpunpckhwd ymmG, ymmA, ymmE ; ymmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F + ; 0O 1O 2O 0P 0Q 1Q 2Q 0R 0S 1S 2S 0T 0U 1U 2U 0V) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07 + ; 0G 1G 2G 0H 0I 1I 2I 0J 0K 1K 2K 0L 0M 1M 2M 0N) + + vpsrldq ymmE, ymmE, 2 ; ymmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F 2G 0H + ; 2I 0J 2K 0L 2M 0N 2O 0P 2Q 0R 2S 0T 2U 0V -- --) + + vpsrldq ymmB, ymmD, 2 ; ymmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F 1H 2H + ; 1J 2J 1L 2L 1N 2N 1P 2P 1R 2R 1T 2T 1V 2V -- --) + vpunpckhwd ymmC, ymmD, ymmH ; ymmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F 0G 1G + ; 1P 2P 0Q 1Q 1R 2R 0S 1S 1T 2T 0U 1U 1V 2V -- --) + vpunpcklwd ymmD, ymmD, ymmH ; ymmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18 + ; 1H 2H 0I 1I 1J 2J 0K 1K 1L 2L 0M 1M 1N 2N 0O 1O) + + vpunpckhwd ymmF, ymmE, ymmB ; ymmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F 2G 0H 1H 2H + ; 2Q 0R 1R 2R 2S 0T 1T 2T 2U 0V 1V 2V -- -- -- --) + vpunpcklwd ymmE, ymmE, ymmB ; ymmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29 + ; 2I 0J 1J 2J 2K 0L 1L 2L 2M 0N 1N 2N 2O 0P 1P 2P) + + vpshufd ymmH, ymmA, 0x4E ; ymmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03 + ; 0K 1K 2K 0L 0M 1M 2M 0N 0G 1G 2G 0H 0I 1I 2I 0J) + vpunpckldq ymmA, ymmA, ymmD ; ymmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14 + ; 0G 1G 2G 0H 1H 2H 0I 1I 0I 1I 2I 0J 1J 2J 0K 1K) + vpunpckhdq ymmD, ymmD, ymmE ; ymmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29 + ; 1L 2L 0M 1M 2M 0N 1N 2N 1N 2N 0O 1O 2O 0P 1P 2P) + vpunpckldq ymmE, ymmE, ymmH ; ymmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07 + ; 2I 0J 1J 2J 0K 1K 2K 0L 2K 0L 1L 2L 0M 1M 2M 0N) + + vpshufd ymmH, ymmG, 0x4E ; ymmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B + ; 0S 1S 2S 0T 0U 1U 2U 0V 0O 1O 2O 0P 0Q 1Q 2Q 0R) + vpunpckldq ymmG, ymmG, ymmC ; ymmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C + ; 0O 1O 2O 0P 1P 2P 0Q 1Q 0Q 1Q 2Q 0R 1R 2R 0S 1S) + vpunpckhdq ymmC, ymmC, ymmF ; ymmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F 0G 1G 2G 0H 1H 2H + ; 1T 2T 0U 1U 2U 0V 1V 2V 1V 2V -- -- -- -- -- --) + vpunpckldq ymmF, ymmF, ymmH ; ymmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 2S 0T 1T 2T 0U 1U 2U 0V) + + vpunpcklqdq ymmH, ymmA, ymmE ; ymmH=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vpunpcklqdq ymmG, ymmD, ymmG ; ymmG=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A + ; 1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q) + vpunpcklqdq ymmC, ymmF, ymmC ; ymmC=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + vperm2i128 ymmA, ymmH, ymmG, 0x20 ; ymmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05 + ; 15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + vperm2i128 ymmD, ymmC, ymmH, 0x30 ; ymmD=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F + ; 0G 1G 2G 0H 1H 2H 0I 1I 2I 0J 1J 2J 0K 1K 2K 0L) + vperm2i128 ymmF, ymmG, ymmC, 0x31 ; ymmF=(1L 2L 0M 1M 2M 0N 1N 2N 0O 1O 2O 0P 1P 2P 0Q 1Q + ; 2Q 0R 1R 2R 0S 1S 2S 0T 1T 2T 0U 1U 2U 0V 1V 2V) + + cmp rcx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test rdi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmF +.out0: + add rdi, byte RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub rcx, byte SIZEOF_YMMWORD + jz near .endcolumn + + add rsi, byte SIZEOF_YMMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add rbx, byte SIZEOF_YMMWORD ; inptr1 + add rdx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + +.column_st64: + lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE + cmp rcx, byte 2*SIZEOF_YMMWORD + jb short .column_st32 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + add rdi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmF + sub rcx, byte 2*SIZEOF_YMMWORD + jmp short .column_st31 +.column_st32: + cmp rcx, byte SIZEOF_YMMWORD + jb short .column_st31 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + add rdi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub rcx, byte SIZEOF_YMMWORD + jmp short .column_st31 +.column_st31: + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st15 + vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + add rdi, byte SIZEOF_XMMWORD ; outptr + vperm2i128 ymmA, ymmA, ymmA, 1 + sub rcx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_MMWORD + jb short .column_st7 + vmovq XMM_MMWORD [rdi], xmmA + add rdi, byte SIZEOF_MMWORD + sub rcx, byte SIZEOF_MMWORD + vpsrldq xmmA, xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_DWORD + jb short .column_st3 + vmovd XMM_DWORD [rdi], xmmA + add rdi, byte SIZEOF_DWORD + sub rcx, byte SIZEOF_DWORD + vpsrldq xmmA, xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of rax to the output when it has enough + ; space. + vmovd eax, xmmA + cmp rcx, byte SIZEOF_WORD + jb short .column_st1 + mov word [rdi], ax + add rdi, byte SIZEOF_WORD + sub rcx, byte SIZEOF_WORD + shr rax, 16 +.column_st1: + ; Store the lower 1 byte of rax to the output when it has enough + ; space. + test rcx, rcx + jz short .endcolumn + mov byte [rdi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + vpcmpeqb ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpcmpeqb ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%else + vpxor ymm6, ymm6, ymm6 ; ymm6=XE=X(02468ACE********GIKMOQSU********) + vpxor ymm7, ymm7, ymm7 ; ymm7=XO=X(13579BDF********HJLNPRTV********) +%endif + ; ymmA=(00 02 04 06 08 0A 0C 0E ** 0G 0I 0K 0M 0O 0Q 0S 0U **) + ; ymmB=(01 03 05 07 09 0B 0D 0F ** 0H 0J 0L 0N 0P 0R 0T 0V **) + ; ymmC=(10 12 14 16 18 1A 1C 1E ** 1G 1I 1K 1M 1O 1Q 1S 1U **) + ; ymmD=(11 13 15 17 19 1B 1D 1F ** 1H 1J 1L 1N 1P 1R 1T 1V **) + ; ymmE=(20 22 24 26 28 2A 2C 2E ** 2G 2I 2K 2M 2O 2Q 2S 2U **) + ; ymmF=(21 23 25 27 29 2B 2D 2F ** 2H 2J 2L 2N 2P 2R 2T 2V **) + ; ymmG=(30 32 34 36 38 3A 3C 3E ** 3G 3I 3K 3M 3O 3Q 3S 3U **) + ; ymmH=(31 33 35 37 39 3B 3D 3F ** 3H 3J 3L 3N 3P 3R 3T 3V **) + + vpunpcklbw ymmA, ymmA, ymmC ; ymmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E + ; 0G 1G 0I 1I 0K 1K 0M 1M 0O 1O 0Q 1Q 0S 1S 0U 1U) + vpunpcklbw ymmE, ymmE, ymmG ; ymmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E + ; 2G 3G 2I 3I 2K 3K 2M 3M 2O 3O 2Q 3Q 2S 3S 2U 3U) + vpunpcklbw ymmB, ymmB, ymmD ; ymmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F + ; 0H 1H 0J 1J 0L 1L 0N 1N 0P 1P 0R 1R 0T 1T 0V 1V) + vpunpcklbw ymmF, ymmF, ymmH ; ymmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F + ; 2H 3H 2J 3J 2L 3L 2N 3N 2P 3P 2R 3R 2T 3T 2V 3V) + + vpunpckhwd ymmC, ymmA, ymmE ; ymmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E + ; 0O 1O 2O 3O 0Q 1Q 2Q 3Q 0S 1S 2S 3S 0U 1U 2U 3U) + vpunpcklwd ymmA, ymmA, ymmE ; ymmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36 + ; 0G 1G 2G 3G 0I 1I 2I 3I 0K 1K 2K 3K 0M 1M 2M 3M) + vpunpckhwd ymmG, ymmB, ymmF ; ymmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F + ; 0P 1P 2P 3P 0R 1R 2R 3R 0T 1T 2T 3T 0V 1V 2V 3V) + vpunpcklwd ymmB, ymmB, ymmF ; ymmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37 + ; 0H 1H 2H 3H 0J 1J 2J 3J 0L 1L 2L 3L 0N 1N 2N 3N) + + vpunpckhdq ymmE, ymmA, ymmB ; ymmE=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37 + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vpunpckldq ymmB, ymmA, ymmB ; ymmB=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J) + vpunpckhdq ymmF, ymmC, ymmG ; ymmF=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + vpunpckldq ymmG, ymmC, ymmG ; ymmG=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R) + + vperm2i128 ymmA, ymmB, ymmE, 0x20 ; ymmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33 + ; 04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + vperm2i128 ymmD, ymmG, ymmF, 0x20 ; ymmD=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B + ; 0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + vperm2i128 ymmC, ymmB, ymmE, 0x31 ; ymmC=(0G 1G 2G 3G 0H 1H 2H 3H 0I 1I 2I 3I 0J 1J 2J 3J + ; 0K 1K 2K 3K 0L 1L 2L 3L 0M 1M 2M 3M 0N 1N 2N 3N) + vperm2i128 ymmH, ymmG, ymmF, 0x31 ; ymmH=(0O 1O 2O 3O 0P 1P 2P 3P 0Q 1Q 2Q 3Q 0R 1R 2R 3R + ; 0S 1S 2S 3S 0T 1T 2T 3T 0U 1U 2U 3U 0V 1V 2V 3V) + + cmp rcx, byte SIZEOF_YMMWORD + jb short .column_st64 + + test rdi, SIZEOF_YMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + vmovntdq YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovntdq YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovntdq YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC + vmovntdq YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymmC + vmovdqu YMMWORD [rdi+3*SIZEOF_YMMWORD], ymmH +.out0: + add rdi, RGB_PIXELSIZE*SIZEOF_YMMWORD ; outptr + sub rcx, byte SIZEOF_YMMWORD + jz near .endcolumn + + add rsi, byte SIZEOF_YMMWORD ; inptr0 + dec al + jnz near .Yloop_2nd + + add rbx, byte SIZEOF_YMMWORD ; inptr1 + add rdx, byte SIZEOF_YMMWORD ; inptr2 + jmp near .columnloop + +.column_st64: + cmp rcx, byte SIZEOF_YMMWORD/2 + jb short .column_st32 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymmD + add rdi, byte 2*SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmC + vmovdqa ymmD, ymmH + sub rcx, byte SIZEOF_YMMWORD/2 +.column_st32: + cmp rcx, byte SIZEOF_YMMWORD/4 + jb short .column_st16 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymmA + add rdi, byte SIZEOF_YMMWORD ; outptr + vmovdqa ymmA, ymmD + sub rcx, byte SIZEOF_YMMWORD/4 +.column_st16: + cmp rcx, byte SIZEOF_YMMWORD/8 + jb short .column_st15 + vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + add rdi, byte SIZEOF_XMMWORD ; outptr + vperm2i128 ymmA, ymmA, ymmA, 1 + sub rcx, byte SIZEOF_YMMWORD/8 +.column_st15: + ; Store two pixels (8 bytes) of ymmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_YMMWORD/16 + jb short .column_st7 + vmovq MMWORD [rdi], xmmA + add rdi, byte SIZEOF_YMMWORD/16*4 + sub rcx, byte SIZEOF_YMMWORD/16 + vpsrldq xmmA, SIZEOF_YMMWORD/16*4 +.column_st7: + ; Store one pixel (4 bytes) of ymmA to the output when it has enough + ; space. + test rcx, rcx + jz short .endcolumn + vmovd XMM_DWORD [rdi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + +.endcolumn: + sfence ; flush the write buffer + +.return: + pop rbx + vzeroupper + UNCOLLECT_ARGS 4 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v2_merged_upsample_avx2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +; r10d = JDIMENSION output_width +; r11 = JSAMPIMAGE input_buf +; r12d = JDIMENSION in_row_group_ctr +; r13 = JSAMPARRAY output_buf + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_merged_upsample_avx2) + +EXTN(jsimd_h2v2_merged_upsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + push rbx + + mov eax, r10d + + mov rdi, r11 + mov ecx, r12d + mov rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] + mov rdi, r13 + lea rsi, [rsi+rcx*SIZEOF_JSAMPROW] + + sub rsp, SIZEOF_JSAMPARRAY*4 + mov JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY], rsip ; intpr00 + mov JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY], rbxp ; intpr1 + mov JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY], rdxp ; intpr2 + mov rbx, rsp + + push rdi + push rcx + push rax + + %ifdef WIN64 + mov r8, rcx + mov r9, rdi + mov rcx, rax + mov rdx, rbx + %else + mov rdx, rcx + mov rcx, rdi + mov rdi, rax + mov rsi, rbx + %endif + + call EXTN(jsimd_h2v1_merged_upsample_avx2) + + pop rax + pop rcx + pop rdi + mov rsip, JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY] + + add rdi, byte SIZEOF_JSAMPROW ; outptr1 + add rsi, byte SIZEOF_JSAMPROW ; inptr01 + + mov JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY], rsip ; intpr00 + mov JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY], rbxp ; intpr1 + mov JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY], rdxp ; intpr2 + mov rbx, rsp + + push rdi + push rcx + push rax + + %ifdef WIN64 + mov r8, rcx + mov r9, rdi + mov rcx, rax + mov rdx, rbx + %else + mov rdx, rcx + mov rcx, rdi + mov rdi, rax + mov rsi, rbx + %endif + + call EXTN(jsimd_h2v1_merged_upsample_avx2) + + pop rax + pop rcx + pop rdi + mov rsip, JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY] + add rsp, SIZEOF_JSAMPARRAY*4 + + pop rbx + UNCOLLECT_ARGS 4 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-sse2.asm new file mode 100644 index 0000000..abd22e2 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdmrgext-sse2.asm @@ -0,0 +1,536 @@ +; +; jdmrgext.asm - merged upsampling/color conversion (64-bit SSE2) +; +; Copyright 2009, 2012 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2012, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jcolsamp.inc" + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 1:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v1_merged_upsample_sse2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +; r10d = JDIMENSION output_width +; r11 = JSAMPIMAGE input_buf +; r12d = JDIMENSION in_row_group_ctr +; r13 = JSAMPARRAY output_buf + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 3 + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_merged_upsample_sse2) + +EXTN(jsimd_h2v1_merged_upsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 4 + push rbx + + mov ecx, r10d ; col + test rcx, rcx + jz near .return + + push rcx + + mov rdi, r11 + mov ecx, r12d + mov rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] + mov rdi, r13 + mov rsip, JSAMPROW [rsi+rcx*SIZEOF_JSAMPROW] ; inptr0 + mov rbxp, JSAMPROW [rbx+rcx*SIZEOF_JSAMPROW] ; inptr1 + mov rdxp, JSAMPROW [rdx+rcx*SIZEOF_JSAMPROW] ; inptr2 + mov rdip, JSAMPROW [rdi] ; outptr + + pop rcx ; col + +.columnloop: + + movdqa xmm6, XMMWORD [rbx] ; xmm6=Cb(0123456789ABCDEF) + movdqa xmm7, XMMWORD [rdx] ; xmm7=Cr(0123456789ABCDEF) + + pxor xmm1, xmm1 ; xmm1=(all 0's) + pcmpeqw xmm3, xmm3 + psllw xmm3, 7 ; xmm3={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + movdqa xmm4, xmm6 + punpckhbw xmm6, xmm1 ; xmm6=Cb(89ABCDEF)=CbH + punpcklbw xmm4, xmm1 ; xmm4=Cb(01234567)=CbL + movdqa xmm0, xmm7 + punpckhbw xmm7, xmm1 ; xmm7=Cr(89ABCDEF)=CrH + punpcklbw xmm0, xmm1 ; xmm0=Cr(01234567)=CrL + + paddw xmm6, xmm3 + paddw xmm4, xmm3 + paddw xmm7, xmm3 + paddw xmm0, xmm3 + + ; (Original) + ; R = Y + 1.40200 * Cr + ; G = Y - 0.34414 * Cb - 0.71414 * Cr + ; B = Y + 1.77200 * Cb + ; + ; (This implementation) + ; R = Y + 0.40200 * Cr + Cr + ; G = Y - 0.34414 * Cb + 0.28586 * Cr - Cr + ; B = Y - 0.22800 * Cb + Cb + Cb + + movdqa xmm5, xmm6 ; xmm5=CbH + movdqa xmm2, xmm4 ; xmm2=CbL + paddw xmm6, xmm6 ; xmm6=2*CbH + paddw xmm4, xmm4 ; xmm4=2*CbL + movdqa xmm1, xmm7 ; xmm1=CrH + movdqa xmm3, xmm0 ; xmm3=CrL + paddw xmm7, xmm7 ; xmm7=2*CrH + paddw xmm0, xmm0 ; xmm0=2*CrL + + pmulhw xmm6, [rel PW_MF0228] ; xmm6=(2*CbH * -FIX(0.22800)) + pmulhw xmm4, [rel PW_MF0228] ; xmm4=(2*CbL * -FIX(0.22800)) + pmulhw xmm7, [rel PW_F0402] ; xmm7=(2*CrH * FIX(0.40200)) + pmulhw xmm0, [rel PW_F0402] ; xmm0=(2*CrL * FIX(0.40200)) + + paddw xmm6, [rel PW_ONE] + paddw xmm4, [rel PW_ONE] + psraw xmm6, 1 ; xmm6=(CbH * -FIX(0.22800)) + psraw xmm4, 1 ; xmm4=(CbL * -FIX(0.22800)) + paddw xmm7, [rel PW_ONE] + paddw xmm0, [rel PW_ONE] + psraw xmm7, 1 ; xmm7=(CrH * FIX(0.40200)) + psraw xmm0, 1 ; xmm0=(CrL * FIX(0.40200)) + + paddw xmm6, xmm5 + paddw xmm4, xmm2 + paddw xmm6, xmm5 ; xmm6=(CbH * FIX(1.77200))=(B-Y)H + paddw xmm4, xmm2 ; xmm4=(CbL * FIX(1.77200))=(B-Y)L + paddw xmm7, xmm1 ; xmm7=(CrH * FIX(1.40200))=(R-Y)H + paddw xmm0, xmm3 ; xmm0=(CrL * FIX(1.40200))=(R-Y)L + + movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=(B-Y)H + movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(R-Y)H + + movdqa xmm6, xmm5 + movdqa xmm7, xmm2 + punpcklwd xmm5, xmm1 + punpckhwd xmm6, xmm1 + pmaddwd xmm5, [rel PW_MF0344_F0285] + pmaddwd xmm6, [rel PW_MF0344_F0285] + punpcklwd xmm2, xmm3 + punpckhwd xmm7, xmm3 + pmaddwd xmm2, [rel PW_MF0344_F0285] + pmaddwd xmm7, [rel PW_MF0344_F0285] + + paddd xmm5, [rel PD_ONEHALF] + paddd xmm6, [rel PD_ONEHALF] + psrad xmm5, SCALEBITS + psrad xmm6, SCALEBITS + paddd xmm2, [rel PD_ONEHALF] + paddd xmm7, [rel PD_ONEHALF] + psrad xmm2, SCALEBITS + psrad xmm7, SCALEBITS + + packssdw xmm5, xmm6 ; xmm5=CbH*-FIX(0.344)+CrH*FIX(0.285) + packssdw xmm2, xmm7 ; xmm2=CbL*-FIX(0.344)+CrL*FIX(0.285) + psubw xmm5, xmm1 ; xmm5=CbH*-FIX(0.344)+CrH*-FIX(0.714)=(G-Y)H + psubw xmm2, xmm3 ; xmm2=CbL*-FIX(0.344)+CrL*-FIX(0.714)=(G-Y)L + + movdqa XMMWORD [wk(2)], xmm5 ; wk(2)=(G-Y)H + + mov al, 2 ; Yctr + jmp short .Yloop_1st + +.Yloop_2nd: + movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(R-Y)H + movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(G-Y)H + movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(B-Y)H + +.Yloop_1st: + movdqa xmm7, XMMWORD [rsi] ; xmm7=Y(0123456789ABCDEF) + + pcmpeqw xmm6, xmm6 + psrlw xmm6, BYTE_BIT ; xmm6={0xFF 0x00 0xFF 0x00 ..} + pand xmm6, xmm7 ; xmm6=Y(02468ACE)=YE + psrlw xmm7, BYTE_BIT ; xmm7=Y(13579BDF)=YO + + movdqa xmm1, xmm0 ; xmm1=xmm0=(R-Y)(L/H) + movdqa xmm3, xmm2 ; xmm3=xmm2=(G-Y)(L/H) + movdqa xmm5, xmm4 ; xmm5=xmm4=(B-Y)(L/H) + + paddw xmm0, xmm6 ; xmm0=((R-Y)+YE)=RE=R(02468ACE) + paddw xmm1, xmm7 ; xmm1=((R-Y)+YO)=RO=R(13579BDF) + packuswb xmm0, xmm0 ; xmm0=R(02468ACE********) + packuswb xmm1, xmm1 ; xmm1=R(13579BDF********) + + paddw xmm2, xmm6 ; xmm2=((G-Y)+YE)=GE=G(02468ACE) + paddw xmm3, xmm7 ; xmm3=((G-Y)+YO)=GO=G(13579BDF) + packuswb xmm2, xmm2 ; xmm2=G(02468ACE********) + packuswb xmm3, xmm3 ; xmm3=G(13579BDF********) + + paddw xmm4, xmm6 ; xmm4=((B-Y)+YE)=BE=B(02468ACE) + paddw xmm5, xmm7 ; xmm5=((B-Y)+YO)=BO=B(13579BDF) + packuswb xmm4, xmm4 ; xmm4=B(02468ACE********) + packuswb xmm5, xmm5 ; xmm5=B(13579BDF********) + +%if RGB_PIXELSIZE == 3 ; --------------- + + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(** ** ** ** ** ** ** ** **), xmmH=(** ** ** ** ** ** ** ** **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmB ; xmmE=(20 01 22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F) + punpcklbw xmmD, xmmF ; xmmD=(11 21 13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F) + + movdqa xmmG, xmmA + movdqa xmmH, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 01 02 12 22 03 04 14 24 05 06 16 26 07) + punpckhwd xmmG, xmmE ; xmmG=(08 18 28 09 0A 1A 2A 0B 0C 1C 2C 0D 0E 1E 2E 0F) + + psrldq xmmH, 2 ; xmmH=(02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E -- --) + psrldq xmmE, 2 ; xmmE=(22 03 24 05 26 07 28 09 2A 0B 2C 0D 2E 0F -- --) + + movdqa xmmC, xmmD + movdqa xmmB, xmmD + punpcklwd xmmD, xmmH ; xmmD=(11 21 02 12 13 23 04 14 15 25 06 16 17 27 08 18) + punpckhwd xmmC, xmmH ; xmmC=(19 29 0A 1A 1B 2B 0C 1C 1D 2D 0E 1E 1F 2F -- --) + + psrldq xmmB, 2 ; xmmB=(13 23 15 25 17 27 19 29 1B 2B 1D 2D 1F 2F -- --) + + movdqa xmmF, xmmE + punpcklwd xmmE, xmmB ; xmmE=(22 03 13 23 24 05 15 25 26 07 17 27 28 09 19 29) + punpckhwd xmmF, xmmB ; xmmF=(2A 0B 1B 2B 2C 0D 1D 2D 2E 0F 1F 2F -- -- -- --) + + pshufd xmmH, xmmA, 0x4E ; xmmH=(04 14 24 05 06 16 26 07 00 10 20 01 02 12 22 03) + movdqa xmmB, xmmE + punpckldq xmmA, xmmD ; xmmA=(00 10 20 01 11 21 02 12 02 12 22 03 13 23 04 14) + punpckldq xmmE, xmmH ; xmmE=(22 03 13 23 04 14 24 05 24 05 15 25 06 16 26 07) + punpckhdq xmmD, xmmB ; xmmD=(15 25 06 16 26 07 17 27 17 27 08 18 28 09 19 29) + + pshufd xmmH, xmmG, 0x4E ; xmmH=(0C 1C 2C 0D 0E 1E 2E 0F 08 18 28 09 0A 1A 2A 0B) + movdqa xmmB, xmmF + punpckldq xmmG, xmmC ; xmmG=(08 18 28 09 19 29 0A 1A 0A 1A 2A 0B 1B 2B 0C 1C) + punpckldq xmmF, xmmH ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 2C 0D 1D 2D 0E 1E 2E 0F) + punpckhdq xmmC, xmmB ; xmmC=(1D 2D 0E 1E 2E 0F 1F 2F 1F 2F -- -- -- -- -- --) + + punpcklqdq xmmA, xmmE ; xmmA=(00 10 20 01 11 21 02 12 22 03 13 23 04 14 24 05) + punpcklqdq xmmD, xmmG ; xmmD=(15 25 06 16 26 07 17 27 08 18 28 09 19 29 0A 1A) + punpcklqdq xmmF, xmmC ; xmmF=(2A 0B 1B 2B 0C 1C 2C 0D 1D 2D 0E 1E 2E 0F 1F 2F) + + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test rdi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmF +.out0: + add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub rcx, byte SIZEOF_XMMWORD + jz near .endcolumn + + add rsi, byte SIZEOF_XMMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add rbx, byte SIZEOF_XMMWORD ; inptr1 + add rdx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + +.column_st32: + lea rcx, [rcx+rcx*2] ; imul ecx, RGB_PIXELSIZE + cmp rcx, byte 2*SIZEOF_XMMWORD + jb short .column_st16 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + add rdi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmF + sub rcx, byte 2*SIZEOF_XMMWORD + jmp short .column_st15 +.column_st16: + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st15 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + add rdi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub rcx, byte SIZEOF_XMMWORD +.column_st15: + ; Store the lower 8 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_MMWORD + jb short .column_st7 + movq XMM_MMWORD [rdi], xmmA + add rdi, byte SIZEOF_MMWORD + sub rcx, byte SIZEOF_MMWORD + psrldq xmmA, SIZEOF_MMWORD +.column_st7: + ; Store the lower 4 bytes of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_DWORD + jb short .column_st3 + movd XMM_DWORD [rdi], xmmA + add rdi, byte SIZEOF_DWORD + sub rcx, byte SIZEOF_DWORD + psrldq xmmA, SIZEOF_DWORD +.column_st3: + ; Store the lower 2 bytes of rax to the output when it has enough + ; space. + movd eax, xmmA + cmp rcx, byte SIZEOF_WORD + jb short .column_st1 + mov word [rdi], ax + add rdi, byte SIZEOF_WORD + sub rcx, byte SIZEOF_WORD + shr rax, 16 +.column_st1: + ; Store the lower 1 byte of rax to the output when it has enough + ; space. + test rcx, rcx + jz short .endcolumn + mov byte [rdi], al + +%else ; RGB_PIXELSIZE == 4 ; ----------- + +%ifdef RGBX_FILLER_0XFF + pcmpeqb xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pcmpeqb xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%else + pxor xmm6, xmm6 ; xmm6=XE=X(02468ACE********) + pxor xmm7, xmm7 ; xmm7=XO=X(13579BDF********) +%endif + ; xmmA=(00 02 04 06 08 0A 0C 0E **), xmmB=(01 03 05 07 09 0B 0D 0F **) + ; xmmC=(10 12 14 16 18 1A 1C 1E **), xmmD=(11 13 15 17 19 1B 1D 1F **) + ; xmmE=(20 22 24 26 28 2A 2C 2E **), xmmF=(21 23 25 27 29 2B 2D 2F **) + ; xmmG=(30 32 34 36 38 3A 3C 3E **), xmmH=(31 33 35 37 39 3B 3D 3F **) + + punpcklbw xmmA, xmmC ; xmmA=(00 10 02 12 04 14 06 16 08 18 0A 1A 0C 1C 0E 1E) + punpcklbw xmmE, xmmG ; xmmE=(20 30 22 32 24 34 26 36 28 38 2A 3A 2C 3C 2E 3E) + punpcklbw xmmB, xmmD ; xmmB=(01 11 03 13 05 15 07 17 09 19 0B 1B 0D 1D 0F 1F) + punpcklbw xmmF, xmmH ; xmmF=(21 31 23 33 25 35 27 37 29 39 2B 3B 2D 3D 2F 3F) + + movdqa xmmC, xmmA + punpcklwd xmmA, xmmE ; xmmA=(00 10 20 30 02 12 22 32 04 14 24 34 06 16 26 36) + punpckhwd xmmC, xmmE ; xmmC=(08 18 28 38 0A 1A 2A 3A 0C 1C 2C 3C 0E 1E 2E 3E) + movdqa xmmG, xmmB + punpcklwd xmmB, xmmF ; xmmB=(01 11 21 31 03 13 23 33 05 15 25 35 07 17 27 37) + punpckhwd xmmG, xmmF ; xmmG=(09 19 29 39 0B 1B 2B 3B 0D 1D 2D 3D 0F 1F 2F 3F) + + movdqa xmmD, xmmA + punpckldq xmmA, xmmB ; xmmA=(00 10 20 30 01 11 21 31 02 12 22 32 03 13 23 33) + punpckhdq xmmD, xmmB ; xmmD=(04 14 24 34 05 15 25 35 06 16 26 36 07 17 27 37) + movdqa xmmH, xmmC + punpckldq xmmC, xmmG ; xmmC=(08 18 28 38 09 19 29 39 0A 1A 2A 3A 0B 1B 2B 3B) + punpckhdq xmmH, xmmG ; xmmH=(0C 1C 2C 3C 0D 1D 2D 3D 0E 1E 2E 3E 0F 1F 2F 3F) + + cmp rcx, byte SIZEOF_XMMWORD + jb short .column_st32 + + test rdi, SIZEOF_XMMWORD-1 + jnz short .out1 + ; --(aligned)------------------- + movntdq XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movntdq XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movntdq XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC + movntdq XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH + jmp short .out0 +.out1: ; --(unaligned)----------------- + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + movdqu XMMWORD [rdi+2*SIZEOF_XMMWORD], xmmC + movdqu XMMWORD [rdi+3*SIZEOF_XMMWORD], xmmH +.out0: + add rdi, byte RGB_PIXELSIZE*SIZEOF_XMMWORD ; outptr + sub rcx, byte SIZEOF_XMMWORD + jz near .endcolumn + + add rsi, byte SIZEOF_XMMWORD ; inptr0 + dec al ; Yctr + jnz near .Yloop_2nd + + add rbx, byte SIZEOF_XMMWORD ; inptr1 + add rdx, byte SIZEOF_XMMWORD ; inptr2 + jmp near .columnloop + +.column_st32: + cmp rcx, byte SIZEOF_XMMWORD/2 + jb short .column_st16 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + movdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmmD + add rdi, byte 2*SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmC + movdqa xmmD, xmmH + sub rcx, byte SIZEOF_XMMWORD/2 +.column_st16: + cmp rcx, byte SIZEOF_XMMWORD/4 + jb short .column_st15 + movdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmmA + add rdi, byte SIZEOF_XMMWORD ; outptr + movdqa xmmA, xmmD + sub rcx, byte SIZEOF_XMMWORD/4 +.column_st15: + ; Store two pixels (8 bytes) of xmmA to the output when it has enough + ; space. + cmp rcx, byte SIZEOF_XMMWORD/8 + jb short .column_st7 + movq XMM_MMWORD [rdi], xmmA + add rdi, byte SIZEOF_XMMWORD/8*4 + sub rcx, byte SIZEOF_XMMWORD/8 + psrldq xmmA, SIZEOF_XMMWORD/8*4 +.column_st7: + ; Store one pixel (4 bytes) of xmmA to the output when it has enough + ; space. + test rcx, rcx + jz short .endcolumn + movd XMM_DWORD [rdi], xmmA + +%endif ; RGB_PIXELSIZE ; --------------- + +.endcolumn: + sfence ; flush the write buffer + +.return: + pop rbx + UNCOLLECT_ARGS 4 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Upsample and color convert for the case of 2:1 horizontal and 2:1 vertical. +; +; GLOBAL(void) +; jsimd_h2v2_merged_upsample_sse2(JDIMENSION output_width, +; JSAMPIMAGE input_buf, +; JDIMENSION in_row_group_ctr, +; JSAMPARRAY output_buf); +; + +; r10d = JDIMENSION output_width +; r11 = JSAMPIMAGE input_buf +; r12d = JDIMENSION in_row_group_ctr +; r13 = JSAMPARRAY output_buf + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_merged_upsample_sse2) + +EXTN(jsimd_h2v2_merged_upsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + push rbx + + mov eax, r10d + + mov rdi, r11 + mov ecx, r12d + mov rsip, JSAMPARRAY [rdi+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rdi+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rdi+2*SIZEOF_JSAMPARRAY] + mov rdi, r13 + lea rsi, [rsi+rcx*SIZEOF_JSAMPROW] + + sub rsp, SIZEOF_JSAMPARRAY*4 + mov JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY], rsip ; intpr00 + mov JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY], rbxp ; intpr1 + mov JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY], rdxp ; intpr2 + mov rbx, rsp + + push rdi + push rcx + push rax + + %ifdef WIN64 + mov r8, rcx + mov r9, rdi + mov rcx, rax + mov rdx, rbx + %else + mov rdx, rcx + mov rcx, rdi + mov rdi, rax + mov rsi, rbx + %endif + + call EXTN(jsimd_h2v1_merged_upsample_sse2) + + pop rax + pop rcx + pop rdi + mov rsip, JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY] + + add rdi, byte SIZEOF_JSAMPROW ; outptr1 + add rsi, byte SIZEOF_JSAMPROW ; inptr01 + + mov JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY], rsip ; intpr00 + mov JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY], rbxp ; intpr1 + mov JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY], rdxp ; intpr2 + mov rbx, rsp + + push rdi + push rcx + push rax + + %ifdef WIN64 + mov r8, rcx + mov r9, rdi + mov rcx, rax + mov rdx, rbx + %else + mov rdx, rcx + mov rcx, rdi + mov rdi, rax + mov rsi, rbx + %endif + + call EXTN(jsimd_h2v1_merged_upsample_sse2) + + pop rax + pop rcx + pop rdi + mov rsip, JSAMPARRAY [rsp+0*SIZEOF_JSAMPARRAY] + mov rbxp, JSAMPARRAY [rsp+1*SIZEOF_JSAMPARRAY] + mov rdxp, JSAMPARRAY [rsp+2*SIZEOF_JSAMPARRAY] + add rsp, SIZEOF_JSAMPARRAY*4 + + pop rbx + UNCOLLECT_ARGS 4 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-avx2.asm new file mode 100644 index 0000000..6ae4cf8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-avx2.asm @@ -0,0 +1,694 @@ +; +; jdsample.asm - upsampling (64-bit AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2015, Intel Corporation. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fancy_upsample_avx2) + +EXTN(jconst_fancy_upsample_avx2): + +PW_ONE times 16 dw 1 +PW_TWO times 16 dw 2 +PW_THREE times 16 dw 3 +PW_SEVEN times 16 dw 7 +PW_EIGHT times 16 dw 8 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical. +; +; The upsampling algorithm is linear interpolation between pixel centers, +; also known as a "triangle filter". This is a good compromise between +; speed and visual quality. The centers of the output pixels are 1/4 and 3/4 +; of the way between input pixel centers. +; +; GLOBAL(void) +; jsimd_h2v1_fancy_upsample_avx2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION downsampled_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_fancy_upsample_avx2) + +EXTN(jsimd_h2v1_fancy_upsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + PUSH_XMM 3 + COLLECT_ARGS 4 + + mov eax, r11d ; colctr + test rax, rax + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz near .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data + + vpxor ymm0, ymm0, ymm0 ; ymm0=(all 0's) + vpcmpeqb xmm9, xmm9, xmm9 + vpsrldq xmm10, xmm9, (SIZEOF_XMMWORD-1) ; (ff -- -- -- ... -- --) LSB is ff + + vpslldq xmm9, xmm9, (SIZEOF_XMMWORD-1) + vperm2i128 ymm9, ymm9, ymm9, 1 ; (---- ---- ... ---- ---- ff) MSB is ff + +.rowloop: + push rax ; colctr + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr + + test rax, SIZEOF_YMMWORD-1 + jz short .skip + mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample +.skip: + vpand ymm7, ymm10, YMMWORD [rsi+0*SIZEOF_YMMWORD] + + add rax, byte SIZEOF_YMMWORD-1 + and rax, byte -SIZEOF_YMMWORD + cmp rax, byte SIZEOF_YMMWORD + ja short .columnloop + +.columnloop_last: + vpand ymm6, ymm9, YMMWORD [rsi+0*SIZEOF_YMMWORD] + jmp short .upsample + +.columnloop: + vmovdqu ymm6, YMMWORD [rsi+1*SIZEOF_YMMWORD] + vperm2i128 ymm6, ymm0, ymm6, 0x20 + vpslldq ymm6, ymm6, 15 + +.upsample: + vmovdqu ymm1, YMMWORD [rsi+0*SIZEOF_YMMWORD] ; ymm1=( 0 1 2 ... 29 30 31) + + vperm2i128 ymm2, ymm0, ymm1, 0x20 + vpalignr ymm2, ymm1, ymm2, 15 ; ymm2=(-- 0 1 ... 28 29 30) + vperm2i128 ymm4, ymm0, ymm1, 0x03 + vpalignr ymm3, ymm4, ymm1, 1 ; ymm3=( 1 2 3 ... 30 31 --) + + vpor ymm2, ymm2, ymm7 ; ymm2=(-1 0 1 ... 28 29 30) + vpor ymm3, ymm3, ymm6 ; ymm3=( 1 2 3 ... 30 31 32) + + vpsrldq ymm7, ymm4, (SIZEOF_XMMWORD-1) ; ymm7=(31 -- -- ... -- -- --) + + vpunpckhbw ymm4, ymm1, ymm0 ; ymm4=( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm5, ymm1, ymm0 ; ymm5=( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm1, ymm5, ymm4, 0x20 ; ymm1=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm5, ymm2, ymm0 ; ymm5=( 7 8 9 10 11 12 13 14 23 24 25 26 27 28 29 30) + vpunpcklbw ymm6, ymm2, ymm0 ; ymm6=(-1 0 1 2 3 4 5 6 15 16 17 18 19 20 21 22) + vperm2i128 ymm2, ymm6, ymm5, 0x20 ; ymm2=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vpunpckhbw ymm6, ymm3, ymm0 ; ymm6=( 1 2 3 4 5 6 7 8 17 18 19 20 21 22 23 24) + vpunpcklbw ymm8, ymm3, ymm0 ; ymm8=( 9 10 11 12 13 14 15 16 25 26 27 28 29 30 31 32) + vperm2i128 ymm3, ymm8, ymm6, 0x20 ; ymm3=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16) + vperm2i128 ymm6, ymm8, ymm6, 0x31 ; ymm6=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32) + + vpmullw ymm1, ymm1, [rel PW_THREE] + vpmullw ymm4, ymm4, [rel PW_THREE] + vpaddw ymm2, ymm2, [rel PW_ONE] + vpaddw ymm5, ymm5, [rel PW_ONE] + vpaddw ymm3, ymm3, [rel PW_TWO] + vpaddw ymm6, ymm6, [rel PW_TWO] + + vpaddw ymm2, ymm2, ymm1 + vpaddw ymm5, ymm5, ymm4 + vpsrlw ymm2, ymm2, 2 ; ymm2=OutLE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30) + vpsrlw ymm5, ymm5, 2 ; ymm5=OutHE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62) + vpaddw ymm3, ymm3, ymm1 + vpaddw ymm6, ymm6, ymm4 + vpsrlw ymm3, ymm3, 2 ; ymm3=OutLO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31) + vpsrlw ymm6, ymm6, 2 ; ymm6=OutHO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63) + + vpsllw ymm3, ymm3, BYTE_BIT + vpsllw ymm6, ymm6, BYTE_BIT + vpor ymm2, ymm2, ymm3 ; ymm2=OutL=( 0 1 2 ... 29 30 31) + vpor ymm5, ymm5, ymm6 ; ymm5=OutH=(32 33 34 ... 61 62 63) + + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm2 + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm5 + + sub rax, byte SIZEOF_YMMWORD + add rsi, byte 1*SIZEOF_YMMWORD ; inptr + add rdi, byte 2*SIZEOF_YMMWORD ; outptr + cmp rax, byte SIZEOF_YMMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop rsi + pop rdi + pop rax + + add rsi, byte SIZEOF_JSAMPROW ; input_data + add rdi, byte SIZEOF_JSAMPROW ; output_data + dec rcx ; rowctr + jg near .rowloop + +.return: + vzeroupper + UNCOLLECT_ARGS 4 + POP_XMM 3 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical. +; Again a triangle filter; see comments for h2v1 case, above. +; +; GLOBAL(void) +; jsimd_h2v2_fancy_upsample_avx2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION downsampled_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_YMMWORD ; ymmword wk[WK_NUM] +%define WK_NUM 4 + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_fancy_upsample_avx2) + +EXTN(jsimd_h2v2_fancy_upsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_YMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, (SIZEOF_YMMWORD * WK_NUM) + PUSH_XMM 3 + COLLECT_ARGS 4 + push rbx + + mov eax, r11d ; colctr + test rax, rax + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz near .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data +.rowloop: + push rax ; colctr + push rcx + push rdi + push rsi + + mov rcxp, JSAMPROW [rsi-1*SIZEOF_JSAMPROW] ; inptr1(above) + mov rbxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0 + mov rsip, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1(below) + mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0 + mov rdip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1 + + vpxor ymm8, ymm8, ymm8 ; ymm8=(all 0's) + vpcmpeqb xmm9, xmm9, xmm9 + vpsrldq xmm10, xmm9, (SIZEOF_XMMWORD-2) ; (ffff ---- ---- ... ---- ----) LSB is ffff + vpslldq xmm9, xmm9, (SIZEOF_XMMWORD-2) + vperm2i128 ymm9, ymm9, ymm9, 1 ; (---- ---- ... ---- ---- ffff) MSB is ffff + + test rax, SIZEOF_YMMWORD-1 + jz short .skip + push rdx + mov dl, JSAMPLE [rcx+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rcx+rax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [rbx+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rbx+rax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample + pop rdx +.skip: + ; -- process the first column block + + vmovdqu ymm0, YMMWORD [rbx+0*SIZEOF_YMMWORD] ; ymm0=row[ 0][0] + vmovdqu ymm1, YMMWORD [rcx+0*SIZEOF_YMMWORD] ; ymm1=row[-1][0] + vmovdqu ymm2, YMMWORD [rsi+0*SIZEOF_YMMWORD] ; ymm2=row[+1][0] + + vpunpckhbw ymm4, ymm0, ymm8 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm5, ymm0, ymm8 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm5, ymm1, ymm8 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm6, ymm1, ymm8 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm6, ymm2, ymm8 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm3, ymm2, ymm8 ; ymm3=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm2, ymm3, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm6, ymm3, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpmullw ymm0, ymm0, [rel PW_THREE] + vpmullw ymm4, ymm4, [rel PW_THREE] + + vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vmovdqu YMMWORD [rdx+0*SIZEOF_YMMWORD], ymm1 ; temporarily save + vmovdqu YMMWORD [rdx+1*SIZEOF_YMMWORD], ymm5 ; the intermediate data + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm2 + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm6 + + vpand ymm1, ymm1, ymm10 ; ymm1=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vpand ymm2, ymm2, ymm10 ; ymm2=( 0 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + + vmovdqa YMMWORD [wk(0)], ymm1 + vmovdqa YMMWORD [wk(1)], ymm2 + + add rax, byte SIZEOF_YMMWORD-1 + and rax, byte -SIZEOF_YMMWORD + cmp rax, byte SIZEOF_YMMWORD + ja short .columnloop + +.columnloop_last: + ; -- process the last column block + + vpand ymm1, ymm9, YMMWORD [rdx+1*SIZEOF_YMMWORD] + vpand ymm2, ymm9, YMMWORD [rdi+1*SIZEOF_YMMWORD] + + vmovdqa YMMWORD [wk(2)], ymm1 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31) + vmovdqa YMMWORD [wk(3)], ymm2 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 31) + + jmp near .upsample + +.columnloop: + ; -- process the next column block + + vmovdqu ymm0, YMMWORD [rbx+1*SIZEOF_YMMWORD] ; ymm0=row[ 0][1] + vmovdqu ymm1, YMMWORD [rcx+1*SIZEOF_YMMWORD] ; ymm1=row[-1][1] + vmovdqu ymm2, YMMWORD [rsi+1*SIZEOF_YMMWORD] ; ymm2=row[+1][1] + + vpunpckhbw ymm4, ymm0, ymm8 ; ymm4=row[ 0]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm5, ymm0, ymm8 ; ymm5=row[ 0]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm0, ymm5, ymm4, 0x20 ; ymm0=row[ 0]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm4, ymm5, ymm4, 0x31 ; ymm4=row[ 0](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm5, ymm1, ymm8 ; ymm5=row[-1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm6, ymm1, ymm8 ; ymm6=row[-1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm1, ymm6, ymm5, 0x20 ; ymm1=row[-1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm5, ymm6, ymm5, 0x31 ; ymm5=row[-1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpunpckhbw ymm6, ymm2, ymm8 ; ymm6=row[+1]( 8 9 10 11 12 13 14 15 24 25 26 27 28 29 30 31) + vpunpcklbw ymm7, ymm2, ymm8 ; ymm7=row[+1]( 0 1 2 3 4 5 6 7 16 17 18 19 20 21 22 23) + vperm2i128 ymm2, ymm7, ymm6, 0x20 ; ymm2=row[+1]( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vperm2i128 ymm6, ymm7, ymm6, 0x31 ; ymm6=row[+1](16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vpmullw ymm0, ymm0, [rel PW_THREE] + vpmullw ymm4, ymm4, [rel PW_THREE] + + vpaddw ymm1, ymm1, ymm0 ; ymm1=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm5, ymm5, ymm4 ; ymm5=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + vpaddw ymm2, ymm2, ymm0 ; ymm2=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vpaddw ymm6, ymm6, ymm4 ; ymm6=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vmovdqu YMMWORD [rdx+2*SIZEOF_YMMWORD], ymm1 ; temporarily save + vmovdqu YMMWORD [rdx+3*SIZEOF_YMMWORD], ymm5 ; the intermediate data + vmovdqu YMMWORD [rdi+2*SIZEOF_YMMWORD], ymm2 + vmovdqu YMMWORD [rdi+3*SIZEOF_YMMWORD], ymm6 + + vperm2i128 ymm1, ymm8, ymm1, 0x20 + vpslldq ymm1, ymm1, 14 ; ymm1=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0) + vperm2i128 ymm2, ymm8, ymm2, 0x20 + vpslldq ymm2, ymm2, 14 ; ymm2=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 0) + + vmovdqa YMMWORD [wk(2)], ymm1 + vmovdqa YMMWORD [wk(3)], ymm2 + +.upsample: + ; -- process the upper row + + vmovdqu ymm7, YMMWORD [rdx+0*SIZEOF_YMMWORD] ; ymm7=Int0L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vmovdqu ymm3, YMMWORD [rdx+1*SIZEOF_YMMWORD] ; ymm3=Int0H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vperm2i128 ymm0, ymm8, ymm7, 0x03 + vpalignr ymm0, ymm0, ymm7, 2 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --) + vperm2i128 ymm4, ymm8, ymm3, 0x20 + vpslldq ymm4, ymm4, 14 ; ymm4=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16) + + vperm2i128 ymm5, ymm8, ymm7, 0x03 + vpsrldq ymm5, ymm5, 14 ; ymm5=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm6, ymm8, ymm3, 0x20 + vpalignr ymm6, ymm3, ymm6, 14 ; ymm6=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vpor ymm0, ymm0, ymm4 ; ymm0=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16) + vpor ymm5, ymm5, ymm6 ; ymm5=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vperm2i128 ymm2, ymm8, ymm3, 0x03 + vpalignr ymm2, ymm2, ymm3, 2 ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --) + vperm2i128 ymm4, ymm8, ymm3, 0x03 + vpsrldq ymm4, ymm4, 14 ; ymm4=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm1, ymm8, ymm7, 0x20 + vpalignr ymm1, ymm7, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + + vpor ymm1, ymm1, YMMWORD [wk(0)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + vpor ymm2, ymm2, YMMWORD [wk(2)] ; ymm2=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32) + + vmovdqa YMMWORD [wk(0)], ymm4 + + vpmullw ymm7, ymm7, [rel PW_THREE] + vpmullw ymm3, ymm3, [rel PW_THREE] + vpaddw ymm1, ymm1, [rel PW_EIGHT] + vpaddw ymm5, ymm5, [rel PW_EIGHT] + vpaddw ymm0, ymm0, [rel PW_SEVEN] + vpaddw ymm2, [rel PW_SEVEN] + + vpaddw ymm1, ymm1, ymm7 + vpaddw ymm5, ymm5, ymm3 + vpsrlw ymm1, ymm1, 4 ; ymm1=Out0LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30) + vpsrlw ymm5, ymm5, 4 ; ymm5=Out0HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62) + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm2, ymm2, ymm3 + vpsrlw ymm0, ymm0, 4 ; ymm0=Out0LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31) + vpsrlw ymm2, ymm2, 4 ; ymm2=Out0HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63) + + vpsllw ymm0, ymm0, BYTE_BIT + vpsllw ymm2, ymm2, BYTE_BIT + vpor ymm1, ymm1, ymm0 ; ymm1=Out0L=( 0 1 2 ... 29 30 31) + vpor ymm5, ymm5, ymm2 ; ymm5=Out0H=(32 33 34 ... 61 62 63) + + vmovdqu YMMWORD [rdx+0*SIZEOF_YMMWORD], ymm1 + vmovdqu YMMWORD [rdx+1*SIZEOF_YMMWORD], ymm5 + + ; -- process the lower row + + vmovdqu ymm6, YMMWORD [rdi+0*SIZEOF_YMMWORD] ; ymm6=Int1L=( 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15) + vmovdqu ymm4, YMMWORD [rdi+1*SIZEOF_YMMWORD] ; ymm4=Int1H=(16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31) + + vperm2i128 ymm7, ymm8, ymm6, 0x03 + vpalignr ymm7, ymm7, ymm6, 2 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 --) + vperm2i128 ymm3, ymm8, ymm4, 0x20 + vpslldq ymm3, ymm3, 14 ; ymm3=(-- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 16) + + vperm2i128 ymm0, ymm8, ymm6, 0x03 + vpsrldq ymm0, ymm0, 14 ; ymm0=(15 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm2, ymm8, ymm4, 0x20 + vpalignr ymm2, ymm4, ymm2, 14 ; ymm2=(-- 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vpor ymm7, ymm7, ymm3 ; ymm7=( 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16) + vpor ymm0, ymm0, ymm2 ; ymm0=(15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30) + + vperm2i128 ymm5, ymm8, ymm4, 0x03 + vpalignr ymm5, ymm5, ymm4, 2 ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 --) + vperm2i128 ymm3, ymm8, ymm4, 0x03 + vpsrldq ymm3, ymm3, 14 ; ymm3=(31 -- -- -- -- -- -- -- -- -- -- -- -- -- -- --) + vperm2i128 ymm1, ymm8, ymm6, 0x20 + vpalignr ymm1, ymm6, ymm1, 14 ; ymm1=(-- 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + + vpor ymm1, ymm1, YMMWORD [wk(1)] ; ymm1=(-1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14) + vpor ymm5, ymm5, YMMWORD [wk(3)] ; ymm5=(17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32) + + vmovdqa YMMWORD [wk(1)], ymm3 + + vpmullw ymm6, ymm6, [rel PW_THREE] + vpmullw ymm4, ymm4, [rel PW_THREE] + vpaddw ymm1, ymm1, [rel PW_EIGHT] + vpaddw ymm0, ymm0, [rel PW_EIGHT] + vpaddw ymm7, ymm7, [rel PW_SEVEN] + vpaddw ymm5, ymm5, [rel PW_SEVEN] + + vpaddw ymm1, ymm1, ymm6 + vpaddw ymm0, ymm0, ymm4 + vpsrlw ymm1, ymm1, 4 ; ymm1=Out1LE=( 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30) + vpsrlw ymm0, ymm0, 4 ; ymm0=Out1HE=(32 34 36 38 40 42 44 46 48 50 52 54 56 58 60 62) + vpaddw ymm7, ymm7, ymm6 + vpaddw ymm5, ymm5, ymm4 + vpsrlw ymm7, ymm7, 4 ; ymm7=Out1LO=( 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31) + vpsrlw ymm5, ymm5, 4 ; ymm5=Out1HO=(33 35 37 39 41 43 45 47 49 51 53 55 57 59 61 63) + + vpsllw ymm7, ymm7, BYTE_BIT + vpsllw ymm5, ymm5, BYTE_BIT + vpor ymm1, ymm1, ymm7 ; ymm1=Out1L=( 0 1 2 ... 29 30 31) + vpor ymm0, ymm0, ymm5 ; ymm0=Out1H=(32 33 34 ... 61 62 63) + + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm1 + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm0 + + sub rax, byte SIZEOF_YMMWORD + add rcx, byte 1*SIZEOF_YMMWORD ; inptr1(above) + add rbx, byte 1*SIZEOF_YMMWORD ; inptr0 + add rsi, byte 1*SIZEOF_YMMWORD ; inptr1(below) + add rdx, byte 2*SIZEOF_YMMWORD ; outptr0 + add rdi, byte 2*SIZEOF_YMMWORD ; outptr1 + cmp rax, byte SIZEOF_YMMWORD + ja near .columnloop + test rax, rax + jnz near .columnloop_last + + pop rsi + pop rdi + pop rcx + pop rax + + add rsi, byte 1*SIZEOF_JSAMPROW ; input_data + add rdi, byte 2*SIZEOF_JSAMPROW ; output_data + sub rcx, byte 2 ; rowctr + jg near .rowloop + +.return: + pop rbx + vzeroupper + UNCOLLECT_ARGS 4 + POP_XMM 3 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 1:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v1_upsample_avx2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION output_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_upsample_avx2) + +EXTN(jsimd_h2v1_upsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + + mov edx, r11d + add rdx, byte (SIZEOF_YMMWORD-1) + and rdx, -SIZEOF_YMMWORD + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz short .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data +.rowloop: + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr + mov rax, rdx ; colctr +.columnloop: + + cmp rax, byte SIZEOF_YMMWORD + ja near .above_16 + + vmovdqu xmm0, XMMWORD [rsi+0*SIZEOF_YMMWORD] + vpunpckhbw xmm1, xmm0, xmm0 + vpunpcklbw xmm0, xmm0, xmm0 + + vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0 + vmovdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1 + + jmp short .nextrow + +.above_16: + vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD] + + vpermq ymm0, ymm0, 0xd8 + vpunpckhbw ymm1, ymm0, ymm0 + vpunpcklbw ymm0, ymm0, ymm0 + + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0 + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm1 + + sub rax, byte 2*SIZEOF_YMMWORD + jz short .nextrow + + add rsi, byte SIZEOF_YMMWORD ; inptr + add rdi, byte 2*SIZEOF_YMMWORD ; outptr + jmp short .columnloop + +.nextrow: + pop rsi + pop rdi + + add rsi, byte SIZEOF_JSAMPROW ; input_data + add rdi, byte SIZEOF_JSAMPROW ; output_data + dec rcx ; rowctr + jg short .rowloop + +.return: + vzeroupper + UNCOLLECT_ARGS 4 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 2:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v2_upsample_avx2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION output_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_upsample_avx2) + +EXTN(jsimd_h2v2_upsample_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + push rbx + + mov edx, r11d + add rdx, byte (SIZEOF_YMMWORD-1) + and rdx, -SIZEOF_YMMWORD + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz near .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data +.rowloop: + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rbxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0 + mov rdip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1 + mov rax, rdx ; colctr +.columnloop: + + cmp rax, byte SIZEOF_YMMWORD + ja short .above_16 + + vmovdqu xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD] + vpunpckhbw xmm1, xmm0, xmm0 + vpunpcklbw xmm0, xmm0, xmm0 + + vmovdqu XMMWORD [rbx+0*SIZEOF_XMMWORD], xmm0 + vmovdqu XMMWORD [rbx+1*SIZEOF_XMMWORD], xmm1 + vmovdqu XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0 + vmovdqu XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1 + + jmp near .nextrow + +.above_16: + vmovdqu ymm0, YMMWORD [rsi+0*SIZEOF_YMMWORD] + + vpermq ymm0, ymm0, 0xd8 + vpunpckhbw ymm1, ymm0, ymm0 + vpunpcklbw ymm0, ymm0, ymm0 + + vmovdqu YMMWORD [rbx+0*SIZEOF_YMMWORD], ymm0 + vmovdqu YMMWORD [rbx+1*SIZEOF_YMMWORD], ymm1 + vmovdqu YMMWORD [rdi+0*SIZEOF_YMMWORD], ymm0 + vmovdqu YMMWORD [rdi+1*SIZEOF_YMMWORD], ymm1 + + sub rax, byte 2*SIZEOF_YMMWORD + jz short .nextrow + + add rsi, byte SIZEOF_YMMWORD ; inptr + add rbx, 2*SIZEOF_YMMWORD ; outptr0 + add rdi, 2*SIZEOF_YMMWORD ; outptr1 + jmp short .columnloop + +.nextrow: + pop rsi + pop rdi + + add rsi, byte 1*SIZEOF_JSAMPROW ; input_data + add rdi, byte 2*SIZEOF_JSAMPROW ; output_data + sub rcx, byte 2 ; rowctr + jg near .rowloop + +.return: + pop rbx + vzeroupper + UNCOLLECT_ARGS 4 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-sse2.asm new file mode 100644 index 0000000..54c560f --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jdsample-sse2.asm @@ -0,0 +1,663 @@ +; +; jdsample.asm - upsampling (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fancy_upsample_sse2) + +EXTN(jconst_fancy_upsample_sse2): + +PW_ONE times 8 dw 1 +PW_TWO times 8 dw 2 +PW_THREE times 8 dw 3 +PW_SEVEN times 8 dw 7 +PW_EIGHT times 8 dw 8 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Fancy processing for the common case of 2:1 horizontal and 1:1 vertical. +; +; The upsampling algorithm is linear interpolation between pixel centers, +; also known as a "triangle filter". This is a good compromise between +; speed and visual quality. The centers of the output pixels are 1/4 and 3/4 +; of the way between input pixel centers. +; +; GLOBAL(void) +; jsimd_h2v1_fancy_upsample_sse2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION downsampled_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_fancy_upsample_sse2) + +EXTN(jsimd_h2v1_fancy_upsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + + mov eax, r11d ; colctr + test rax, rax + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz near .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data +.rowloop: + push rax ; colctr + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr + + test rax, SIZEOF_XMMWORD-1 + jz short .skip + mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample +.skip: + pxor xmm0, xmm0 ; xmm0=(all 0's) + pcmpeqb xmm7, xmm7 + psrldq xmm7, (SIZEOF_XMMWORD-1) + pand xmm7, XMMWORD [rsi+0*SIZEOF_XMMWORD] + + add rax, byte SIZEOF_XMMWORD-1 + and rax, byte -SIZEOF_XMMWORD + cmp rax, byte SIZEOF_XMMWORD + ja short .columnloop + +.columnloop_last: + pcmpeqb xmm6, xmm6 + pslldq xmm6, (SIZEOF_XMMWORD-1) + pand xmm6, XMMWORD [rsi+0*SIZEOF_XMMWORD] + jmp short .upsample + +.columnloop: + movdqa xmm6, XMMWORD [rsi+1*SIZEOF_XMMWORD] + pslldq xmm6, (SIZEOF_XMMWORD-1) + +.upsample: + movdqa xmm1, XMMWORD [rsi+0*SIZEOF_XMMWORD] + movdqa xmm2, xmm1 + movdqa xmm3, xmm1 ; xmm1=( 0 1 2 ... 13 14 15) + pslldq xmm2, 1 ; xmm2=(-- 0 1 ... 12 13 14) + psrldq xmm3, 1 ; xmm3=( 1 2 3 ... 14 15 --) + + por xmm2, xmm7 ; xmm2=(-1 0 1 ... 12 13 14) + por xmm3, xmm6 ; xmm3=( 1 2 3 ... 14 15 16) + + movdqa xmm7, xmm1 + psrldq xmm7, (SIZEOF_XMMWORD-1) ; xmm7=(15 -- -- ... -- -- --) + + movdqa xmm4, xmm1 + punpcklbw xmm1, xmm0 ; xmm1=( 0 1 2 3 4 5 6 7) + punpckhbw xmm4, xmm0 ; xmm4=( 8 9 10 11 12 13 14 15) + movdqa xmm5, xmm2 + punpcklbw xmm2, xmm0 ; xmm2=(-1 0 1 2 3 4 5 6) + punpckhbw xmm5, xmm0 ; xmm5=( 7 8 9 10 11 12 13 14) + movdqa xmm6, xmm3 + punpcklbw xmm3, xmm0 ; xmm3=( 1 2 3 4 5 6 7 8) + punpckhbw xmm6, xmm0 ; xmm6=( 9 10 11 12 13 14 15 16) + + pmullw xmm1, [rel PW_THREE] + pmullw xmm4, [rel PW_THREE] + paddw xmm2, [rel PW_ONE] + paddw xmm5, [rel PW_ONE] + paddw xmm3, [rel PW_TWO] + paddw xmm6, [rel PW_TWO] + + paddw xmm2, xmm1 + paddw xmm5, xmm4 + psrlw xmm2, 2 ; xmm2=OutLE=( 0 2 4 6 8 10 12 14) + psrlw xmm5, 2 ; xmm5=OutHE=(16 18 20 22 24 26 28 30) + paddw xmm3, xmm1 + paddw xmm6, xmm4 + psrlw xmm3, 2 ; xmm3=OutLO=( 1 3 5 7 9 11 13 15) + psrlw xmm6, 2 ; xmm6=OutHO=(17 19 21 23 25 27 29 31) + + psllw xmm3, BYTE_BIT + psllw xmm6, BYTE_BIT + por xmm2, xmm3 ; xmm2=OutL=( 0 1 2 ... 13 14 15) + por xmm5, xmm6 ; xmm5=OutH=(16 17 18 ... 29 30 31) + + movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm5 + + sub rax, byte SIZEOF_XMMWORD + add rsi, byte 1*SIZEOF_XMMWORD ; inptr + add rdi, byte 2*SIZEOF_XMMWORD ; outptr + cmp rax, byte SIZEOF_XMMWORD + ja near .columnloop + test eax, eax + jnz near .columnloop_last + + pop rsi + pop rdi + pop rax + + add rsi, byte SIZEOF_JSAMPROW ; input_data + add rdi, byte SIZEOF_JSAMPROW ; output_data + dec rcx ; rowctr + jg near .rowloop + +.return: + UNCOLLECT_ARGS 4 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Fancy processing for the common case of 2:1 horizontal and 2:1 vertical. +; Again a triangle filter; see comments for h2v1 case, above. +; +; GLOBAL(void) +; jsimd_h2v2_fancy_upsample_sse2(int max_v_samp_factor, +; JDIMENSION downsampled_width, +; JSAMPARRAY input_data, +; JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION downsampled_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 4 + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_fancy_upsample_sse2) + +EXTN(jsimd_h2v2_fancy_upsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 4 + push rbx + + mov eax, r11d ; colctr + test rax, rax + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz near .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data +.rowloop: + push rax ; colctr + push rcx + push rdi + push rsi + + mov rcxp, JSAMPROW [rsi-1*SIZEOF_JSAMPROW] ; inptr1(above) + mov rbxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; inptr0 + mov rsip, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; inptr1(below) + mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0 + mov rdip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1 + + test rax, SIZEOF_XMMWORD-1 + jz short .skip + push rdx + mov dl, JSAMPLE [rcx+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rcx+rax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [rbx+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rbx+rax*SIZEOF_JSAMPLE], dl + mov dl, JSAMPLE [rsi+(rax-1)*SIZEOF_JSAMPLE] + mov JSAMPLE [rsi+rax*SIZEOF_JSAMPLE], dl ; insert a dummy sample + pop rdx +.skip: + ; -- process the first column block + + movdqa xmm0, XMMWORD [rbx+0*SIZEOF_XMMWORD] ; xmm0=row[ 0][0] + movdqa xmm1, XMMWORD [rcx+0*SIZEOF_XMMWORD] ; xmm1=row[-1][0] + movdqa xmm2, XMMWORD [rsi+0*SIZEOF_XMMWORD] ; xmm2=row[+1][0] + + pxor xmm3, xmm3 ; xmm3=(all 0's) + movdqa xmm4, xmm0 + punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7) + punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15) + movdqa xmm5, xmm1 + punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15) + movdqa xmm6, xmm2 + punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15) + + pmullw xmm0, [rel PW_THREE] + pmullw xmm4, [rel PW_THREE] + + pcmpeqb xmm7, xmm7 + psrldq xmm7, (SIZEOF_XMMWORD-2) + + paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7) + paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15) + paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7) + paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15) + + movdqa XMMWORD [rdx+0*SIZEOF_XMMWORD], xmm1 ; temporarily save + movdqa XMMWORD [rdx+1*SIZEOF_XMMWORD], xmm5 ; the intermediate data + movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm6 + + pand xmm1, xmm7 ; xmm1=( 0 -- -- -- -- -- -- --) + pand xmm2, xmm7 ; xmm2=( 0 -- -- -- -- -- -- --) + + movdqa XMMWORD [wk(0)], xmm1 + movdqa XMMWORD [wk(1)], xmm2 + + add rax, byte SIZEOF_XMMWORD-1 + and rax, byte -SIZEOF_XMMWORD + cmp rax, byte SIZEOF_XMMWORD + ja short .columnloop + +.columnloop_last: + ; -- process the last column block + + pcmpeqb xmm1, xmm1 + pslldq xmm1, (SIZEOF_XMMWORD-2) + movdqa xmm2, xmm1 + + pand xmm1, XMMWORD [rdx+1*SIZEOF_XMMWORD] + pand xmm2, XMMWORD [rdi+1*SIZEOF_XMMWORD] + + movdqa XMMWORD [wk(2)], xmm1 ; xmm1=(-- -- -- -- -- -- -- 15) + movdqa XMMWORD [wk(3)], xmm2 ; xmm2=(-- -- -- -- -- -- -- 15) + + jmp near .upsample + +.columnloop: + ; -- process the next column block + + movdqa xmm0, XMMWORD [rbx+1*SIZEOF_XMMWORD] ; xmm0=row[ 0][1] + movdqa xmm1, XMMWORD [rcx+1*SIZEOF_XMMWORD] ; xmm1=row[-1][1] + movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD] ; xmm2=row[+1][1] + + pxor xmm3, xmm3 ; xmm3=(all 0's) + movdqa xmm4, xmm0 + punpcklbw xmm0, xmm3 ; xmm0=row[ 0]( 0 1 2 3 4 5 6 7) + punpckhbw xmm4, xmm3 ; xmm4=row[ 0]( 8 9 10 11 12 13 14 15) + movdqa xmm5, xmm1 + punpcklbw xmm1, xmm3 ; xmm1=row[-1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm5, xmm3 ; xmm5=row[-1]( 8 9 10 11 12 13 14 15) + movdqa xmm6, xmm2 + punpcklbw xmm2, xmm3 ; xmm2=row[+1]( 0 1 2 3 4 5 6 7) + punpckhbw xmm6, xmm3 ; xmm6=row[+1]( 8 9 10 11 12 13 14 15) + + pmullw xmm0, [rel PW_THREE] + pmullw xmm4, [rel PW_THREE] + + paddw xmm1, xmm0 ; xmm1=Int0L=( 0 1 2 3 4 5 6 7) + paddw xmm5, xmm4 ; xmm5=Int0H=( 8 9 10 11 12 13 14 15) + paddw xmm2, xmm0 ; xmm2=Int1L=( 0 1 2 3 4 5 6 7) + paddw xmm6, xmm4 ; xmm6=Int1H=( 8 9 10 11 12 13 14 15) + + movdqa XMMWORD [rdx+2*SIZEOF_XMMWORD], xmm1 ; temporarily save + movdqa XMMWORD [rdx+3*SIZEOF_XMMWORD], xmm5 ; the intermediate data + movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm6 + + pslldq xmm1, (SIZEOF_XMMWORD-2) ; xmm1=(-- -- -- -- -- -- -- 0) + pslldq xmm2, (SIZEOF_XMMWORD-2) ; xmm2=(-- -- -- -- -- -- -- 0) + + movdqa XMMWORD [wk(2)], xmm1 + movdqa XMMWORD [wk(3)], xmm2 + +.upsample: + ; -- process the upper row + + movdqa xmm7, XMMWORD [rdx+0*SIZEOF_XMMWORD] + movdqa xmm3, XMMWORD [rdx+1*SIZEOF_XMMWORD] + + movdqa xmm0, xmm7 ; xmm7=Int0L=( 0 1 2 3 4 5 6 7) + movdqa xmm4, xmm3 ; xmm3=Int0H=( 8 9 10 11 12 13 14 15) + psrldq xmm0, 2 ; xmm0=( 1 2 3 4 5 6 7 --) + pslldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(-- -- -- -- -- -- -- 8) + movdqa xmm5, xmm7 + movdqa xmm6, xmm3 + psrldq xmm5, (SIZEOF_XMMWORD-2) ; xmm5=( 7 -- -- -- -- -- -- --) + pslldq xmm6, 2 ; xmm6=(-- 8 9 10 11 12 13 14) + + por xmm0, xmm4 ; xmm0=( 1 2 3 4 5 6 7 8) + por xmm5, xmm6 ; xmm5=( 7 8 9 10 11 12 13 14) + + movdqa xmm1, xmm7 + movdqa xmm2, xmm3 + pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6) + psrldq xmm2, 2 ; xmm2=( 9 10 11 12 13 14 15 --) + movdqa xmm4, xmm3 + psrldq xmm4, (SIZEOF_XMMWORD-2) ; xmm4=(15 -- -- -- -- -- -- --) + + por xmm1, XMMWORD [wk(0)] ; xmm1=(-1 0 1 2 3 4 5 6) + por xmm2, XMMWORD [wk(2)] ; xmm2=( 9 10 11 12 13 14 15 16) + + movdqa XMMWORD [wk(0)], xmm4 + + pmullw xmm7, [rel PW_THREE] + pmullw xmm3, [rel PW_THREE] + paddw xmm1, [rel PW_EIGHT] + paddw xmm5, [rel PW_EIGHT] + paddw xmm0, [rel PW_SEVEN] + paddw xmm2, [rel PW_SEVEN] + + paddw xmm1, xmm7 + paddw xmm5, xmm3 + psrlw xmm1, 4 ; xmm1=Out0LE=( 0 2 4 6 8 10 12 14) + psrlw xmm5, 4 ; xmm5=Out0HE=(16 18 20 22 24 26 28 30) + paddw xmm0, xmm7 + paddw xmm2, xmm3 + psrlw xmm0, 4 ; xmm0=Out0LO=( 1 3 5 7 9 11 13 15) + psrlw xmm2, 4 ; xmm2=Out0HO=(17 19 21 23 25 27 29 31) + + psllw xmm0, BYTE_BIT + psllw xmm2, BYTE_BIT + por xmm1, xmm0 ; xmm1=Out0L=( 0 1 2 ... 13 14 15) + por xmm5, xmm2 ; xmm5=Out0H=(16 17 18 ... 29 30 31) + + movdqa XMMWORD [rdx+0*SIZEOF_XMMWORD], xmm1 + movdqa XMMWORD [rdx+1*SIZEOF_XMMWORD], xmm5 + + ; -- process the lower row + + movdqa xmm6, XMMWORD [rdi+0*SIZEOF_XMMWORD] + movdqa xmm4, XMMWORD [rdi+1*SIZEOF_XMMWORD] + + movdqa xmm7, xmm6 ; xmm6=Int1L=( 0 1 2 3 4 5 6 7) + movdqa xmm3, xmm4 ; xmm4=Int1H=( 8 9 10 11 12 13 14 15) + psrldq xmm7, 2 ; xmm7=( 1 2 3 4 5 6 7 --) + pslldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(-- -- -- -- -- -- -- 8) + movdqa xmm0, xmm6 + movdqa xmm2, xmm4 + psrldq xmm0, (SIZEOF_XMMWORD-2) ; xmm0=( 7 -- -- -- -- -- -- --) + pslldq xmm2, 2 ; xmm2=(-- 8 9 10 11 12 13 14) + + por xmm7, xmm3 ; xmm7=( 1 2 3 4 5 6 7 8) + por xmm0, xmm2 ; xmm0=( 7 8 9 10 11 12 13 14) + + movdqa xmm1, xmm6 + movdqa xmm5, xmm4 + pslldq xmm1, 2 ; xmm1=(-- 0 1 2 3 4 5 6) + psrldq xmm5, 2 ; xmm5=( 9 10 11 12 13 14 15 --) + movdqa xmm3, xmm4 + psrldq xmm3, (SIZEOF_XMMWORD-2) ; xmm3=(15 -- -- -- -- -- -- --) + + por xmm1, XMMWORD [wk(1)] ; xmm1=(-1 0 1 2 3 4 5 6) + por xmm5, XMMWORD [wk(3)] ; xmm5=( 9 10 11 12 13 14 15 16) + + movdqa XMMWORD [wk(1)], xmm3 + + pmullw xmm6, [rel PW_THREE] + pmullw xmm4, [rel PW_THREE] + paddw xmm1, [rel PW_EIGHT] + paddw xmm0, [rel PW_EIGHT] + paddw xmm7, [rel PW_SEVEN] + paddw xmm5, [rel PW_SEVEN] + + paddw xmm1, xmm6 + paddw xmm0, xmm4 + psrlw xmm1, 4 ; xmm1=Out1LE=( 0 2 4 6 8 10 12 14) + psrlw xmm0, 4 ; xmm0=Out1HE=(16 18 20 22 24 26 28 30) + paddw xmm7, xmm6 + paddw xmm5, xmm4 + psrlw xmm7, 4 ; xmm7=Out1LO=( 1 3 5 7 9 11 13 15) + psrlw xmm5, 4 ; xmm5=Out1HO=(17 19 21 23 25 27 29 31) + + psllw xmm7, BYTE_BIT + psllw xmm5, BYTE_BIT + por xmm1, xmm7 ; xmm1=Out1L=( 0 1 2 ... 13 14 15) + por xmm0, xmm5 ; xmm0=Out1H=(16 17 18 ... 29 30 31) + + movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm1 + movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm0 + + sub rax, byte SIZEOF_XMMWORD + add rcx, byte 1*SIZEOF_XMMWORD ; inptr1(above) + add rbx, byte 1*SIZEOF_XMMWORD ; inptr0 + add rsi, byte 1*SIZEOF_XMMWORD ; inptr1(below) + add rdx, byte 2*SIZEOF_XMMWORD ; outptr0 + add rdi, byte 2*SIZEOF_XMMWORD ; outptr1 + cmp rax, byte SIZEOF_XMMWORD + ja near .columnloop + test rax, rax + jnz near .columnloop_last + + pop rsi + pop rdi + pop rcx + pop rax + + add rsi, byte 1*SIZEOF_JSAMPROW ; input_data + add rdi, byte 2*SIZEOF_JSAMPROW ; output_data + sub rcx, byte 2 ; rowctr + jg near .rowloop + +.return: + pop rbx + UNCOLLECT_ARGS 4 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 1:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v1_upsample_sse2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION output_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v1_upsample_sse2) + +EXTN(jsimd_h2v1_upsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + + mov edx, r11d + add rdx, byte (2*SIZEOF_XMMWORD)-1 + and rdx, byte -(2*SIZEOF_XMMWORD) + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz short .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data +.rowloop: + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rdip, JSAMPROW [rdi] ; outptr + mov rax, rdx ; colctr +.columnloop: + + movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD] + + movdqa xmm1, xmm0 + punpcklbw xmm0, xmm0 + punpckhbw xmm1, xmm1 + + movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0 + movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1 + + sub rax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD] + + movdqa xmm3, xmm2 + punpcklbw xmm2, xmm2 + punpckhbw xmm3, xmm3 + + movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm3 + + sub rax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + add rsi, byte 2*SIZEOF_XMMWORD ; inptr + add rdi, byte 4*SIZEOF_XMMWORD ; outptr + jmp short .columnloop + +.nextrow: + pop rsi + pop rdi + + add rsi, byte SIZEOF_JSAMPROW ; input_data + add rdi, byte SIZEOF_JSAMPROW ; output_data + dec rcx ; rowctr + jg short .rowloop + +.return: + UNCOLLECT_ARGS 4 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Fast processing for the common case of 2:1 horizontal and 2:1 vertical. +; It's still a box filter. +; +; GLOBAL(void) +; jsimd_h2v2_upsample_sse2(int max_v_samp_factor, JDIMENSION output_width, +; JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr); +; + +; r10 = int max_v_samp_factor +; r11d = JDIMENSION output_width +; r12 = JSAMPARRAY input_data +; r13 = JSAMPARRAY *output_data_ptr + + align 32 + GLOBAL_FUNCTION(jsimd_h2v2_upsample_sse2) + +EXTN(jsimd_h2v2_upsample_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + push rbx + + mov edx, r11d + add rdx, byte (2*SIZEOF_XMMWORD)-1 + and rdx, byte -(2*SIZEOF_XMMWORD) + jz near .return + + mov rcx, r10 ; rowctr + test rcx, rcx + jz near .return + + mov rsi, r12 ; input_data + mov rdi, r13 + mov rdip, JSAMPARRAY [rdi] ; output_data +.rowloop: + push rdi + push rsi + + mov rsip, JSAMPROW [rsi] ; inptr + mov rbxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] ; outptr0 + mov rdip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] ; outptr1 + mov rax, rdx ; colctr +.columnloop: + + movdqa xmm0, XMMWORD [rsi+0*SIZEOF_XMMWORD] + + movdqa xmm1, xmm0 + punpcklbw xmm0, xmm0 + punpckhbw xmm1, xmm1 + + movdqa XMMWORD [rbx+0*SIZEOF_XMMWORD], xmm0 + movdqa XMMWORD [rbx+1*SIZEOF_XMMWORD], xmm1 + movdqa XMMWORD [rdi+0*SIZEOF_XMMWORD], xmm0 + movdqa XMMWORD [rdi+1*SIZEOF_XMMWORD], xmm1 + + sub rax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + movdqa xmm2, XMMWORD [rsi+1*SIZEOF_XMMWORD] + + movdqa xmm3, xmm2 + punpcklbw xmm2, xmm2 + punpckhbw xmm3, xmm3 + + movdqa XMMWORD [rbx+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [rbx+3*SIZEOF_XMMWORD], xmm3 + movdqa XMMWORD [rdi+2*SIZEOF_XMMWORD], xmm2 + movdqa XMMWORD [rdi+3*SIZEOF_XMMWORD], xmm3 + + sub rax, byte 2*SIZEOF_XMMWORD + jz short .nextrow + + add rsi, byte 2*SIZEOF_XMMWORD ; inptr + add rbx, byte 4*SIZEOF_XMMWORD ; outptr0 + add rdi, byte 4*SIZEOF_XMMWORD ; outptr1 + jmp short .columnloop + +.nextrow: + pop rsi + pop rdi + + add rsi, byte 1*SIZEOF_JSAMPROW ; input_data + add rdi, byte 2*SIZEOF_JSAMPROW ; output_data + sub rcx, byte 2 ; rowctr + jg near .rowloop + +.return: + pop rbx + UNCOLLECT_ARGS 4 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctflt-sse.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctflt-sse.asm new file mode 100644 index 0000000..58a1f55 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctflt-sse.asm @@ -0,0 +1,353 @@ +; +; jfdctflt.asm - floating-point FDCT (64-bit SSE) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a floating-point implementation of the forward DCT +; (Discrete Cosine Transform). The following code is based directly on +; the IJG's original jfdctflt.c; see the jfdctflt.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%macro unpcklps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5) + shufps %1, %2, 0x44 +%endmacro + +%macro unpckhps2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7) + shufps %1, %2, 0xEE +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_float_sse) + +EXTN(jconst_fdct_float_sse): + +PD_0_382 times 4 dd 0.382683432365089771728460 +PD_0_707 times 4 dd 0.707106781186547524400844 +PD_0_541 times 4 dd 0.541196100146196984399723 +PD_1_306 times 4 dd 1.306562964876376527856643 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_float_sse(FAST_FLOAT *data) +; + +; r10 = FAST_FLOAT *data + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_float_sse) + +EXTN(jsimd_fdct_float_sse): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 1 + + ; ---- Pass 1: process rows. + + mov rdx, r10 ; (FAST_FLOAT *) + mov rcx, DCTSIZE/4 +.rowloop: + + movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)] + + ; xmm0=(20 21 22 23), xmm2=(24 25 26 27) + ; xmm1=(30 31 32 33), xmm3=(34 35 36 37) + + movaps xmm4, xmm0 ; transpose coefficients(phase 1) + unpcklps xmm0, xmm1 ; xmm0=(20 30 21 31) + unpckhps xmm4, xmm1 ; xmm4=(22 32 23 33) + movaps xmm5, xmm2 ; transpose coefficients(phase 1) + unpcklps xmm2, xmm3 ; xmm2=(24 34 25 35) + unpckhps xmm5, xmm3 ; xmm5=(26 36 27 37) + + movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)] + + ; xmm6=(00 01 02 03), xmm1=(04 05 06 07) + ; xmm7=(10 11 12 13), xmm3=(14 15 16 17) + + movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 32 23 33) + movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(24 34 25 35) + + movaps xmm4, xmm6 ; transpose coefficients(phase 1) + unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11) + unpckhps xmm4, xmm7 ; xmm4=(02 12 03 13) + movaps xmm2, xmm1 ; transpose coefficients(phase 1) + unpcklps xmm1, xmm3 ; xmm1=(04 14 05 15) + unpckhps xmm2, xmm3 ; xmm2=(06 16 07 17) + + movaps xmm7, xmm6 ; transpose coefficients(phase 2) + unpcklps2 xmm6, xmm0 ; xmm6=(00 10 20 30)=data0 + unpckhps2 xmm7, xmm0 ; xmm7=(01 11 21 31)=data1 + movaps xmm3, xmm2 ; transpose coefficients(phase 2) + unpcklps2 xmm2, xmm5 ; xmm2=(06 16 26 36)=data6 + unpckhps2 xmm3, xmm5 ; xmm3=(07 17 27 37)=data7 + + movaps xmm0, xmm7 + movaps xmm5, xmm6 + subps xmm7, xmm2 ; xmm7=data1-data6=tmp6 + subps xmm6, xmm3 ; xmm6=data0-data7=tmp7 + addps xmm0, xmm2 ; xmm0=data1+data6=tmp1 + addps xmm5, xmm3 ; xmm5=data0+data7=tmp0 + + movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 32 23 33) + movaps xmm3, XMMWORD [wk(1)] ; xmm3=(24 34 25 35) + movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6 + movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 + + movaps xmm7, xmm4 ; transpose coefficients(phase 2) + unpcklps2 xmm4, xmm2 ; xmm4=(02 12 22 32)=data2 + unpckhps2 xmm7, xmm2 ; xmm7=(03 13 23 33)=data3 + movaps xmm6, xmm1 ; transpose coefficients(phase 2) + unpcklps2 xmm1, xmm3 ; xmm1=(04 14 24 34)=data4 + unpckhps2 xmm6, xmm3 ; xmm6=(05 15 25 35)=data5 + + movaps xmm2, xmm7 + movaps xmm3, xmm4 + addps xmm7, xmm1 ; xmm7=data3+data4=tmp3 + addps xmm4, xmm6 ; xmm4=data2+data5=tmp2 + subps xmm2, xmm1 ; xmm2=data3-data4=tmp4 + subps xmm3, xmm6 ; xmm3=data2-data5=tmp5 + + ; -- Even part + + movaps xmm1, xmm5 + movaps xmm6, xmm0 + subps xmm5, xmm7 ; xmm5=tmp13 + subps xmm0, xmm4 ; xmm0=tmp12 + addps xmm1, xmm7 ; xmm1=tmp10 + addps xmm6, xmm4 ; xmm6=tmp11 + + addps xmm0, xmm5 + mulps xmm0, [rel PD_0_707] ; xmm0=z1 + + movaps xmm7, xmm1 + movaps xmm4, xmm5 + subps xmm1, xmm6 ; xmm1=data4 + subps xmm5, xmm0 ; xmm5=data6 + addps xmm7, xmm6 ; xmm7=data0 + addps xmm4, xmm0 ; xmm4=data2 + + movaps XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(2,1,rdx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4 + + ; -- Odd part + + movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6 + movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7 + + addps xmm2, xmm3 ; xmm2=tmp10 + addps xmm3, xmm6 ; xmm3=tmp11 + addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7 + + mulps xmm3, [rel PD_0_707] ; xmm3=z3 + + movaps xmm1, xmm2 ; xmm1=tmp10 + subps xmm2, xmm6 + mulps xmm2, [rel PD_0_382] ; xmm2=z5 + mulps xmm1, [rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196) + mulps xmm6, [rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562) + addps xmm1, xmm2 ; xmm1=z2 + addps xmm6, xmm2 ; xmm6=z4 + + movaps xmm5, xmm0 + subps xmm0, xmm3 ; xmm0=z13 + addps xmm5, xmm3 ; xmm5=z11 + + movaps xmm7, xmm0 + movaps xmm4, xmm5 + subps xmm0, xmm1 ; xmm0=data3 + subps xmm5, xmm6 ; xmm5=data7 + addps xmm7, xmm1 ; xmm7=data5 + addps xmm4, xmm6 ; xmm4=data1 + + movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(3,1,rdx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4 + + add rdx, 4*DCTSIZE*SIZEOF_FAST_FLOAT + dec rcx + jnz near .rowloop + + ; ---- Pass 2: process columns. + + mov rdx, r10 ; (FAST_FLOAT *) + mov rcx, DCTSIZE/4 +.columnloop: + + movaps xmm0, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)] + + ; xmm0=(02 12 22 32), xmm2=(42 52 62 72) + ; xmm1=(03 13 23 33), xmm3=(43 53 63 73) + + movaps xmm4, xmm0 ; transpose coefficients(phase 1) + unpcklps xmm0, xmm1 ; xmm0=(02 03 12 13) + unpckhps xmm4, xmm1 ; xmm4=(22 23 32 33) + movaps xmm5, xmm2 ; transpose coefficients(phase 1) + unpcklps xmm2, xmm3 ; xmm2=(42 43 52 53) + unpckhps xmm5, xmm3 ; xmm5=(62 63 72 73) + + movaps xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm7, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)] + + ; xmm6=(00 10 20 30), xmm1=(40 50 60 70) + ; xmm7=(01 11 21 31), xmm3=(41 51 61 71) + + movaps XMMWORD [wk(0)], xmm4 ; wk(0)=(22 23 32 33) + movaps XMMWORD [wk(1)], xmm2 ; wk(1)=(42 43 52 53) + + movaps xmm4, xmm6 ; transpose coefficients(phase 1) + unpcklps xmm6, xmm7 ; xmm6=(00 01 10 11) + unpckhps xmm4, xmm7 ; xmm4=(20 21 30 31) + movaps xmm2, xmm1 ; transpose coefficients(phase 1) + unpcklps xmm1, xmm3 ; xmm1=(40 41 50 51) + unpckhps xmm2, xmm3 ; xmm2=(60 61 70 71) + + movaps xmm7, xmm6 ; transpose coefficients(phase 2) + unpcklps2 xmm6, xmm0 ; xmm6=(00 01 02 03)=data0 + unpckhps2 xmm7, xmm0 ; xmm7=(10 11 12 13)=data1 + movaps xmm3, xmm2 ; transpose coefficients(phase 2) + unpcklps2 xmm2, xmm5 ; xmm2=(60 61 62 63)=data6 + unpckhps2 xmm3, xmm5 ; xmm3=(70 71 72 73)=data7 + + movaps xmm0, xmm7 + movaps xmm5, xmm6 + subps xmm7, xmm2 ; xmm7=data1-data6=tmp6 + subps xmm6, xmm3 ; xmm6=data0-data7=tmp7 + addps xmm0, xmm2 ; xmm0=data1+data6=tmp1 + addps xmm5, xmm3 ; xmm5=data0+data7=tmp0 + + movaps xmm2, XMMWORD [wk(0)] ; xmm2=(22 23 32 33) + movaps xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53) + movaps XMMWORD [wk(0)], xmm7 ; wk(0)=tmp6 + movaps XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 + + movaps xmm7, xmm4 ; transpose coefficients(phase 2) + unpcklps2 xmm4, xmm2 ; xmm4=(20 21 22 23)=data2 + unpckhps2 xmm7, xmm2 ; xmm7=(30 31 32 33)=data3 + movaps xmm6, xmm1 ; transpose coefficients(phase 2) + unpcklps2 xmm1, xmm3 ; xmm1=(40 41 42 43)=data4 + unpckhps2 xmm6, xmm3 ; xmm6=(50 51 52 53)=data5 + + movaps xmm2, xmm7 + movaps xmm3, xmm4 + addps xmm7, xmm1 ; xmm7=data3+data4=tmp3 + addps xmm4, xmm6 ; xmm4=data2+data5=tmp2 + subps xmm2, xmm1 ; xmm2=data3-data4=tmp4 + subps xmm3, xmm6 ; xmm3=data2-data5=tmp5 + + ; -- Even part + + movaps xmm1, xmm5 + movaps xmm6, xmm0 + subps xmm5, xmm7 ; xmm5=tmp13 + subps xmm0, xmm4 ; xmm0=tmp12 + addps xmm1, xmm7 ; xmm1=tmp10 + addps xmm6, xmm4 ; xmm6=tmp11 + + addps xmm0, xmm5 + mulps xmm0, [rel PD_0_707] ; xmm0=z1 + + movaps xmm7, xmm1 + movaps xmm4, xmm5 + subps xmm1, xmm6 ; xmm1=data4 + subps xmm5, xmm0 ; xmm5=data6 + addps xmm7, xmm6 ; xmm7=data0 + addps xmm4, xmm0 ; xmm4=data2 + + movaps XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FAST_FLOAT)], xmm4 + + ; -- Odd part + + movaps xmm6, XMMWORD [wk(0)] ; xmm6=tmp6 + movaps xmm0, XMMWORD [wk(1)] ; xmm0=tmp7 + + addps xmm2, xmm3 ; xmm2=tmp10 + addps xmm3, xmm6 ; xmm3=tmp11 + addps xmm6, xmm0 ; xmm6=tmp12, xmm0=tmp7 + + mulps xmm3, [rel PD_0_707] ; xmm3=z3 + + movaps xmm1, xmm2 ; xmm1=tmp10 + subps xmm2, xmm6 + mulps xmm2, [rel PD_0_382] ; xmm2=z5 + mulps xmm1, [rel PD_0_541] ; xmm1=MULTIPLY(tmp10,FIX_0_541196) + mulps xmm6, [rel PD_1_306] ; xmm6=MULTIPLY(tmp12,FIX_1_306562) + addps xmm1, xmm2 ; xmm1=z2 + addps xmm6, xmm2 ; xmm6=z4 + + movaps xmm5, xmm0 + subps xmm0, xmm3 ; xmm0=z13 + addps xmm5, xmm3 ; xmm5=z11 + + movaps xmm7, xmm0 + movaps xmm4, xmm5 + subps xmm0, xmm1 ; xmm0=data3 + subps xmm5, xmm6 ; xmm5=data7 + addps xmm7, xmm1 ; xmm7=data5 + addps xmm4, xmm6 ; xmm4=data1 + + movaps XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FAST_FLOAT)], xmm7 + movaps XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)], xmm4 + + add rdx, byte 4*SIZEOF_FAST_FLOAT + dec rcx + jnz near .columnloop + + UNCOLLECT_ARGS 1 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctfst-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctfst-sse2.asm new file mode 100644 index 0000000..3b92d4e --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctfst-sse2.asm @@ -0,0 +1,387 @@ +; +; jfdctfst.asm - fast integer FDCT (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a fast, not so accurate integer implementation of +; the forward DCT (Discrete Cosine Transform). The following code is +; based directly on the IJG's original jfdctfst.c; see the jfdctfst.c +; for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 8 ; 14 is also OK. + +%if CONST_BITS == 8 +F_0_382 equ 98 ; FIX(0.382683433) +F_0_541 equ 139 ; FIX(0.541196100) +F_0_707 equ 181 ; FIX(0.707106781) +F_1_306 equ 334 ; FIX(1.306562965) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_382 equ DESCALE( 410903207, 30 - CONST_BITS) ; FIX(0.382683433) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_707 equ DESCALE( 759250124, 30 - CONST_BITS) ; FIX(0.707106781) +F_1_306 equ DESCALE(1402911301, 30 - CONST_BITS) ; FIX(1.306562965) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + +; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow) +; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw) + +%define PRE_MULTIPLY_SCALE_BITS 2 +%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS) + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_ifast_sse2) + +EXTN(jconst_fdct_ifast_sse2): + +PW_F0707 times 8 dw F_0_707 << CONST_SHIFT +PW_F0382 times 8 dw F_0_382 << CONST_SHIFT +PW_F0541 times 8 dw F_0_541 << CONST_SHIFT +PW_F1306 times 8 dw F_1_306 << CONST_SHIFT + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_ifast_sse2(DCTELEM *data) +; + +; r10 = DCTELEM *data + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_ifast_sse2) + +EXTN(jsimd_fdct_ifast_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 1 + + ; ---- Pass 1: process rows. + + mov rdx, r10 ; (DCTELEM *) + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm2, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)] + + ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27) + ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37) + + movdqa xmm4, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13) + punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17) + movdqa xmm5, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33) + punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37) + + movdqa xmm6, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm7, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)] + + ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62) + ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63) + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33) + movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37) + + movdqa xmm2, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53) + punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57) + movdqa xmm5, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73) + punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77) + + movdqa xmm7, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71) + punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73) + movdqa xmm3, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75) + punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77) + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33) + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37) + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(42 52 62 72 43 53 63 73) + movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=(44 54 64 74 45 55 65 75) + + movdqa xmm7, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31) + punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33) + movdqa xmm2, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35) + punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37) + + movdqa xmm1, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0 + punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1 + movdqa xmm5, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6 + punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7 + + movdqa xmm6, xmm1 + movdqa xmm3, xmm0 + psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6 + psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7 + paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1 + paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0 + + movdqa xmm2, XMMWORD [wk(0)] ; xmm2=(42 52 62 72 43 53 63 73) + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(44 54 64 74 45 55 65 75) + movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7 + + movdqa xmm1, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2 + punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3 + movdqa xmm0, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4 + punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5 + + movdqa xmm2, xmm1 + movdqa xmm5, xmm7 + paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3 + paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2 + psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4 + psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm4, xmm3 + movdqa xmm0, xmm6 + psubw xmm3, xmm1 ; xmm3=tmp13 + psubw xmm6, xmm7 ; xmm6=tmp12 + paddw xmm4, xmm1 ; xmm4=tmp10 + paddw xmm0, xmm7 ; xmm0=tmp11 + + paddw xmm6, xmm3 + psllw xmm6, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm6, [rel PW_F0707] ; xmm6=z1 + + movdqa xmm1, xmm4 + movdqa xmm7, xmm3 + psubw xmm4, xmm0 ; xmm4=data4 + psubw xmm3, xmm6 ; xmm3=data6 + paddw xmm1, xmm0 ; xmm1=data0 + paddw xmm7, xmm6 ; xmm7=data2 + + movdqa xmm0, XMMWORD [wk(0)] ; xmm0=tmp6 + movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp7 + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=data4 + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=data6 + + ; -- Odd part + + paddw xmm2, xmm5 ; xmm2=tmp10 + paddw xmm5, xmm0 ; xmm5=tmp11 + paddw xmm0, xmm6 ; xmm0=tmp12, xmm6=tmp7 + + psllw xmm2, PRE_MULTIPLY_SCALE_BITS + psllw xmm0, PRE_MULTIPLY_SCALE_BITS + + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm5, [rel PW_F0707] ; xmm5=z3 + + movdqa xmm4, xmm2 ; xmm4=tmp10 + psubw xmm2, xmm0 + pmulhw xmm2, [rel PW_F0382] ; xmm2=z5 + pmulhw xmm4, [rel PW_F0541] ; xmm4=MULTIPLY(tmp10,FIX_0_541196) + pmulhw xmm0, [rel PW_F1306] ; xmm0=MULTIPLY(tmp12,FIX_1_306562) + paddw xmm4, xmm2 ; xmm4=z2 + paddw xmm0, xmm2 ; xmm0=z4 + + movdqa xmm3, xmm6 + psubw xmm6, xmm5 ; xmm6=z13 + paddw xmm3, xmm5 ; xmm3=z11 + + movdqa xmm2, xmm6 + movdqa xmm5, xmm3 + psubw xmm6, xmm4 ; xmm6=data3 + psubw xmm3, xmm0 ; xmm3=data7 + paddw xmm2, xmm4 ; xmm2=data5 + paddw xmm5, xmm0 ; xmm5=data1 + + ; ---- Pass 2: process columns. + + ; xmm1=(00 10 20 30 40 50 60 70), xmm7=(02 12 22 32 42 52 62 72) + ; xmm5=(01 11 21 31 41 51 61 71), xmm6=(03 13 23 33 43 53 63 73) + + movdqa xmm4, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm5 ; xmm1=(00 01 10 11 20 21 30 31) + punpckhwd xmm4, xmm5 ; xmm4=(40 41 50 51 60 61 70 71) + movdqa xmm0, xmm7 ; transpose coefficients(phase 1) + punpcklwd xmm7, xmm6 ; xmm7=(02 03 12 13 22 23 32 33) + punpckhwd xmm0, xmm6 ; xmm0=(42 43 52 53 62 63 72 73) + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=col4 + movdqa xmm6, XMMWORD [wk(1)] ; xmm6=col6 + + ; xmm5=(04 14 24 34 44 54 64 74), xmm6=(06 16 26 36 46 56 66 76) + ; xmm2=(05 15 25 35 45 55 65 75), xmm3=(07 17 27 37 47 57 67 77) + + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=(02 03 12 13 22 23 32 33) + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(42 43 52 53 62 63 72 73) + + movdqa xmm7, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm2 ; xmm5=(04 05 14 15 24 25 34 35) + punpckhwd xmm7, xmm2 ; xmm7=(44 45 54 55 64 65 74 75) + movdqa xmm0, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm3 ; xmm6=(06 07 16 17 26 27 36 37) + punpckhwd xmm0, xmm3 ; xmm0=(46 47 56 57 66 67 76 77) + + movdqa xmm2, xmm5 ; transpose coefficients(phase 2) + punpckldq xmm5, xmm6 ; xmm5=(04 05 06 07 14 15 16 17) + punpckhdq xmm2, xmm6 ; xmm2=(24 25 26 27 34 35 36 37) + movdqa xmm3, xmm7 ; transpose coefficients(phase 2) + punpckldq xmm7, xmm0 ; xmm7=(44 45 46 47 54 55 56 57) + punpckhdq xmm3, xmm0 ; xmm3=(64 65 66 67 74 75 76 77) + + movdqa xmm6, XMMWORD [wk(0)] ; xmm6=(02 03 12 13 22 23 32 33) + movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(42 43 52 53 62 63 72 73) + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(24 25 26 27 34 35 36 37) + movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=(44 45 46 47 54 55 56 57) + + movdqa xmm2, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 10 11 12 13) + punpckhdq xmm2, xmm6 ; xmm2=(20 21 22 23 30 31 32 33) + movdqa xmm7, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm0 ; xmm4=(40 41 42 43 50 51 52 53) + punpckhdq xmm7, xmm0 ; xmm7=(60 61 62 63 70 71 72 73) + + movdqa xmm6, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm5 ; xmm1=(00 01 02 03 04 05 06 07)=data0 + punpckhqdq xmm6, xmm5 ; xmm6=(10 11 12 13 14 15 16 17)=data1 + movdqa xmm0, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm3 ; xmm7=(60 61 62 63 64 65 66 67)=data6 + punpckhqdq xmm0, xmm3 ; xmm0=(70 71 72 73 74 75 76 77)=data7 + + movdqa xmm5, xmm6 + movdqa xmm3, xmm1 + psubw xmm6, xmm7 ; xmm6=data1-data6=tmp6 + psubw xmm1, xmm0 ; xmm1=data0-data7=tmp7 + paddw xmm5, xmm7 ; xmm5=data1+data6=tmp1 + paddw xmm3, xmm0 ; xmm3=data0+data7=tmp0 + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(24 25 26 27 34 35 36 37) + movdqa xmm0, XMMWORD [wk(1)] ; xmm0=(44 45 46 47 54 55 56 57) + movdqa XMMWORD [wk(0)], xmm6 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=tmp7 + + movdqa xmm6, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm7 ; xmm2=(20 21 22 23 24 25 26 27)=data2 + punpckhqdq xmm6, xmm7 ; xmm6=(30 31 32 33 34 35 36 37)=data3 + movdqa xmm1, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm0 ; xmm4=(40 41 42 43 44 45 46 47)=data4 + punpckhqdq xmm1, xmm0 ; xmm1=(50 51 52 53 54 55 56 57)=data5 + + movdqa xmm7, xmm6 + movdqa xmm0, xmm2 + paddw xmm6, xmm4 ; xmm6=data3+data4=tmp3 + paddw xmm2, xmm1 ; xmm2=data2+data5=tmp2 + psubw xmm7, xmm4 ; xmm7=data3-data4=tmp4 + psubw xmm0, xmm1 ; xmm0=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm4, xmm3 + movdqa xmm1, xmm5 + psubw xmm3, xmm6 ; xmm3=tmp13 + psubw xmm5, xmm2 ; xmm5=tmp12 + paddw xmm4, xmm6 ; xmm4=tmp10 + paddw xmm1, xmm2 ; xmm1=tmp11 + + paddw xmm5, xmm3 + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm5, [rel PW_F0707] ; xmm5=z1 + + movdqa xmm6, xmm4 + movdqa xmm2, xmm3 + psubw xmm4, xmm1 ; xmm4=data4 + psubw xmm3, xmm5 ; xmm3=data6 + paddw xmm6, xmm1 ; xmm6=data0 + paddw xmm2, xmm5 ; xmm2=data2 + + movdqa XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)], xmm4 + movdqa XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)], xmm3 + movdqa XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)], xmm6 + movdqa XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)], xmm2 + + ; -- Odd part + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=tmp6 + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7 + + paddw xmm7, xmm0 ; xmm7=tmp10 + paddw xmm0, xmm1 ; xmm0=tmp11 + paddw xmm1, xmm5 ; xmm1=tmp12, xmm5=tmp7 + + psllw xmm7, PRE_MULTIPLY_SCALE_BITS + psllw xmm1, PRE_MULTIPLY_SCALE_BITS + + psllw xmm0, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm0, [rel PW_F0707] ; xmm0=z3 + + movdqa xmm4, xmm7 ; xmm4=tmp10 + psubw xmm7, xmm1 + pmulhw xmm7, [rel PW_F0382] ; xmm7=z5 + pmulhw xmm4, [rel PW_F0541] ; xmm4=MULTIPLY(tmp10,FIX_0_541196) + pmulhw xmm1, [rel PW_F1306] ; xmm1=MULTIPLY(tmp12,FIX_1_306562) + paddw xmm4, xmm7 ; xmm4=z2 + paddw xmm1, xmm7 ; xmm1=z4 + + movdqa xmm3, xmm5 + psubw xmm5, xmm0 ; xmm5=z13 + paddw xmm3, xmm0 ; xmm3=z11 + + movdqa xmm6, xmm5 + movdqa xmm2, xmm3 + psubw xmm5, xmm4 ; xmm5=data3 + psubw xmm3, xmm1 ; xmm3=data7 + paddw xmm6, xmm4 ; xmm6=data5 + paddw xmm2, xmm1 ; xmm2=data1 + + movdqa XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)], xmm5 + movdqa XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)], xmm3 + movdqa XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)], xmm6 + movdqa XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)], xmm2 + + UNCOLLECT_ARGS 1 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-avx2.asm new file mode 100644 index 0000000..0c45286 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-avx2.asm @@ -0,0 +1,316 @@ +; +; jfdctint.asm - accurate integer FDCT (64-bit AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2018, 2020, 2024, D. R. Commander. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; forward DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jfdctint.c; see the jfdctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit matrix transpose using AVX2 instructions +; %1-%4: Input/output registers +; %5-%8: Temp registers + +%macro DOTRANSPOSE 8 + ; %1=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47) + ; %2=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57) + ; %3=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67) + ; %4=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77) + + vpunpcklwd %5, %1, %2 + vpunpckhwd %6, %1, %2 + vpunpcklwd %7, %3, %4 + vpunpckhwd %8, %3, %4 + ; transpose coefficients(phase 1) + ; %5=(00 10 01 11 02 12 03 13 40 50 41 51 42 52 43 53) + ; %6=(04 14 05 15 06 16 07 17 44 54 45 55 46 56 47 57) + ; %7=(20 30 21 31 22 32 23 33 60 70 61 71 62 72 63 73) + ; %8=(24 34 25 35 26 36 27 37 64 74 65 75 66 76 67 77) + + vpunpckldq %1, %5, %7 + vpunpckhdq %2, %5, %7 + vpunpckldq %3, %6, %8 + vpunpckhdq %4, %6, %8 + ; transpose coefficients(phase 2) + ; %1=(00 10 20 30 01 11 21 31 40 50 60 70 41 51 61 71) + ; %2=(02 12 22 32 03 13 23 33 42 52 62 72 43 53 63 73) + ; %3=(04 14 24 34 05 15 25 35 44 54 64 74 45 55 65 75) + ; %4=(06 16 26 36 07 17 27 37 46 56 66 76 47 57 67 77) + + vpermq %1, %1, 0x8D + vpermq %2, %2, 0x8D + vpermq %3, %3, 0xD8 + vpermq %4, %4, 0xD8 + ; transpose coefficients(phase 3) + ; %1=(01 11 21 31 41 51 61 71 00 10 20 30 40 50 60 70) + ; %2=(03 13 23 33 43 53 63 73 02 12 22 32 42 52 62 72) + ; %3=(04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75) + ; %4=(06 16 26 36 46 56 66 76 07 17 27 37 47 57 67 77) +%endmacro + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit accurate integer forward DCT using AVX2 instructions +; %1-%4: Input/output registers +; %5-%8: Temp registers +; %9: Pass (1 or 2) + +%macro DODCT 9 + vpsubw %5, %1, %4 ; %5=data1_0-data6_7=tmp6_7 + vpaddw %6, %1, %4 ; %6=data1_0+data6_7=tmp1_0 + vpaddw %7, %2, %3 ; %7=data3_2+data4_5=tmp3_2 + vpsubw %8, %2, %3 ; %8=data3_2-data4_5=tmp4_5 + + ; -- Even part + + vperm2i128 %6, %6, %6, 0x01 ; %6=tmp0_1 + vpaddw %1, %6, %7 ; %1=tmp0_1+tmp3_2=tmp10_11 + vpsubw %6, %6, %7 ; %6=tmp0_1-tmp3_2=tmp13_12 + + vperm2i128 %7, %1, %1, 0x01 ; %7=tmp11_10 + vpsignw %1, %1, [rel PW_1_NEG1] ; %1=tmp10_neg11 + vpaddw %7, %7, %1 ; %7=(tmp10+tmp11)_(tmp10-tmp11) +%if %9 == 1 + vpsllw %1, %7, PASS1_BITS ; %1=data0_4 +%else + vpaddw %7, %7, [rel PW_DESCALE_P2X] + vpsraw %1, %7, PASS1_BITS ; %1=data0_4 +%endif + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + vperm2i128 %7, %6, %6, 0x01 ; %7=tmp12_13 + vpunpcklwd %2, %6, %7 + vpunpckhwd %6, %6, %7 + vpmaddwd %2, %2, [rel PW_F130_F054_MF130_F054] ; %2=data2_6L + vpmaddwd %6, %6, [rel PW_F130_F054_MF130_F054] ; %6=data2_6H + + vpaddd %2, %2, [rel PD_DESCALE_P %+ %9] + vpaddd %6, %6, [rel PD_DESCALE_P %+ %9] + vpsrad %2, %2, DESCALE_P %+ %9 + vpsrad %6, %6, DESCALE_P %+ %9 + + vpackssdw %3, %2, %6 ; %6=data2_6 + + ; -- Odd part + + vpaddw %7, %8, %5 ; %7=tmp4_5+tmp6_7=z3_4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + vperm2i128 %2, %7, %7, 0x01 ; %2=z4_3 + vpunpcklwd %6, %7, %2 + vpunpckhwd %7, %7, %2 + vpmaddwd %6, %6, [rel PW_MF078_F117_F078_F117] ; %6=z3_4L + vpmaddwd %7, %7, [rel PW_MF078_F117_F078_F117] ; %7=z3_4H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + vperm2i128 %4, %5, %5, 0x01 ; %4=tmp7_6 + vpunpcklwd %2, %8, %4 + vpunpckhwd %4, %8, %4 + vpmaddwd %2, %2, [rel PW_MF060_MF089_MF050_MF256] ; %2=tmp4_5L + vpmaddwd %4, %4, [rel PW_MF060_MF089_MF050_MF256] ; %4=tmp4_5H + + vpaddd %2, %2, %6 ; %2=data7_5L + vpaddd %4, %4, %7 ; %4=data7_5H + + vpaddd %2, %2, [rel PD_DESCALE_P %+ %9] + vpaddd %4, %4, [rel PD_DESCALE_P %+ %9] + vpsrad %2, %2, DESCALE_P %+ %9 + vpsrad %4, %4, DESCALE_P %+ %9 + + vpackssdw %4, %2, %4 ; %4=data7_5 + + vperm2i128 %2, %8, %8, 0x01 ; %2=tmp5_4 + vpunpcklwd %8, %5, %2 + vpunpckhwd %5, %5, %2 + vpmaddwd %8, %8, [rel PW_F050_MF256_F060_MF089] ; %8=tmp6_7L + vpmaddwd %5, %5, [rel PW_F050_MF256_F060_MF089] ; %5=tmp6_7H + + vpaddd %8, %8, %6 ; %8=data3_1L + vpaddd %5, %5, %7 ; %5=data3_1H + + vpaddd %8, %8, [rel PD_DESCALE_P %+ %9] + vpaddd %5, %5, [rel PD_DESCALE_P %+ %9] + vpsrad %8, %8, DESCALE_P %+ %9 + vpsrad %5, %5, DESCALE_P %+ %9 + + vpackssdw %2, %8, %5 ; %2=data3_1 +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_islow_avx2) + +EXTN(jconst_fdct_islow_avx2): + +PW_F130_F054_MF130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 + times 4 dw (F_0_541 - F_1_847), F_0_541 +PW_MF078_F117_F078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 + times 4 dw (F_1_175 - F_0_390), F_1_175 +PW_MF060_MF089_MF050_MF256 times 4 dw (F_0_298 - F_0_899), -F_0_899 + times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_F050_MF256_F060_MF089 times 4 dw (F_3_072 - F_2_562), -F_2_562 + times 4 dw (F_1_501 - F_0_899), -F_0_899 +PD_DESCALE_P1 times 8 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 8 dd 1 << (DESCALE_P2 - 1) +PW_DESCALE_P2X times 16 dw 1 << (PASS1_BITS - 1) +PW_1_NEG1 times 8 dw 1 + times 8 dw -1 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_islow_avx2(DCTELEM *data) +; + +; r10 = DCTELEM *data + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_islow_avx2) + +EXTN(jsimd_fdct_islow_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 1 + + ; ---- Pass 1: process rows. + + vmovdqu ymm4, YMMWORD [YMMBLOCK(0,0,r10,SIZEOF_DCTELEM)] + vmovdqu ymm5, YMMWORD [YMMBLOCK(2,0,r10,SIZEOF_DCTELEM)] + vmovdqu ymm6, YMMWORD [YMMBLOCK(4,0,r10,SIZEOF_DCTELEM)] + vmovdqu ymm7, YMMWORD [YMMBLOCK(6,0,r10,SIZEOF_DCTELEM)] + ; ymm4=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + ; ymm5=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + ; ymm6=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) + ; ymm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) + + vperm2i128 ymm0, ymm4, ymm6, 0x20 + vperm2i128 ymm1, ymm4, ymm6, 0x31 + vperm2i128 ymm2, ymm5, ymm7, 0x20 + vperm2i128 ymm3, ymm5, ymm7, 0x31 + ; ymm0=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47) + ; ymm1=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57) + ; ymm2=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67) + ; ymm3=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77) + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7 + + DODCT ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7, 1 + ; ymm0=data0_4, ymm1=data3_1, ymm2=data2_6, ymm3=data7_5 + + ; ---- Pass 2: process columns. + + vperm2i128 ymm4, ymm1, ymm3, 0x20 ; ymm4=data3_7 + vperm2i128 ymm1, ymm1, ymm3, 0x31 ; ymm1=data1_5 + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7 + + DODCT ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7, 2 + ; ymm0=data0_4, ymm1=data3_1, ymm2=data2_6, ymm4=data7_5 + + vperm2i128 ymm3, ymm0, ymm1, 0x30 ; ymm3=data0_1 + vperm2i128 ymm5, ymm2, ymm1, 0x20 ; ymm5=data2_3 + vperm2i128 ymm6, ymm0, ymm4, 0x31 ; ymm6=data4_5 + vperm2i128 ymm7, ymm2, ymm4, 0x21 ; ymm7=data6_7 + + vmovdqu YMMWORD [YMMBLOCK(0,0,r10,SIZEOF_DCTELEM)], ymm3 + vmovdqu YMMWORD [YMMBLOCK(2,0,r10,SIZEOF_DCTELEM)], ymm5 + vmovdqu YMMWORD [YMMBLOCK(4,0,r10,SIZEOF_DCTELEM)], ymm6 + vmovdqu YMMWORD [YMMBLOCK(6,0,r10,SIZEOF_DCTELEM)], ymm7 + + vzeroupper + UNCOLLECT_ARGS 1 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-sse2.asm new file mode 100644 index 0000000..3a6be02 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jfdctint-sse2.asm @@ -0,0 +1,617 @@ +; +; jfdctint.asm - accurate integer FDCT (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2020, 2024, D. R. Commander. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; forward DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jfdctint.c; see the jfdctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_fdct_islow_sse2) + +EXTN(jconst_fdct_islow_sse2): + +PW_F130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 +PW_F054_MF130 times 4 dw F_0_541, (F_0_541 - F_1_847) +PW_MF078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 +PW_F117_F078 times 4 dw F_1_175, (F_1_175 - F_0_390) +PW_MF060_MF089 times 4 dw (F_0_298 - F_0_899), -F_0_899 +PW_MF089_F060 times 4 dw -F_0_899, (F_1_501 - F_0_899) +PW_MF050_MF256 times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF256_F050 times 4 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2 - 1) +PW_DESCALE_P2X times 8 dw 1 << (PASS1_BITS - 1) + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform the forward DCT on one block of samples. +; +; GLOBAL(void) +; jsimd_fdct_islow_sse2(DCTELEM *data) +; + +; r10 = DCTELEM *data + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD ; xmmword wk[WK_NUM] +%define WK_NUM 6 + + align 32 + GLOBAL_FUNCTION(jsimd_fdct_islow_sse2) + +EXTN(jsimd_fdct_islow_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 1 + + ; ---- Pass 1: process rows. + + mov rdx, r10 ; (DCTELEM *) + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm2, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)] + + ; xmm0=(00 01 02 03 04 05 06 07), xmm2=(20 21 22 23 24 25 26 27) + ; xmm1=(10 11 12 13 14 15 16 17), xmm3=(30 31 32 33 34 35 36 37) + + movdqa xmm4, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm1 ; xmm0=(00 10 01 11 02 12 03 13) + punpckhwd xmm4, xmm1 ; xmm4=(04 14 05 15 06 16 07 17) + movdqa xmm5, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm3 ; xmm2=(20 30 21 31 22 32 23 33) + punpckhwd xmm5, xmm3 ; xmm5=(24 34 25 35 26 36 27 37) + + movdqa xmm6, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm7, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm1, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)] + + ; xmm6=( 4 12 20 28 36 44 52 60), xmm1=( 6 14 22 30 38 46 54 62) + ; xmm7=( 5 13 21 29 37 45 53 61), xmm3=( 7 15 23 31 39 47 55 63) + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=(20 30 21 31 22 32 23 33) + movdqa XMMWORD [wk(1)], xmm5 ; wk(1)=(24 34 25 35 26 36 27 37) + + movdqa xmm2, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm7 ; xmm6=(40 50 41 51 42 52 43 53) + punpckhwd xmm2, xmm7 ; xmm2=(44 54 45 55 46 56 47 57) + movdqa xmm5, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm3 ; xmm1=(60 70 61 71 62 72 63 73) + punpckhwd xmm5, xmm3 ; xmm5=(64 74 65 75 66 76 67 77) + + movdqa xmm7, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm1 ; xmm6=(40 50 60 70 41 51 61 71) + punpckhdq xmm7, xmm1 ; xmm7=(42 52 62 72 43 53 63 73) + movdqa xmm3, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm5 ; xmm2=(44 54 64 74 45 55 65 75) + punpckhdq xmm3, xmm5 ; xmm3=(46 56 66 76 47 57 67 77) + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(20 30 21 31 22 32 23 33) + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=(24 34 25 35 26 36 27 37) + movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=(42 52 62 72 43 53 63 73) + movdqa XMMWORD [wk(3)], xmm2 ; wk(3)=(44 54 64 74 45 55 65 75) + + movdqa xmm7, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm1 ; xmm0=(00 10 20 30 01 11 21 31) + punpckhdq xmm7, xmm1 ; xmm7=(02 12 22 32 03 13 23 33) + movdqa xmm2, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm5 ; xmm4=(04 14 24 34 05 15 25 35) + punpckhdq xmm2, xmm5 ; xmm2=(06 16 26 36 07 17 27 37) + + movdqa xmm1, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm6 ; xmm0=(00 10 20 30 40 50 60 70)=data0 + punpckhqdq xmm1, xmm6 ; xmm1=(01 11 21 31 41 51 61 71)=data1 + movdqa xmm5, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm3 ; xmm2=(06 16 26 36 46 56 66 76)=data6 + punpckhqdq xmm5, xmm3 ; xmm5=(07 17 27 37 47 57 67 77)=data7 + + movdqa xmm6, xmm1 + movdqa xmm3, xmm0 + psubw xmm1, xmm2 ; xmm1=data1-data6=tmp6 + psubw xmm0, xmm5 ; xmm0=data0-data7=tmp7 + paddw xmm6, xmm2 ; xmm6=data1+data6=tmp1 + paddw xmm3, xmm5 ; xmm3=data0+data7=tmp0 + + movdqa xmm2, XMMWORD [wk(2)] ; xmm2=(42 52 62 72 43 53 63 73) + movdqa xmm5, XMMWORD [wk(3)] ; xmm5=(44 54 64 74 45 55 65 75) + movdqa XMMWORD [wk(0)], xmm1 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp7 + + movdqa xmm1, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm2 ; xmm7=(02 12 22 32 42 52 62 72)=data2 + punpckhqdq xmm1, xmm2 ; xmm1=(03 13 23 33 43 53 63 73)=data3 + movdqa xmm0, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm5 ; xmm4=(04 14 24 34 44 54 64 74)=data4 + punpckhqdq xmm0, xmm5 ; xmm0=(05 15 25 35 45 55 65 75)=data5 + + movdqa xmm2, xmm1 + movdqa xmm5, xmm7 + paddw xmm1, xmm4 ; xmm1=data3+data4=tmp3 + paddw xmm7, xmm0 ; xmm7=data2+data5=tmp2 + psubw xmm2, xmm4 ; xmm2=data3-data4=tmp4 + psubw xmm5, xmm0 ; xmm5=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm4, xmm3 + movdqa xmm0, xmm6 + paddw xmm3, xmm1 ; xmm3=tmp10 + paddw xmm6, xmm7 ; xmm6=tmp11 + psubw xmm4, xmm1 ; xmm4=tmp13 + psubw xmm0, xmm7 ; xmm0=tmp12 + + movdqa xmm1, xmm3 + paddw xmm3, xmm6 ; xmm3=tmp10+tmp11 + psubw xmm1, xmm6 ; xmm1=tmp10-tmp11 + + psllw xmm3, PASS1_BITS ; xmm3=data0 + psllw xmm1, PASS1_BITS ; xmm1=data4 + + movdqa XMMWORD [wk(2)], xmm3 ; wk(2)=data0 + movdqa XMMWORD [wk(3)], xmm1 ; wk(3)=data4 + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + movdqa xmm7, xmm4 ; xmm4=tmp13 + movdqa xmm6, xmm4 + punpcklwd xmm7, xmm0 ; xmm0=tmp12 + punpckhwd xmm6, xmm0 + movdqa xmm4, xmm7 + movdqa xmm0, xmm6 + pmaddwd xmm7, [rel PW_F130_F054] ; xmm7=data2L + pmaddwd xmm6, [rel PW_F130_F054] ; xmm6=data2H + pmaddwd xmm4, [rel PW_F054_MF130] ; xmm4=data6L + pmaddwd xmm0, [rel PW_F054_MF130] ; xmm0=data6H + + paddd xmm7, [rel PD_DESCALE_P1] + paddd xmm6, [rel PD_DESCALE_P1] + psrad xmm7, DESCALE_P1 + psrad xmm6, DESCALE_P1 + paddd xmm4, [rel PD_DESCALE_P1] + paddd xmm0, [rel PD_DESCALE_P1] + psrad xmm4, DESCALE_P1 + psrad xmm0, DESCALE_P1 + + packssdw xmm7, xmm6 ; xmm7=data2 + packssdw xmm4, xmm0 ; xmm4=data6 + + movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=data2 + movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=data6 + + ; -- Odd part + + movdqa xmm3, XMMWORD [wk(0)] ; xmm3=tmp6 + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp7 + + movdqa xmm6, xmm2 ; xmm2=tmp4 + movdqa xmm0, xmm5 ; xmm5=tmp5 + paddw xmm6, xmm3 ; xmm6=z3 + paddw xmm0, xmm1 ; xmm0=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm7, xmm6 + movdqa xmm4, xmm6 + punpcklwd xmm7, xmm0 + punpckhwd xmm4, xmm0 + movdqa xmm6, xmm7 + movdqa xmm0, xmm4 + pmaddwd xmm7, [rel PW_MF078_F117] ; xmm7=z3L + pmaddwd xmm4, [rel PW_MF078_F117] ; xmm4=z3H + pmaddwd xmm6, [rel PW_F117_F078] ; xmm6=z4L + pmaddwd xmm0, [rel PW_F117_F078] ; xmm0=z4H + + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=z3L + movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=z3H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + movdqa xmm7, xmm2 + movdqa xmm4, xmm2 + punpcklwd xmm7, xmm1 + punpckhwd xmm4, xmm1 + movdqa xmm2, xmm7 + movdqa xmm1, xmm4 + pmaddwd xmm7, [rel PW_MF060_MF089] ; xmm7=tmp4L + pmaddwd xmm4, [rel PW_MF060_MF089] ; xmm4=tmp4H + pmaddwd xmm2, [rel PW_MF089_F060] ; xmm2=tmp7L + pmaddwd xmm1, [rel PW_MF089_F060] ; xmm1=tmp7H + + paddd xmm7, XMMWORD [wk(0)] ; xmm7=data7L + paddd xmm4, XMMWORD [wk(1)] ; xmm4=data7H + paddd xmm2, xmm6 ; xmm2=data1L + paddd xmm1, xmm0 ; xmm1=data1H + + paddd xmm7, [rel PD_DESCALE_P1] + paddd xmm4, [rel PD_DESCALE_P1] + psrad xmm7, DESCALE_P1 + psrad xmm4, DESCALE_P1 + paddd xmm2, [rel PD_DESCALE_P1] + paddd xmm1, [rel PD_DESCALE_P1] + psrad xmm2, DESCALE_P1 + psrad xmm1, DESCALE_P1 + + packssdw xmm7, xmm4 ; xmm7=data7 + packssdw xmm2, xmm1 ; xmm2=data1 + + movdqa xmm4, xmm5 + movdqa xmm1, xmm5 + punpcklwd xmm4, xmm3 + punpckhwd xmm1, xmm3 + movdqa xmm5, xmm4 + movdqa xmm3, xmm1 + pmaddwd xmm4, [rel PW_MF050_MF256] ; xmm4=tmp5L + pmaddwd xmm1, [rel PW_MF050_MF256] ; xmm1=tmp5H + pmaddwd xmm5, [rel PW_MF256_F050] ; xmm5=tmp6L + pmaddwd xmm3, [rel PW_MF256_F050] ; xmm3=tmp6H + + paddd xmm4, xmm6 ; xmm4=data5L + paddd xmm1, xmm0 ; xmm1=data5H + paddd xmm5, XMMWORD [wk(0)] ; xmm5=data3L + paddd xmm3, XMMWORD [wk(1)] ; xmm3=data3H + + paddd xmm4, [rel PD_DESCALE_P1] + paddd xmm1, [rel PD_DESCALE_P1] + psrad xmm4, DESCALE_P1 + psrad xmm1, DESCALE_P1 + paddd xmm5, [rel PD_DESCALE_P1] + paddd xmm3, [rel PD_DESCALE_P1] + psrad xmm5, DESCALE_P1 + psrad xmm3, DESCALE_P1 + + packssdw xmm4, xmm1 ; xmm4=data5 + packssdw xmm5, xmm3 ; xmm5=data3 + + ; ---- Pass 2: process columns. + + movdqa xmm6, XMMWORD [wk(2)] ; xmm6=col0 + movdqa xmm0, XMMWORD [wk(4)] ; xmm0=col2 + + ; xmm6=(00 10 20 30 40 50 60 70), xmm0=(02 12 22 32 42 52 62 72) + ; xmm2=(01 11 21 31 41 51 61 71), xmm5=(03 13 23 33 43 53 63 73) + + movdqa xmm1, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm2 ; xmm6=(00 01 10 11 20 21 30 31) + punpckhwd xmm1, xmm2 ; xmm1=(40 41 50 51 60 61 70 71) + movdqa xmm3, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm5 ; xmm0=(02 03 12 13 22 23 32 33) + punpckhwd xmm3, xmm5 ; xmm3=(42 43 52 53 62 63 72 73) + + movdqa xmm2, XMMWORD [wk(3)] ; xmm2=col4 + movdqa xmm5, XMMWORD [wk(5)] ; xmm5=col6 + + ; xmm2=(04 14 24 34 44 54 64 74), xmm5=(06 16 26 36 46 56 66 76) + ; xmm4=(05 15 25 35 45 55 65 75), xmm7=(07 17 27 37 47 57 67 77) + + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=(02 03 12 13 22 23 32 33) + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(42 43 52 53 62 63 72 73) + + movdqa xmm0, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm4 ; xmm2=(04 05 14 15 24 25 34 35) + punpckhwd xmm0, xmm4 ; xmm0=(44 45 54 55 64 65 74 75) + movdqa xmm3, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm7 ; xmm5=(06 07 16 17 26 27 36 37) + punpckhwd xmm3, xmm7 ; xmm3=(46 47 56 57 66 67 76 77) + + movdqa xmm4, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm5 ; xmm2=(04 05 06 07 14 15 16 17) + punpckhdq xmm4, xmm5 ; xmm4=(24 25 26 27 34 35 36 37) + movdqa xmm7, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm3 ; xmm0=(44 45 46 47 54 55 56 57) + punpckhdq xmm7, xmm3 ; xmm7=(64 65 66 67 74 75 76 77) + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=(02 03 12 13 22 23 32 33) + movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(42 43 52 53 62 63 72 73) + movdqa XMMWORD [wk(2)], xmm4 ; wk(2)=(24 25 26 27 34 35 36 37) + movdqa XMMWORD [wk(3)], xmm0 ; wk(3)=(44 45 46 47 54 55 56 57) + + movdqa xmm4, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm5 ; xmm6=(00 01 02 03 10 11 12 13) + punpckhdq xmm4, xmm5 ; xmm4=(20 21 22 23 30 31 32 33) + movdqa xmm0, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm3 ; xmm1=(40 41 42 43 50 51 52 53) + punpckhdq xmm0, xmm3 ; xmm0=(60 61 62 63 70 71 72 73) + + movdqa xmm5, xmm6 ; transpose coefficients(phase 3) + punpcklqdq xmm6, xmm2 ; xmm6=(00 01 02 03 04 05 06 07)=data0 + punpckhqdq xmm5, xmm2 ; xmm5=(10 11 12 13 14 15 16 17)=data1 + movdqa xmm3, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm7 ; xmm0=(60 61 62 63 64 65 66 67)=data6 + punpckhqdq xmm3, xmm7 ; xmm3=(70 71 72 73 74 75 76 77)=data7 + + movdqa xmm2, xmm5 + movdqa xmm7, xmm6 + psubw xmm5, xmm0 ; xmm5=data1-data6=tmp6 + psubw xmm6, xmm3 ; xmm6=data0-data7=tmp7 + paddw xmm2, xmm0 ; xmm2=data1+data6=tmp1 + paddw xmm7, xmm3 ; xmm7=data0+data7=tmp0 + + movdqa xmm0, XMMWORD [wk(2)] ; xmm0=(24 25 26 27 34 35 36 37) + movdqa xmm3, XMMWORD [wk(3)] ; xmm3=(44 45 46 47 54 55 56 57) + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=tmp6 + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp7 + + movdqa xmm5, xmm4 ; transpose coefficients(phase 3) + punpcklqdq xmm4, xmm0 ; xmm4=(20 21 22 23 24 25 26 27)=data2 + punpckhqdq xmm5, xmm0 ; xmm5=(30 31 32 33 34 35 36 37)=data3 + movdqa xmm6, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm3 ; xmm1=(40 41 42 43 44 45 46 47)=data4 + punpckhqdq xmm6, xmm3 ; xmm6=(50 51 52 53 54 55 56 57)=data5 + + movdqa xmm0, xmm5 + movdqa xmm3, xmm4 + paddw xmm5, xmm1 ; xmm5=data3+data4=tmp3 + paddw xmm4, xmm6 ; xmm4=data2+data5=tmp2 + psubw xmm0, xmm1 ; xmm0=data3-data4=tmp4 + psubw xmm3, xmm6 ; xmm3=data2-data5=tmp5 + + ; -- Even part + + movdqa xmm1, xmm7 + movdqa xmm6, xmm2 + paddw xmm7, xmm5 ; xmm7=tmp10 + paddw xmm2, xmm4 ; xmm2=tmp11 + psubw xmm1, xmm5 ; xmm1=tmp13 + psubw xmm6, xmm4 ; xmm6=tmp12 + + movdqa xmm5, xmm7 + paddw xmm7, xmm2 ; xmm7=tmp10+tmp11 + psubw xmm5, xmm2 ; xmm5=tmp10-tmp11 + + paddw xmm7, [rel PW_DESCALE_P2X] + paddw xmm5, [rel PW_DESCALE_P2X] + psraw xmm7, PASS1_BITS ; xmm7=data0 + psraw xmm5, PASS1_BITS ; xmm5=data4 + + movdqa XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_DCTELEM)], xmm7 + movdqa XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_DCTELEM)], xmm5 + + ; (Original) + ; z1 = (tmp12 + tmp13) * 0.541196100; + ; data2 = z1 + tmp13 * 0.765366865; + ; data6 = z1 + tmp12 * -1.847759065; + ; + ; (This implementation) + ; data2 = tmp13 * (0.541196100 + 0.765366865) + tmp12 * 0.541196100; + ; data6 = tmp13 * 0.541196100 + tmp12 * (0.541196100 - 1.847759065); + + movdqa xmm4, xmm1 ; xmm1=tmp13 + movdqa xmm2, xmm1 + punpcklwd xmm4, xmm6 ; xmm6=tmp12 + punpckhwd xmm2, xmm6 + movdqa xmm1, xmm4 + movdqa xmm6, xmm2 + pmaddwd xmm4, [rel PW_F130_F054] ; xmm4=data2L + pmaddwd xmm2, [rel PW_F130_F054] ; xmm2=data2H + pmaddwd xmm1, [rel PW_F054_MF130] ; xmm1=data6L + pmaddwd xmm6, [rel PW_F054_MF130] ; xmm6=data6H + + paddd xmm4, [rel PD_DESCALE_P2] + paddd xmm2, [rel PD_DESCALE_P2] + psrad xmm4, DESCALE_P2 + psrad xmm2, DESCALE_P2 + paddd xmm1, [rel PD_DESCALE_P2] + paddd xmm6, [rel PD_DESCALE_P2] + psrad xmm1, DESCALE_P2 + psrad xmm6, DESCALE_P2 + + packssdw xmm4, xmm2 ; xmm4=data2 + packssdw xmm1, xmm6 ; xmm1=data6 + + movdqa XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_DCTELEM)], xmm4 + movdqa XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_DCTELEM)], xmm1 + + ; -- Odd part + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp6 + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp7 + + movdqa xmm2, xmm0 ; xmm0=tmp4 + movdqa xmm6, xmm3 ; xmm3=tmp5 + paddw xmm2, xmm7 ; xmm2=z3 + paddw xmm6, xmm5 ; xmm6=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm4, xmm2 + movdqa xmm1, xmm2 + punpcklwd xmm4, xmm6 + punpckhwd xmm1, xmm6 + movdqa xmm2, xmm4 + movdqa xmm6, xmm1 + pmaddwd xmm4, [rel PW_MF078_F117] ; xmm4=z3L + pmaddwd xmm1, [rel PW_MF078_F117] ; xmm1=z3H + pmaddwd xmm2, [rel PW_F117_F078] ; xmm2=z4L + pmaddwd xmm6, [rel PW_F117_F078] ; xmm6=z4H + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=z3L + movdqa XMMWORD [wk(1)], xmm1 ; wk(1)=z3H + + ; (Original) + ; z1 = tmp4 + tmp7; z2 = tmp5 + tmp6; + ; tmp4 = tmp4 * 0.298631336; tmp5 = tmp5 * 2.053119869; + ; tmp6 = tmp6 * 3.072711026; tmp7 = tmp7 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; data7 = tmp4 + z1 + z3; data5 = tmp5 + z2 + z4; + ; data3 = tmp6 + z2 + z3; data1 = tmp7 + z1 + z4; + ; + ; (This implementation) + ; tmp4 = tmp4 * (0.298631336 - 0.899976223) + tmp7 * -0.899976223; + ; tmp5 = tmp5 * (2.053119869 - 2.562915447) + tmp6 * -2.562915447; + ; tmp6 = tmp5 * -2.562915447 + tmp6 * (3.072711026 - 2.562915447); + ; tmp7 = tmp4 * -0.899976223 + tmp7 * (1.501321110 - 0.899976223); + ; data7 = tmp4 + z3; data5 = tmp5 + z4; + ; data3 = tmp6 + z3; data1 = tmp7 + z4; + + movdqa xmm4, xmm0 + movdqa xmm1, xmm0 + punpcklwd xmm4, xmm5 + punpckhwd xmm1, xmm5 + movdqa xmm0, xmm4 + movdqa xmm5, xmm1 + pmaddwd xmm4, [rel PW_MF060_MF089] ; xmm4=tmp4L + pmaddwd xmm1, [rel PW_MF060_MF089] ; xmm1=tmp4H + pmaddwd xmm0, [rel PW_MF089_F060] ; xmm0=tmp7L + pmaddwd xmm5, [rel PW_MF089_F060] ; xmm5=tmp7H + + paddd xmm4, XMMWORD [wk(0)] ; xmm4=data7L + paddd xmm1, XMMWORD [wk(1)] ; xmm1=data7H + paddd xmm0, xmm2 ; xmm0=data1L + paddd xmm5, xmm6 ; xmm5=data1H + + paddd xmm4, [rel PD_DESCALE_P2] + paddd xmm1, [rel PD_DESCALE_P2] + psrad xmm4, DESCALE_P2 + psrad xmm1, DESCALE_P2 + paddd xmm0, [rel PD_DESCALE_P2] + paddd xmm5, [rel PD_DESCALE_P2] + psrad xmm0, DESCALE_P2 + psrad xmm5, DESCALE_P2 + + packssdw xmm4, xmm1 ; xmm4=data7 + packssdw xmm0, xmm5 ; xmm0=data1 + + movdqa XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_DCTELEM)], xmm4 + movdqa XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_DCTELEM)], xmm0 + + movdqa xmm1, xmm3 + movdqa xmm5, xmm3 + punpcklwd xmm1, xmm7 + punpckhwd xmm5, xmm7 + movdqa xmm3, xmm1 + movdqa xmm7, xmm5 + pmaddwd xmm1, [rel PW_MF050_MF256] ; xmm1=tmp5L + pmaddwd xmm5, [rel PW_MF050_MF256] ; xmm5=tmp5H + pmaddwd xmm3, [rel PW_MF256_F050] ; xmm3=tmp6L + pmaddwd xmm7, [rel PW_MF256_F050] ; xmm7=tmp6H + + paddd xmm1, xmm2 ; xmm1=data5L + paddd xmm5, xmm6 ; xmm5=data5H + paddd xmm3, XMMWORD [wk(0)] ; xmm3=data3L + paddd xmm7, XMMWORD [wk(1)] ; xmm7=data3H + + paddd xmm1, [rel PD_DESCALE_P2] + paddd xmm5, [rel PD_DESCALE_P2] + psrad xmm1, DESCALE_P2 + psrad xmm5, DESCALE_P2 + paddd xmm3, [rel PD_DESCALE_P2] + paddd xmm7, [rel PD_DESCALE_P2] + psrad xmm3, DESCALE_P2 + psrad xmm7, DESCALE_P2 + + packssdw xmm1, xmm5 ; xmm1=data5 + packssdw xmm3, xmm7 ; xmm3=data3 + + movdqa XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_DCTELEM)], xmm1 + movdqa XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_DCTELEM)], xmm3 + + UNCOLLECT_ARGS 1 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctflt-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctflt-sse2.asm new file mode 100644 index 0000000..1443734 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctflt-sse2.asm @@ -0,0 +1,478 @@ +; +; jidctflt.asm - floating-point IDCT (64-bit SSE & SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a floating-point implementation of the inverse DCT +; (Discrete Cosine Transform). The following code is based directly on +; the IJG's original jidctflt.c; see the jidctflt.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%macro UNPCKLPS2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(0 1 4 5) + shufps %1, %2, 0x44 +%endmacro + +%macro UNPCKHPS2 2 ; %1=(0 1 2 3) / %2=(4 5 6 7) => %1=(2 3 6 7) + shufps %1, %2, 0xEE +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_float_sse2) + +EXTN(jconst_idct_float_sse2): + +PD_1_414 times 4 dd 1.414213562373095048801689 +PD_1_847 times 4 dd 1.847759065022573512256366 +PD_1_082 times 4 dd 1.082392200292393968799446 +PD_M2_613 times 4 dd -2.613125929752753055713286 +PD_RNDINT_MAGIC times 4 dd 100663296.0 ; (float)(0x00C00000 << 3) +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_float_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +; r10 = void *dct_table +; r11 = JCOEFPTR coef_block +; r12 = JSAMPARRAY output_buf +; r13d = JDIMENSION output_col + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 +%define workspace wk(0) - DCTSIZE2 * SIZEOF_FAST_FLOAT + ; FAST_FLOAT workspace[DCTSIZE2] + + align 32 + GLOBAL_FUNCTION(jsimd_idct_float_sse2) + +EXTN(jsimd_idct_float_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + lea rsp, [workspace] + COLLECT_ARGS 4 + push rbx + + ; ---- Pass 1: process columns from input, store into work array. + + mov rdx, r10 ; quantptr + mov rsi, r11 ; inptr + lea rdi, [workspace] ; FAST_FLOAT *wsptr + mov rcx, DCTSIZE/4 ; ctr +.columnloop: +%ifndef NO_ZERO_COLUMN_TEST_FLOAT_SSE + mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movq xmm1, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movq xmm2, XMM_MMWORD [MMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + movq xmm3, XMM_MMWORD [MMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + movq xmm4, XMM_MMWORD [MMBLOCK(4,0,rsi,SIZEOF_JCOEF)] + movq xmm5, XMM_MMWORD [MMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + movq xmm6, XMM_MMWORD [MMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + movq xmm7, XMM_MMWORD [MMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + por xmm1, xmm2 + por xmm3, xmm4 + por xmm5, xmm6 + por xmm1, xmm3 + por xmm5, xmm7 + por xmm1, xmm5 + packsswb xmm1, xmm1 + movd eax, xmm1 + test rax, rax + jnz short .columnDCT + + ; -- AC terms all zero + + movq xmm0, XMM_MMWORD [MMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03) + cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03) + + mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm1, xmm0 + movaps xmm2, xmm0 + movaps xmm3, xmm0 + + shufps xmm0, xmm0, 0x00 ; xmm0=(00 00 00 00) + shufps xmm1, xmm1, 0x55 ; xmm1=(01 01 01 01) + shufps xmm2, xmm2, 0xAA ; xmm2=(02 02 02 02) + shufps xmm3, xmm3, 0xFF ; xmm3=(03 03 03 03) + + movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(2,1,rdi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(3,1,rdi,SIZEOF_FAST_FLOAT)], xmm3 + jmp near .nextcolumn +%endif +.columnDCT: + + ; -- Even part + + movq xmm0, XMM_MMWORD [MMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + movq xmm1, XMM_MMWORD [MMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + movq xmm2, XMM_MMWORD [MMBLOCK(4,0,rsi,SIZEOF_JCOEF)] + movq xmm3, XMM_MMWORD [MMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + punpcklwd xmm1, xmm1 ; xmm1=(20 20 21 21 22 22 23 23) + psrad xmm0, (DWORD_BIT-WORD_BIT) ; xmm0=in0=(00 01 02 03) + psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in2=(20 21 22 23) + cvtdq2ps xmm0, xmm0 ; xmm0=in0=(00 01 02 03) + cvtdq2ps xmm1, xmm1 ; xmm1=in2=(20 21 22 23) + + punpcklwd xmm2, xmm2 ; xmm2=(40 40 41 41 42 42 43 43) + punpcklwd xmm3, xmm3 ; xmm3=(60 60 61 61 62 62 63 63) + psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in4=(40 41 42 43) + psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in6=(60 61 62 63) + cvtdq2ps xmm2, xmm2 ; xmm2=in4=(40 41 42 43) + cvtdq2ps xmm3, xmm3 ; xmm3=in6=(60 61 62 63) + + mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm4, xmm0 + movaps xmm5, xmm1 + subps xmm0, xmm2 ; xmm0=tmp11 + subps xmm1, xmm3 + addps xmm4, xmm2 ; xmm4=tmp10 + addps xmm5, xmm3 ; xmm5=tmp13 + + mulps xmm1, [rel PD_1_414] + subps xmm1, xmm5 ; xmm1=tmp12 + + movaps xmm6, xmm4 + movaps xmm7, xmm0 + subps xmm4, xmm5 ; xmm4=tmp3 + subps xmm0, xmm1 ; xmm0=tmp2 + addps xmm6, xmm5 ; xmm6=tmp0 + addps xmm7, xmm1 ; xmm7=tmp1 + + movaps XMMWORD [wk(1)], xmm4 ; tmp3 + movaps XMMWORD [wk(0)], xmm0 ; tmp2 + + ; -- Odd part + + movq xmm2, XMM_MMWORD [MMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movq xmm3, XMM_MMWORD [MMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + movq xmm5, XMM_MMWORD [MMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + movq xmm1, XMM_MMWORD [MMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + + punpcklwd xmm2, xmm2 ; xmm2=(10 10 11 11 12 12 13 13) + punpcklwd xmm3, xmm3 ; xmm3=(30 30 31 31 32 32 33 33) + psrad xmm2, (DWORD_BIT-WORD_BIT) ; xmm2=in1=(10 11 12 13) + psrad xmm3, (DWORD_BIT-WORD_BIT) ; xmm3=in3=(30 31 32 33) + cvtdq2ps xmm2, xmm2 ; xmm2=in1=(10 11 12 13) + cvtdq2ps xmm3, xmm3 ; xmm3=in3=(30 31 32 33) + + punpcklwd xmm5, xmm5 ; xmm5=(50 50 51 51 52 52 53 53) + punpcklwd xmm1, xmm1 ; xmm1=(70 70 71 71 72 72 73 73) + psrad xmm5, (DWORD_BIT-WORD_BIT) ; xmm5=in5=(50 51 52 53) + psrad xmm1, (DWORD_BIT-WORD_BIT) ; xmm1=in7=(70 71 72 73) + cvtdq2ps xmm5, xmm5 ; xmm5=in5=(50 51 52 53) + cvtdq2ps xmm1, xmm1 ; xmm1=in7=(70 71 72 73) + + mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm5, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + mulps xmm1, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_FLOAT_MULT_TYPE)] + + movaps xmm4, xmm2 + movaps xmm0, xmm5 + addps xmm2, xmm1 ; xmm2=z11 + addps xmm5, xmm3 ; xmm5=z13 + subps xmm4, xmm1 ; xmm4=z12 + subps xmm0, xmm3 ; xmm0=z10 + + movaps xmm1, xmm2 + subps xmm2, xmm5 + addps xmm1, xmm5 ; xmm1=tmp7 + + mulps xmm2, [rel PD_1_414] ; xmm2=tmp11 + + movaps xmm3, xmm0 + addps xmm0, xmm4 + mulps xmm0, [rel PD_1_847] ; xmm0=z5 + mulps xmm3, [rel PD_M2_613] ; xmm3=(z10 * -2.613125930) + mulps xmm4, [rel PD_1_082] ; xmm4=(z12 * 1.082392200) + addps xmm3, xmm0 ; xmm3=tmp12 + subps xmm4, xmm0 ; xmm4=tmp10 + + ; -- Final output stage + + subps xmm3, xmm1 ; xmm3=tmp6 + movaps xmm5, xmm6 + movaps xmm0, xmm7 + addps xmm6, xmm1 ; xmm6=data0=(00 01 02 03) + addps xmm7, xmm3 ; xmm7=data1=(10 11 12 13) + subps xmm5, xmm1 ; xmm5=data7=(70 71 72 73) + subps xmm0, xmm3 ; xmm0=data6=(60 61 62 63) + subps xmm2, xmm3 ; xmm2=tmp5 + + movaps xmm1, xmm6 ; transpose coefficients(phase 1) + unpcklps xmm6, xmm7 ; xmm6=(00 10 01 11) + unpckhps xmm1, xmm7 ; xmm1=(02 12 03 13) + movaps xmm3, xmm0 ; transpose coefficients(phase 1) + unpcklps xmm0, xmm5 ; xmm0=(60 70 61 71) + unpckhps xmm3, xmm5 ; xmm3=(62 72 63 73) + + movaps xmm7, XMMWORD [wk(0)] ; xmm7=tmp2 + movaps xmm5, XMMWORD [wk(1)] ; xmm5=tmp3 + + movaps XMMWORD [wk(0)], xmm0 ; wk(0)=(60 70 61 71) + movaps XMMWORD [wk(1)], xmm3 ; wk(1)=(62 72 63 73) + + addps xmm4, xmm2 ; xmm4=tmp4 + movaps xmm0, xmm7 + movaps xmm3, xmm5 + addps xmm7, xmm2 ; xmm7=data2=(20 21 22 23) + addps xmm5, xmm4 ; xmm5=data4=(40 41 42 43) + subps xmm0, xmm2 ; xmm0=data5=(50 51 52 53) + subps xmm3, xmm4 ; xmm3=data3=(30 31 32 33) + + movaps xmm2, xmm7 ; transpose coefficients(phase 1) + unpcklps xmm7, xmm3 ; xmm7=(20 30 21 31) + unpckhps xmm2, xmm3 ; xmm2=(22 32 23 33) + movaps xmm4, xmm5 ; transpose coefficients(phase 1) + unpcklps xmm5, xmm0 ; xmm5=(40 50 41 51) + unpckhps xmm4, xmm0 ; xmm4=(42 52 43 53) + + movaps xmm3, xmm6 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm6, xmm7 ; xmm6=(00 10 20 30) + UNPCKHPS2 xmm3, xmm7 ; xmm3=(01 11 21 31) + movaps xmm0, xmm1 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm1, xmm2 ; xmm1=(02 12 22 32) + UNPCKHPS2 xmm0, xmm2 ; xmm0=(03 13 23 33) + + movaps xmm7, XMMWORD [wk(0)] ; xmm7=(60 70 61 71) + movaps xmm2, XMMWORD [wk(1)] ; xmm2=(62 72 63 73) + + movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm6 + movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_FAST_FLOAT)], xmm1 + movaps XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_FAST_FLOAT)], xmm0 + + movaps xmm6, xmm5 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm5, xmm7 ; xmm5=(40 50 60 70) + UNPCKHPS2 xmm6, xmm7 ; xmm6=(41 51 61 71) + movaps xmm3, xmm4 ; transpose coefficients(phase 2) + UNPCKLPS2 xmm4, xmm2 ; xmm4=(42 52 62 72) + UNPCKHPS2 xmm3, xmm2 ; xmm3=(43 53 63 73) + + movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm5 + movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm6 + movaps XMMWORD [XMMBLOCK(2,1,rdi,SIZEOF_FAST_FLOAT)], xmm4 + movaps XMMWORD [XMMBLOCK(3,1,rdi,SIZEOF_FAST_FLOAT)], xmm3 + +.nextcolumn: + add rsi, byte 4*SIZEOF_JCOEF ; coef_block + add rdx, byte 4*SIZEOF_FLOAT_MULT_TYPE ; quantptr + add rdi, 4*DCTSIZE*SIZEOF_FAST_FLOAT ; wsptr + dec rcx ; ctr + jnz near .columnloop + + ; -- Prefetch the next coefficient block + + prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 0*32] + prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 1*32] + prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 2*32] + prefetchnta [rsi + (DCTSIZE2-8)*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + lea rsi, [workspace] ; FAST_FLOAT *wsptr + mov rdi, r12 ; (JSAMPROW *) + mov eax, r13d + mov rcx, DCTSIZE/4 ; ctr +.rowloop: + + ; -- Even part + + movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_FAST_FLOAT)] + + movaps xmm4, xmm0 + movaps xmm5, xmm1 + subps xmm0, xmm2 ; xmm0=tmp11 + subps xmm1, xmm3 + addps xmm4, xmm2 ; xmm4=tmp10 + addps xmm5, xmm3 ; xmm5=tmp13 + + mulps xmm1, [rel PD_1_414] + subps xmm1, xmm5 ; xmm1=tmp12 + + movaps xmm6, xmm4 + movaps xmm7, xmm0 + subps xmm4, xmm5 ; xmm4=tmp3 + subps xmm0, xmm1 ; xmm0=tmp2 + addps xmm6, xmm5 ; xmm6=tmp0 + addps xmm7, xmm1 ; xmm7=tmp1 + + movaps XMMWORD [wk(1)], xmm4 ; tmp3 + movaps XMMWORD [wk(0)], xmm0 ; tmp2 + + ; -- Odd part + + movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm5, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_FAST_FLOAT)] + + movaps xmm4, xmm2 + movaps xmm0, xmm5 + addps xmm2, xmm1 ; xmm2=z11 + addps xmm5, xmm3 ; xmm5=z13 + subps xmm4, xmm1 ; xmm4=z12 + subps xmm0, xmm3 ; xmm0=z10 + + movaps xmm1, xmm2 + subps xmm2, xmm5 + addps xmm1, xmm5 ; xmm1=tmp7 + + mulps xmm2, [rel PD_1_414] ; xmm2=tmp11 + + movaps xmm3, xmm0 + addps xmm0, xmm4 + mulps xmm0, [rel PD_1_847] ; xmm0=z5 + mulps xmm3, [rel PD_M2_613] ; xmm3=(z10 * -2.613125930) + mulps xmm4, [rel PD_1_082] ; xmm4=(z12 * 1.082392200) + addps xmm3, xmm0 ; xmm3=tmp12 + subps xmm4, xmm0 ; xmm4=tmp10 + + ; -- Final output stage + + subps xmm3, xmm1 ; xmm3=tmp6 + movaps xmm5, xmm6 + movaps xmm0, xmm7 + addps xmm6, xmm1 ; xmm6=data0=(00 10 20 30) + addps xmm7, xmm3 ; xmm7=data1=(01 11 21 31) + subps xmm5, xmm1 ; xmm5=data7=(07 17 27 37) + subps xmm0, xmm3 ; xmm0=data6=(06 16 26 36) + subps xmm2, xmm3 ; xmm2=tmp5 + + movaps xmm1, [rel PD_RNDINT_MAGIC] ; xmm1=[rel PD_RNDINT_MAGIC] + pcmpeqd xmm3, xmm3 + psrld xmm3, WORD_BIT ; xmm3={0xFFFF 0x0000 0xFFFF 0x0000 ..} + + addps xmm6, xmm1 ; xmm6=roundint(data0/8)=(00 ** 10 ** 20 ** 30 **) + addps xmm7, xmm1 ; xmm7=roundint(data1/8)=(01 ** 11 ** 21 ** 31 **) + addps xmm0, xmm1 ; xmm0=roundint(data6/8)=(06 ** 16 ** 26 ** 36 **) + addps xmm5, xmm1 ; xmm5=roundint(data7/8)=(07 ** 17 ** 27 ** 37 **) + + pand xmm6, xmm3 ; xmm6=(00 -- 10 -- 20 -- 30 --) + pslld xmm7, WORD_BIT ; xmm7=(-- 01 -- 11 -- 21 -- 31) + pand xmm0, xmm3 ; xmm0=(06 -- 16 -- 26 -- 36 --) + pslld xmm5, WORD_BIT ; xmm5=(-- 07 -- 17 -- 27 -- 37) + por xmm6, xmm7 ; xmm6=(00 01 10 11 20 21 30 31) + por xmm0, xmm5 ; xmm0=(06 07 16 17 26 27 36 37) + + movaps xmm1, XMMWORD [wk(0)] ; xmm1=tmp2 + movaps xmm3, XMMWORD [wk(1)] ; xmm3=tmp3 + + addps xmm4, xmm2 ; xmm4=tmp4 + movaps xmm7, xmm1 + movaps xmm5, xmm3 + addps xmm1, xmm2 ; xmm1=data2=(02 12 22 32) + addps xmm3, xmm4 ; xmm3=data4=(04 14 24 34) + subps xmm7, xmm2 ; xmm7=data5=(05 15 25 35) + subps xmm5, xmm4 ; xmm5=data3=(03 13 23 33) + + movaps xmm2, [rel PD_RNDINT_MAGIC] ; xmm2=[rel PD_RNDINT_MAGIC] + pcmpeqd xmm4, xmm4 + psrld xmm4, WORD_BIT ; xmm4={0xFFFF 0x0000 0xFFFF 0x0000 ..} + + addps xmm3, xmm2 ; xmm3=roundint(data4/8)=(04 ** 14 ** 24 ** 34 **) + addps xmm7, xmm2 ; xmm7=roundint(data5/8)=(05 ** 15 ** 25 ** 35 **) + addps xmm1, xmm2 ; xmm1=roundint(data2/8)=(02 ** 12 ** 22 ** 32 **) + addps xmm5, xmm2 ; xmm5=roundint(data3/8)=(03 ** 13 ** 23 ** 33 **) + + pand xmm3, xmm4 ; xmm3=(04 -- 14 -- 24 -- 34 --) + pslld xmm7, WORD_BIT ; xmm7=(-- 05 -- 15 -- 25 -- 35) + pand xmm1, xmm4 ; xmm1=(02 -- 12 -- 22 -- 32 --) + pslld xmm5, WORD_BIT ; xmm5=(-- 03 -- 13 -- 23 -- 33) + por xmm3, xmm7 ; xmm3=(04 05 14 15 24 25 34 35) + por xmm1, xmm5 ; xmm1=(02 03 12 13 22 23 32 33) + + movdqa xmm2, [rel PB_CENTERJSAMP] ; xmm2=[rel PB_CENTERJSAMP] + + packsswb xmm6, xmm3 ; xmm6=(00 01 10 11 20 21 30 31 04 05 14 15 24 25 34 35) + packsswb xmm1, xmm0 ; xmm1=(02 03 12 13 22 23 32 33 06 07 16 17 26 27 36 37) + paddb xmm6, xmm2 + paddb xmm1, xmm2 + + movdqa xmm4, xmm6 ; transpose coefficients(phase 2) + punpcklwd xmm6, xmm1 ; xmm6=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) + punpckhwd xmm4, xmm1 ; xmm4=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) + + movdqa xmm7, xmm6 ; transpose coefficients(phase 3) + punpckldq xmm6, xmm4 ; xmm6=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + punpckhdq xmm7, xmm4 ; xmm7=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + + pshufd xmm5, xmm6, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + pshufd xmm3, xmm7, 0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + + mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] + mov rbxp, JSAMPROW [rdi+2*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE], xmm7 + mov rdxp, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] + mov rbxp, JSAMPROW [rdi+3*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm5 + movq XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE], xmm3 + + add rsi, byte 4*SIZEOF_FAST_FLOAT ; wsptr + add rdi, byte 4*SIZEOF_JSAMPROW + dec rcx ; ctr + jnz near .rowloop + + pop rbx + UNCOLLECT_ARGS 4 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctfst-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctfst-sse2.asm new file mode 100644 index 0000000..cffabb8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctfst-sse2.asm @@ -0,0 +1,487 @@ +; +; jidctfst.asm - fast integer IDCT (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a fast, not so accurate integer implementation of +; the inverse DCT (Discrete Cosine Transform). The following code is +; based directly on the IJG's original jidctfst.c; see the jidctfst.c +; for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 8 ; 14 is also OK. +%define PASS1_BITS 2 + +%if IFAST_SCALE_BITS != PASS1_BITS +%error "'IFAST_SCALE_BITS' must be equal to 'PASS1_BITS'." +%endif + +%if CONST_BITS == 8 +F_1_082 equ 277 ; FIX(1.082392200) +F_1_414 equ 362 ; FIX(1.414213562) +F_1_847 equ 473 ; FIX(1.847759065) +F_2_613 equ 669 ; FIX(2.613125930) +F_1_613 equ (F_2_613 - 256) ; FIX(2.613125930) - FIX(1) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_1_082 equ DESCALE(1162209775, 30 - CONST_BITS) ; FIX(1.082392200) +F_1_414 equ DESCALE(1518500249, 30 - CONST_BITS) ; FIX(1.414213562) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_2_613 equ DESCALE(2805822602, 30 - CONST_BITS) ; FIX(2.613125930) +F_1_613 equ (F_2_613 - (1 << CONST_BITS)) ; FIX(2.613125930) - FIX(1) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + +; PRE_MULTIPLY_SCALE_BITS <= 2 (to avoid overflow) +; CONST_BITS + CONST_SHIFT + PRE_MULTIPLY_SCALE_BITS == 16 (for pmulhw) + +%define PRE_MULTIPLY_SCALE_BITS 2 +%define CONST_SHIFT (16 - PRE_MULTIPLY_SCALE_BITS - CONST_BITS) + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_ifast_sse2) + +EXTN(jconst_idct_ifast_sse2): + +PW_F1414 times 8 dw F_1_414 << CONST_SHIFT +PW_F1847 times 8 dw F_1_847 << CONST_SHIFT +PW_MF1613 times 8 dw -F_1_613 << CONST_SHIFT +PW_F1082 times 8 dw F_1_082 << CONST_SHIFT +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_ifast_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +; r10 = jpeg_component_info *compptr +; r11 = JCOEFPTR coef_block +; r12 = JSAMPARRAY output_buf +; r13d = JDIMENSION output_col + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_idct_ifast_sse2) + +EXTN(jsimd_idct_ifast_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 4 + + ; ---- Pass 1: process columns from input. + + mov rdx, r10 ; quantptr + mov rsi, r11 ; inptr + +%ifndef NO_ZERO_COLUMN_TEST_IFAST_SSE2 + mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + por xmm1, xmm0 + packsswb xmm1, xmm1 + packsswb xmm1, xmm1 + movd eax, xmm1 + test rax, rax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + movdqa xmm7, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07) + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + punpckhwd xmm7, xmm7 ; xmm7=(04 04 05 05 06 06 07 07) + + pshufd xmm6, xmm0, 0x00 ; xmm6=col0=(00 00 00 00 00 00 00 00) + pshufd xmm2, xmm0, 0x55 ; xmm2=col1=(01 01 01 01 01 01 01 01) + pshufd xmm5, xmm0, 0xAA ; xmm5=col2=(02 02 02 02 02 02 02 02) + pshufd xmm0, xmm0, 0xFF ; xmm0=col3=(03 03 03 03 03 03 03 03) + pshufd xmm1, xmm7, 0x00 ; xmm1=col4=(04 04 04 04 04 04 04 04) + pshufd xmm4, xmm7, 0x55 ; xmm4=col5=(05 05 05 05 05 05 05 05) + pshufd xmm3, xmm7, 0xAA ; xmm3=col6=(06 06 06 06 06 06 06 06) + pshufd xmm7, xmm7, 0xFF ; xmm7=col7=(07 07 07 07 07 07 07 07) + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=col1 + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=col3 + jmp near .column_end +%endif +.columnDCT: + + ; -- Even part + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + + movdqa xmm4, xmm0 + movdqa xmm5, xmm1 + psubw xmm0, xmm2 ; xmm0=tmp11 + psubw xmm1, xmm3 + paddw xmm4, xmm2 ; xmm4=tmp10 + paddw xmm5, xmm3 ; xmm5=tmp13 + + psllw xmm1, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm1, [rel PW_F1414] + psubw xmm1, xmm5 ; xmm1=tmp12 + + movdqa xmm6, xmm4 + movdqa xmm7, xmm0 + psubw xmm4, xmm5 ; xmm4=tmp3 + psubw xmm0, xmm1 ; xmm0=tmp2 + paddw xmm6, xmm5 ; xmm6=tmp0 + paddw xmm7, xmm1 ; xmm7=tmp1 + + movdqa XMMWORD [wk(1)], xmm4 ; wk(1)=tmp3 + movdqa XMMWORD [wk(0)], xmm0 ; wk(0)=tmp2 + + ; -- Odd part + + movdqa xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + movdqa xmm5, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + pmullw xmm5, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_IFAST_MULT_TYPE)] + + movdqa xmm4, xmm2 + movdqa xmm0, xmm5 + psubw xmm2, xmm1 ; xmm2=z12 + psubw xmm5, xmm3 ; xmm5=z10 + paddw xmm4, xmm1 ; xmm4=z11 + paddw xmm0, xmm3 ; xmm0=z13 + + movdqa xmm1, xmm5 ; xmm1=z10(unscaled) + psllw xmm2, PRE_MULTIPLY_SCALE_BITS + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + + movdqa xmm3, xmm4 + psubw xmm4, xmm0 + paddw xmm3, xmm0 ; xmm3=tmp7 + + psllw xmm4, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm4, [rel PW_F1414] ; xmm4=tmp11 + + ; To avoid overflow... + ; + ; (Original) + ; tmp12 = -2.613125930 * z10 + z5; + ; + ; (This implementation) + ; tmp12 = (-1.613125930 - 1) * z10 + z5; + ; = -1.613125930 * z10 - z10 + z5; + + movdqa xmm0, xmm5 + paddw xmm5, xmm2 + pmulhw xmm5, [rel PW_F1847] ; xmm5=z5 + pmulhw xmm0, [rel PW_MF1613] + pmulhw xmm2, [rel PW_F1082] + psubw xmm0, xmm1 + psubw xmm2, xmm5 ; xmm2=tmp10 + paddw xmm0, xmm5 ; xmm0=tmp12 + + ; -- Final output stage + + psubw xmm0, xmm3 ; xmm0=tmp6 + movdqa xmm1, xmm6 + movdqa xmm5, xmm7 + paddw xmm6, xmm3 ; xmm6=data0=(00 01 02 03 04 05 06 07) + paddw xmm7, xmm0 ; xmm7=data1=(10 11 12 13 14 15 16 17) + psubw xmm1, xmm3 ; xmm1=data7=(70 71 72 73 74 75 76 77) + psubw xmm5, xmm0 ; xmm5=data6=(60 61 62 63 64 65 66 67) + psubw xmm4, xmm0 ; xmm4=tmp5 + + movdqa xmm3, xmm6 ; transpose coefficients(phase 1) + punpcklwd xmm6, xmm7 ; xmm6=(00 10 01 11 02 12 03 13) + punpckhwd xmm3, xmm7 ; xmm3=(04 14 05 15 06 16 07 17) + movdqa xmm0, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm1 ; xmm5=(60 70 61 71 62 72 63 73) + punpckhwd xmm0, xmm1 ; xmm0=(64 74 65 75 66 76 67 77) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp2 + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=tmp3 + + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(60 70 61 71 62 72 63 73) + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(64 74 65 75 66 76 67 77) + + paddw xmm2, xmm4 ; xmm2=tmp4 + movdqa xmm5, xmm7 + movdqa xmm0, xmm1 + paddw xmm7, xmm4 ; xmm7=data2=(20 21 22 23 24 25 26 27) + paddw xmm1, xmm2 ; xmm1=data4=(40 41 42 43 44 45 46 47) + psubw xmm5, xmm4 ; xmm5=data5=(50 51 52 53 54 55 56 57) + psubw xmm0, xmm2 ; xmm0=data3=(30 31 32 33 34 35 36 37) + + movdqa xmm4, xmm7 ; transpose coefficients(phase 1) + punpcklwd xmm7, xmm0 ; xmm7=(20 30 21 31 22 32 23 33) + punpckhwd xmm4, xmm0 ; xmm4=(24 34 25 35 26 36 27 37) + movdqa xmm2, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm5 ; xmm1=(40 50 41 51 42 52 43 53) + punpckhwd xmm2, xmm5 ; xmm2=(44 54 45 55 46 56 47 57) + + movdqa xmm0, xmm3 ; transpose coefficients(phase 2) + punpckldq xmm3, xmm4 ; xmm3=(04 14 24 34 05 15 25 35) + punpckhdq xmm0, xmm4 ; xmm0=(06 16 26 36 07 17 27 37) + movdqa xmm5, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm7 ; xmm6=(00 10 20 30 01 11 21 31) + punpckhdq xmm5, xmm7 ; xmm5=(02 12 22 32 03 13 23 33) + + movdqa xmm4, XMMWORD [wk(0)] ; xmm4=(60 70 61 71 62 72 63 73) + movdqa xmm7, XMMWORD [wk(1)] ; xmm7=(64 74 65 75 66 76 67 77) + + movdqa XMMWORD [wk(0)], xmm3 ; wk(0)=(04 14 24 34 05 15 25 35) + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=(06 16 26 36 07 17 27 37) + + movdqa xmm3, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm4 ; xmm1=(40 50 60 70 41 51 61 71) + punpckhdq xmm3, xmm4 ; xmm3=(42 52 62 72 43 53 63 73) + movdqa xmm0, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm7 ; xmm2=(44 54 64 74 45 55 65 75) + punpckhdq xmm0, xmm7 ; xmm0=(46 56 66 76 47 57 67 77) + + movdqa xmm4, xmm6 ; transpose coefficients(phase 3) + punpcklqdq xmm6, xmm1 ; xmm6=col0=(00 10 20 30 40 50 60 70) + punpckhqdq xmm4, xmm1 ; xmm4=col1=(01 11 21 31 41 51 61 71) + movdqa xmm7, xmm5 ; transpose coefficients(phase 3) + punpcklqdq xmm5, xmm3 ; xmm5=col2=(02 12 22 32 42 52 62 72) + punpckhqdq xmm7, xmm3 ; xmm7=col3=(03 13 23 33 43 53 63 73) + + movdqa xmm1, XMMWORD [wk(0)] ; xmm1=(04 14 24 34 05 15 25 35) + movdqa xmm3, XMMWORD [wk(1)] ; xmm3=(06 16 26 36 07 17 27 37) + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=col1 + movdqa XMMWORD [wk(1)], xmm7 ; wk(1)=col3 + + movdqa xmm4, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm2 ; xmm1=col4=(04 14 24 34 44 54 64 74) + punpckhqdq xmm4, xmm2 ; xmm4=col5=(05 15 25 35 45 55 65 75) + movdqa xmm7, xmm3 ; transpose coefficients(phase 3) + punpcklqdq xmm3, xmm0 ; xmm3=col6=(06 16 26 36 46 56 66 76) + punpckhqdq xmm7, xmm0 ; xmm7=col7=(07 17 27 37 47 57 67 77) +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + mov rdi, r12 ; (JSAMPROW *) + mov eax, r13d + + ; -- Even part + + ; xmm6=col0, xmm5=col2, xmm1=col4, xmm3=col6 + + movdqa xmm2, xmm6 + movdqa xmm0, xmm5 + psubw xmm6, xmm1 ; xmm6=tmp11 + psubw xmm5, xmm3 + paddw xmm2, xmm1 ; xmm2=tmp10 + paddw xmm0, xmm3 ; xmm0=tmp13 + + psllw xmm5, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm5, [rel PW_F1414] + psubw xmm5, xmm0 ; xmm5=tmp12 + + movdqa xmm1, xmm2 + movdqa xmm3, xmm6 + psubw xmm2, xmm0 ; xmm2=tmp3 + psubw xmm6, xmm5 ; xmm6=tmp2 + paddw xmm1, xmm0 ; xmm1=tmp0 + paddw xmm3, xmm5 ; xmm3=tmp1 + + movdqa xmm0, XMMWORD [wk(0)] ; xmm0=col1 + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=col3 + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp3 + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=tmp2 + + ; -- Odd part + + ; xmm0=col1, xmm5=col3, xmm4=col5, xmm7=col7 + + movdqa xmm2, xmm0 + movdqa xmm6, xmm4 + psubw xmm0, xmm7 ; xmm0=z12 + psubw xmm4, xmm5 ; xmm4=z10 + paddw xmm2, xmm7 ; xmm2=z11 + paddw xmm6, xmm5 ; xmm6=z13 + + movdqa xmm7, xmm4 ; xmm7=z10(unscaled) + psllw xmm0, PRE_MULTIPLY_SCALE_BITS + psllw xmm4, PRE_MULTIPLY_SCALE_BITS + + movdqa xmm5, xmm2 + psubw xmm2, xmm6 + paddw xmm5, xmm6 ; xmm5=tmp7 + + psllw xmm2, PRE_MULTIPLY_SCALE_BITS + pmulhw xmm2, [rel PW_F1414] ; xmm2=tmp11 + + ; To avoid overflow... + ; + ; (Original) + ; tmp12 = -2.613125930 * z10 + z5; + ; + ; (This implementation) + ; tmp12 = (-1.613125930 - 1) * z10 + z5; + ; = -1.613125930 * z10 - z10 + z5; + + movdqa xmm6, xmm4 + paddw xmm4, xmm0 + pmulhw xmm4, [rel PW_F1847] ; xmm4=z5 + pmulhw xmm6, [rel PW_MF1613] + pmulhw xmm0, [rel PW_F1082] + psubw xmm6, xmm7 + psubw xmm0, xmm4 ; xmm0=tmp10 + paddw xmm6, xmm4 ; xmm6=tmp12 + + ; -- Final output stage + + psubw xmm6, xmm5 ; xmm6=tmp6 + movdqa xmm7, xmm1 + movdqa xmm4, xmm3 + paddw xmm1, xmm5 ; xmm1=data0=(00 10 20 30 40 50 60 70) + paddw xmm3, xmm6 ; xmm3=data1=(01 11 21 31 41 51 61 71) + psraw xmm1, (PASS1_BITS+3) ; descale + psraw xmm3, (PASS1_BITS+3) ; descale + psubw xmm7, xmm5 ; xmm7=data7=(07 17 27 37 47 57 67 77) + psubw xmm4, xmm6 ; xmm4=data6=(06 16 26 36 46 56 66 76) + psraw xmm7, (PASS1_BITS+3) ; descale + psraw xmm4, (PASS1_BITS+3) ; descale + psubw xmm2, xmm6 ; xmm2=tmp5 + + packsswb xmm1, xmm4 ; xmm1=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + packsswb xmm3, xmm7 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + movdqa xmm5, XMMWORD [wk(1)] ; xmm5=tmp2 + movdqa xmm6, XMMWORD [wk(0)] ; xmm6=tmp3 + + paddw xmm0, xmm2 ; xmm0=tmp4 + movdqa xmm4, xmm5 + movdqa xmm7, xmm6 + paddw xmm5, xmm2 ; xmm5=data2=(02 12 22 32 42 52 62 72) + paddw xmm6, xmm0 ; xmm6=data4=(04 14 24 34 44 54 64 74) + psraw xmm5, (PASS1_BITS+3) ; descale + psraw xmm6, (PASS1_BITS+3) ; descale + psubw xmm4, xmm2 ; xmm4=data5=(05 15 25 35 45 55 65 75) + psubw xmm7, xmm0 ; xmm7=data3=(03 13 23 33 43 53 63 73) + psraw xmm4, (PASS1_BITS+3) ; descale + psraw xmm7, (PASS1_BITS+3) ; descale + + movdqa xmm2, [rel PB_CENTERJSAMP] ; xmm2=[rel PB_CENTERJSAMP] + + packsswb xmm5, xmm6 ; xmm5=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74) + packsswb xmm7, xmm4 ; xmm7=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75) + + paddb xmm1, xmm2 + paddb xmm3, xmm2 + paddb xmm5, xmm2 + paddb xmm7, xmm2 + + movdqa xmm0, xmm1 ; transpose coefficients(phase 1) + punpcklbw xmm1, xmm3 ; xmm1=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71) + punpckhbw xmm0, xmm3 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77) + movdqa xmm6, xmm5 ; transpose coefficients(phase 1) + punpcklbw xmm5, xmm7 ; xmm5=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73) + punpckhbw xmm6, xmm7 ; xmm6=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75) + + movdqa xmm4, xmm1 ; transpose coefficients(phase 2) + punpcklwd xmm1, xmm5 ; xmm1=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) + punpckhwd xmm4, xmm5 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73) + movdqa xmm2, xmm6 ; transpose coefficients(phase 2) + punpcklwd xmm6, xmm0 ; xmm6=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) + punpckhwd xmm2, xmm0 ; xmm2=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77) + + movdqa xmm3, xmm1 ; transpose coefficients(phase 3) + punpckldq xmm1, xmm6 ; xmm1=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + punpckhdq xmm3, xmm6 ; xmm3=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + movdqa xmm7, xmm4 ; transpose coefficients(phase 3) + punpckldq xmm4, xmm2 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) + punpckhdq xmm7, xmm2 ; xmm7=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) + + pshufd xmm5, xmm1, 0x4E ; xmm5=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + pshufd xmm0, xmm3, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + pshufd xmm6, xmm4, 0x4E ; xmm6=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) + pshufd xmm2, xmm7, 0x4E ; xmm2=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) + + mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+2*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm1 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3 + mov rdxp, JSAMPROW [rdi+4*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+6*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm7 + + mov rdxp, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+3*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm5 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0 + mov rdxp, JSAMPROW [rdi+5*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+7*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm2 + + UNCOLLECT_ARGS 4 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-avx2.asm new file mode 100644 index 0000000..be3b468 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-avx2.asm @@ -0,0 +1,414 @@ +; +; jidctint.asm - accurate integer IDCT (64-bit AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2018, 2020, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; inverse DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jidctint.c; see the jidctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit inverse matrix transpose using AVX2 instructions +; %1-%4: Input/output registers +; %5-%8: Temp registers + +%macro DOTRANSPOSE 8 + ; %5=(00 10 20 30 40 50 60 70 01 11 21 31 41 51 61 71) + ; %6=(03 13 23 33 43 53 63 73 02 12 22 32 42 52 62 72) + ; %7=(04 14 24 34 44 54 64 74 05 15 25 35 45 55 65 75) + ; %8=(07 17 27 37 47 57 67 77 06 16 26 36 46 56 66 76) + + vpermq %5, %1, 0xD8 + vpermq %6, %2, 0x72 + vpermq %7, %3, 0xD8 + vpermq %8, %4, 0x72 + ; transpose coefficients(phase 1) + ; %5=(00 10 20 30 01 11 21 31 40 50 60 70 41 51 61 71) + ; %6=(02 12 22 32 03 13 23 33 42 52 62 72 43 53 63 73) + ; %7=(04 14 24 34 05 15 25 35 44 54 64 74 45 55 65 75) + ; %8=(06 16 26 36 07 17 27 37 46 56 66 76 47 57 67 77) + + vpunpcklwd %1, %5, %6 + vpunpckhwd %2, %5, %6 + vpunpcklwd %3, %7, %8 + vpunpckhwd %4, %7, %8 + ; transpose coefficients(phase 2) + ; %1=(00 02 10 12 20 22 30 32 40 42 50 52 60 62 70 72) + ; %2=(01 03 11 13 21 23 31 33 41 43 51 53 61 63 71 73) + ; %3=(04 06 14 16 24 26 34 36 44 46 54 56 64 66 74 76) + ; %4=(05 07 15 17 25 27 35 37 45 47 55 57 65 67 75 77) + + vpunpcklwd %5, %1, %2 + vpunpcklwd %6, %3, %4 + vpunpckhwd %7, %1, %2 + vpunpckhwd %8, %3, %4 + ; transpose coefficients(phase 3) + ; %5=(00 01 02 03 10 11 12 13 40 41 42 43 50 51 52 53) + ; %6=(04 05 06 07 14 15 16 17 44 45 46 47 54 55 56 57) + ; %7=(20 21 22 23 30 31 32 33 60 61 62 63 70 71 72 73) + ; %8=(24 25 26 27 34 35 36 37 64 65 66 67 74 75 76 77) + + vpunpcklqdq %1, %5, %6 + vpunpckhqdq %2, %5, %6 + vpunpcklqdq %3, %7, %8 + vpunpckhqdq %4, %7, %8 + ; transpose coefficients(phase 4) + ; %1=(00 01 02 03 04 05 06 07 40 41 42 43 44 45 46 47) + ; %2=(10 11 12 13 14 15 16 17 50 51 52 53 54 55 56 57) + ; %3=(20 21 22 23 24 25 26 27 60 61 62 63 64 65 66 67) + ; %4=(30 31 32 33 34 35 36 37 70 71 72 73 74 75 76 77) +%endmacro + +; -------------------------------------------------------------------------- +; In-place 8x8x16-bit accurate integer inverse DCT using AVX2 instructions +; %1-%4: Input/output registers +; %5-%12: Temp registers +; %9: Pass (1 or 2) + +%macro DODCT 13 + ; -- Even part + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + vperm2i128 %6, %3, %3, 0x01 ; %6=in6_2 + vpunpcklwd %5, %3, %6 ; %5=in26_62L + vpunpckhwd %6, %3, %6 ; %6=in26_62H + vpmaddwd %5, %5, [rel PW_F130_F054_MF130_F054] ; %5=tmp3_2L + vpmaddwd %6, %6, [rel PW_F130_F054_MF130_F054] ; %6=tmp3_2H + + vperm2i128 %7, %1, %1, 0x01 ; %7=in4_0 + vpsignw %1, %1, [rel PW_1_NEG1] + vpaddw %7, %7, %1 ; %7=(in0+in4)_(in0-in4) + + vpxor %1, %1, %1 + vpunpcklwd %8, %1, %7 ; %8=tmp0_1L + vpunpckhwd %1, %1, %7 ; %1=tmp0_1H + vpsrad %8, %8, (16-CONST_BITS) ; vpsrad %8,16 & vpslld %8,CONST_BITS + vpsrad %1, %1, (16-CONST_BITS) ; vpsrad %1,16 & vpslld %1,CONST_BITS + + vpsubd %11, %8, %5 ; %11=tmp0_1L-tmp3_2L=tmp13_12L + vpaddd %9, %8, %5 ; %9=tmp0_1L+tmp3_2L=tmp10_11L + vpsubd %12, %1, %6 ; %12=tmp0_1H-tmp3_2H=tmp13_12H + vpaddd %10, %1, %6 ; %10=tmp0_1H+tmp3_2H=tmp10_11H + + ; -- Odd part + + vpaddw %1, %4, %2 ; %1=in7_5+in3_1=z3_4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + vperm2i128 %8, %1, %1, 0x01 ; %8=z4_3 + vpunpcklwd %7, %1, %8 ; %7=z34_43L + vpunpckhwd %8, %1, %8 ; %8=z34_43H + vpmaddwd %7, %7, [rel PW_MF078_F117_F078_F117] ; %7=z3_4L + vpmaddwd %8, %8, [rel PW_MF078_F117_F078_F117] ; %8=z3_4H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + vperm2i128 %2, %2, %2, 0x01 ; %2=in1_3 + vpunpcklwd %3, %4, %2 ; %3=in71_53L + vpunpckhwd %4, %4, %2 ; %4=in71_53H + + vpmaddwd %5, %3, [rel PW_MF060_MF089_MF050_MF256] ; %5=tmp0_1L + vpmaddwd %6, %4, [rel PW_MF060_MF089_MF050_MF256] ; %6=tmp0_1H + vpaddd %5, %5, %7 ; %5=tmp0_1L+z3_4L=tmp0_1L + vpaddd %6, %6, %8 ; %6=tmp0_1H+z3_4H=tmp0_1H + + vpmaddwd %3, %3, [rel PW_MF089_F060_MF256_F050] ; %3=tmp3_2L + vpmaddwd %4, %4, [rel PW_MF089_F060_MF256_F050] ; %4=tmp3_2H + vperm2i128 %7, %7, %7, 0x01 ; %7=z4_3L + vperm2i128 %8, %8, %8, 0x01 ; %8=z4_3H + vpaddd %7, %3, %7 ; %7=tmp3_2L+z4_3L=tmp3_2L + vpaddd %8, %4, %8 ; %8=tmp3_2H+z4_3H=tmp3_2H + + ; -- Final output stage + + vpaddd %1, %9, %7 ; %1=tmp10_11L+tmp3_2L=data0_1L + vpaddd %2, %10, %8 ; %2=tmp10_11H+tmp3_2H=data0_1H + vpaddd %1, %1, [rel PD_DESCALE_P %+ %13] + vpaddd %2, %2, [rel PD_DESCALE_P %+ %13] + vpsrad %1, %1, DESCALE_P %+ %13 + vpsrad %2, %2, DESCALE_P %+ %13 + vpackssdw %1, %1, %2 ; %1=data0_1 + + vpsubd %3, %9, %7 ; %3=tmp10_11L-tmp3_2L=data7_6L + vpsubd %4, %10, %8 ; %4=tmp10_11H-tmp3_2H=data7_6H + vpaddd %3, %3, [rel PD_DESCALE_P %+ %13] + vpaddd %4, %4, [rel PD_DESCALE_P %+ %13] + vpsrad %3, %3, DESCALE_P %+ %13 + vpsrad %4, %4, DESCALE_P %+ %13 + vpackssdw %4, %3, %4 ; %4=data7_6 + + vpaddd %7, %11, %5 ; %7=tmp13_12L+tmp0_1L=data3_2L + vpaddd %8, %12, %6 ; %8=tmp13_12H+tmp0_1H=data3_2H + vpaddd %7, %7, [rel PD_DESCALE_P %+ %13] + vpaddd %8, %8, [rel PD_DESCALE_P %+ %13] + vpsrad %7, %7, DESCALE_P %+ %13 + vpsrad %8, %8, DESCALE_P %+ %13 + vpackssdw %2, %7, %8 ; %2=data3_2 + + vpsubd %7, %11, %5 ; %7=tmp13_12L-tmp0_1L=data4_5L + vpsubd %8, %12, %6 ; %8=tmp13_12H-tmp0_1H=data4_5H + vpaddd %7, %7, [rel PD_DESCALE_P %+ %13] + vpaddd %8, %8, [rel PD_DESCALE_P %+ %13] + vpsrad %7, %7, DESCALE_P %+ %13 + vpsrad %8, %8, DESCALE_P %+ %13 + vpackssdw %3, %7, %8 ; %3=data4_5 +%endmacro + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_islow_avx2) + +EXTN(jconst_idct_islow_avx2): + +PW_F130_F054_MF130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 + times 4 dw (F_0_541 - F_1_847), F_0_541 +PW_MF078_F117_F078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 + times 4 dw (F_1_175 - F_0_390), F_1_175 +PW_MF060_MF089_MF050_MF256 times 4 dw (F_0_298 - F_0_899), -F_0_899 + times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF089_F060_MF256_F050 times 4 dw -F_0_899, (F_1_501 - F_0_899) + times 4 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 8 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 8 dd 1 << (DESCALE_P2 - 1) +PB_CENTERJSAMP times 32 db CENTERJSAMPLE +PW_1_NEG1 times 8 dw 1 + times 8 dw -1 + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_islow_avx2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +; r10 = jpeg_component_info *compptr +; r11 = JCOEFPTR coef_block +; r12 = JSAMPARRAY output_buf +; r13d = JDIMENSION output_col + + align 32 + GLOBAL_FUNCTION(jsimd_idct_islow_avx2) + +EXTN(jsimd_idct_islow_avx2): + ENDBR64 + push rbp + mov rbp, rsp ; rbp = aligned rbp + PUSH_XMM 4 + COLLECT_ARGS 4 + + ; ---- Pass 1: process columns. + +%ifndef NO_ZERO_COLUMN_TEST_ISLOW_AVX2 + mov eax, dword [DWBLOCK(1,0,r11,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,r11,SIZEOF_JCOEF)] + jnz near .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,r11,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,r11,SIZEOF_JCOEF)] + vpor xmm0, xmm0, XMMWORD [XMMBLOCK(3,0,r11,SIZEOF_JCOEF)] + vpor xmm1, xmm1, XMMWORD [XMMBLOCK(4,0,r11,SIZEOF_JCOEF)] + vpor xmm0, xmm0, XMMWORD [XMMBLOCK(5,0,r11,SIZEOF_JCOEF)] + vpor xmm1, xmm1, XMMWORD [XMMBLOCK(6,0,r11,SIZEOF_JCOEF)] + vpor xmm0, xmm0, XMMWORD [XMMBLOCK(7,0,r11,SIZEOF_JCOEF)] + vpor xmm1, xmm1, xmm0 + vpacksswb xmm1, xmm1, xmm1 + vpacksswb xmm1, xmm1, xmm1 + movd eax, xmm1 + test rax, rax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm5, XMMWORD [XMMBLOCK(0,0,r11,SIZEOF_JCOEF)] + vpmullw xmm5, xmm5, XMMWORD [XMMBLOCK(0,0,r10,SIZEOF_ISLOW_MULT_TYPE)] + + vpsllw xmm5, xmm5, PASS1_BITS + + vpunpcklwd xmm4, xmm5, xmm5 ; xmm4=(00 00 01 01 02 02 03 03) + vpunpckhwd xmm5, xmm5, xmm5 ; xmm5=(04 04 05 05 06 06 07 07) + vinserti128 ymm4, ymm4, xmm5, 1 + + vpshufd ymm0, ymm4, 0x00 ; ymm0=col0_4=(00 00 00 00 00 00 00 00 04 04 04 04 04 04 04 04) + vpshufd ymm1, ymm4, 0x55 ; ymm1=col1_5=(01 01 01 01 01 01 01 01 05 05 05 05 05 05 05 05) + vpshufd ymm2, ymm4, 0xAA ; ymm2=col2_6=(02 02 02 02 02 02 02 02 06 06 06 06 06 06 06 06) + vpshufd ymm3, ymm4, 0xFF ; ymm3=col3_7=(03 03 03 03 03 03 03 03 07 07 07 07 07 07 07 07) + + jmp near .column_end +%endif +.columnDCT: + + vmovdqu ymm4, YMMWORD [YMMBLOCK(0,0,r11,SIZEOF_JCOEF)] ; ymm4=in0_1 + vmovdqu ymm5, YMMWORD [YMMBLOCK(2,0,r11,SIZEOF_JCOEF)] ; ymm5=in2_3 + vmovdqu ymm6, YMMWORD [YMMBLOCK(4,0,r11,SIZEOF_JCOEF)] ; ymm6=in4_5 + vmovdqu ymm7, YMMWORD [YMMBLOCK(6,0,r11,SIZEOF_JCOEF)] ; ymm7=in6_7 + vpmullw ymm4, ymm4, YMMWORD [YMMBLOCK(0,0,r10,SIZEOF_ISLOW_MULT_TYPE)] + vpmullw ymm5, ymm5, YMMWORD [YMMBLOCK(2,0,r10,SIZEOF_ISLOW_MULT_TYPE)] + vpmullw ymm6, ymm6, YMMWORD [YMMBLOCK(4,0,r10,SIZEOF_ISLOW_MULT_TYPE)] + vpmullw ymm7, ymm7, YMMWORD [YMMBLOCK(6,0,r10,SIZEOF_ISLOW_MULT_TYPE)] + + vperm2i128 ymm0, ymm4, ymm6, 0x20 ; ymm0=in0_4 + vperm2i128 ymm1, ymm5, ymm4, 0x31 ; ymm1=in3_1 + vperm2i128 ymm2, ymm5, ymm7, 0x20 ; ymm2=in2_6 + vperm2i128 ymm3, ymm7, ymm6, 0x31 ; ymm3=in7_5 + + DODCT ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7, ymm8, ymm9, ymm10, ymm11, 1 + ; ymm0=data0_1, ymm1=data3_2, ymm2=data4_5, ymm3=data7_6 + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7 + ; ymm0=data0_4, ymm1=data1_5, ymm2=data2_6, ymm3=data3_7 + +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [r11 + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [r11 + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [r11 + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [r11 + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows. + + vperm2i128 ymm4, ymm3, ymm1, 0x31 ; ymm3=in7_5 + vperm2i128 ymm1, ymm3, ymm1, 0x20 ; ymm1=in3_1 + + DODCT ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7, ymm8, ymm9, ymm10, ymm11, 2 + ; ymm0=data0_1, ymm1=data3_2, ymm2=data4_5, ymm4=data7_6 + + DOTRANSPOSE ymm0, ymm1, ymm2, ymm4, ymm3, ymm5, ymm6, ymm7 + ; ymm0=data0_4, ymm1=data1_5, ymm2=data2_6, ymm4=data3_7 + + vpacksswb ymm0, ymm0, ymm1 ; ymm0=data01_45 + vpacksswb ymm1, ymm2, ymm4 ; ymm1=data23_67 + vpaddb ymm0, ymm0, [rel PB_CENTERJSAMP] + vpaddb ymm1, ymm1, [rel PB_CENTERJSAMP] + + vextracti128 xmm6, ymm1, 1 ; xmm3=data67 + vextracti128 xmm4, ymm0, 1 ; xmm2=data45 + vextracti128 xmm2, ymm1, 0 ; xmm1=data23 + vextracti128 xmm0, ymm0, 0 ; xmm0=data01 + + vpshufd xmm1, xmm0, 0x4E ; xmm1=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + vpshufd xmm3, xmm2, 0x4E ; xmm3=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + vpshufd xmm5, xmm4, 0x4E ; xmm5=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) + vpshufd xmm7, xmm6, 0x4E ; xmm7=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) + + vzeroupper + + mov eax, r13d + + mov rdxp, JSAMPROW [r12+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rsip, JSAMPROW [r12+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm0 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm1 + + mov rdxp, JSAMPROW [r12+2*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rsip, JSAMPROW [r12+3*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm2 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3 + + mov rdxp, JSAMPROW [r12+4*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rsip, JSAMPROW [r12+5*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm5 + + mov rdxp, JSAMPROW [r12+6*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rsip, JSAMPROW [r12+7*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm7 + + UNCOLLECT_ARGS 4 + POP_XMM 4 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-sse2.asm new file mode 100644 index 0000000..b186871 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctint-sse2.asm @@ -0,0 +1,843 @@ +; +; jidctint.asm - accurate integer IDCT (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2020, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains a slower but more accurate integer implementation of the +; inverse DCT (Discrete Cosine Transform). The following code is based +; directly on the IJG's original jidctint.c; see the jidctint.c for +; more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1 (CONST_BITS - PASS1_BITS) +%define DESCALE_P2 (CONST_BITS + PASS1_BITS + 3) + +%if CONST_BITS == 13 +F_0_298 equ 2446 ; FIX(0.298631336) +F_0_390 equ 3196 ; FIX(0.390180644) +F_0_541 equ 4433 ; FIX(0.541196100) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_175 equ 9633 ; FIX(1.175875602) +F_1_501 equ 12299 ; FIX(1.501321110) +F_1_847 equ 15137 ; FIX(1.847759065) +F_1_961 equ 16069 ; FIX(1.961570560) +F_2_053 equ 16819 ; FIX(2.053119869) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_072 equ 25172 ; FIX(3.072711026) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_298 equ DESCALE( 320652955, 30 - CONST_BITS) ; FIX(0.298631336) +F_0_390 equ DESCALE( 418953276, 30 - CONST_BITS) ; FIX(0.390180644) +F_0_541 equ DESCALE( 581104887, 30 - CONST_BITS) ; FIX(0.541196100) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_175 equ DESCALE(1262586813, 30 - CONST_BITS) ; FIX(1.175875602) +F_1_501 equ DESCALE(1612031267, 30 - CONST_BITS) ; FIX(1.501321110) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_1_961 equ DESCALE(2106220350, 30 - CONST_BITS) ; FIX(1.961570560) +F_2_053 equ DESCALE(2204520673, 30 - CONST_BITS) ; FIX(2.053119869) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_072 equ DESCALE(3299298341, 30 - CONST_BITS) ; FIX(3.072711026) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_islow_sse2) + +EXTN(jconst_idct_islow_sse2): + +PW_F130_F054 times 4 dw (F_0_541 + F_0_765), F_0_541 +PW_F054_MF130 times 4 dw F_0_541, (F_0_541 - F_1_847) +PW_MF078_F117 times 4 dw (F_1_175 - F_1_961), F_1_175 +PW_F117_F078 times 4 dw F_1_175, (F_1_175 - F_0_390) +PW_MF060_MF089 times 4 dw (F_0_298 - F_0_899), -F_0_899 +PW_MF089_F060 times 4 dw -F_0_899, (F_1_501 - F_0_899) +PW_MF050_MF256 times 4 dw (F_2_053 - F_2_562), -F_2_562 +PW_MF256_F050 times 4 dw -F_2_562, (F_3_072 - F_2_562) +PD_DESCALE_P1 times 4 dd 1 << (DESCALE_P1 - 1) +PD_DESCALE_P2 times 4 dd 1 << (DESCALE_P2 - 1) +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform dequantization and inverse DCT on one block of coefficients. +; +; GLOBAL(void) +; jsimd_idct_islow_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +; r10 = jpeg_component_info *compptr +; r11 = JCOEFPTR coef_block +; r12 = JSAMPARRAY output_buf +; r13d = JDIMENSION output_col + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 12 + + align 32 + GLOBAL_FUNCTION(jsimd_idct_islow_sse2) + +EXTN(jsimd_idct_islow_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 4 + + ; ---- Pass 1: process columns from input. + + mov rdx, r10 ; quantptr + mov rsi, r11 ; inptr + +%ifndef NO_ZERO_COLUMN_TEST_ISLOW_SSE2 + mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] + jnz near .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + por xmm1, xmm0 + packsswb xmm1, xmm1 + packsswb xmm1, xmm1 + movd eax, xmm1 + test rax, rax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm5, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + pmullw xmm5, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + psllw xmm5, PASS1_BITS + + movdqa xmm4, xmm5 ; xmm5=in0=(00 01 02 03 04 05 06 07) + punpcklwd xmm5, xmm5 ; xmm5=(00 00 01 01 02 02 03 03) + punpckhwd xmm4, xmm4 ; xmm4=(04 04 05 05 06 06 07 07) + + pshufd xmm7, xmm5, 0x00 ; xmm7=col0=(00 00 00 00 00 00 00 00) + pshufd xmm6, xmm5, 0x55 ; xmm6=col1=(01 01 01 01 01 01 01 01) + pshufd xmm1, xmm5, 0xAA ; xmm1=col2=(02 02 02 02 02 02 02 02) + pshufd xmm5, xmm5, 0xFF ; xmm5=col3=(03 03 03 03 03 03 03 03) + pshufd xmm0, xmm4, 0x00 ; xmm0=col4=(04 04 04 04 04 04 04 04) + pshufd xmm3, xmm4, 0x55 ; xmm3=col5=(05 05 05 05 05 05 05 05) + pshufd xmm2, xmm4, 0xAA ; xmm2=col6=(06 06 06 06 06 06 06 06) + pshufd xmm4, xmm4, 0xFF ; xmm4=col7=(07 07 07 07 07 07 07 07) + + movdqa XMMWORD [wk(8)], xmm6 ; wk(8)=col1 + movdqa XMMWORD [wk(9)], xmm5 ; wk(9)=col3 + movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 + movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 + jmp near .column_end +%endif +.columnDCT: + + ; -- Even part + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(4,0,rsi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(4,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + movdqa xmm4, xmm1 ; xmm1=in2=z2 + movdqa xmm5, xmm1 + punpcklwd xmm4, xmm3 ; xmm3=in6=z3 + punpckhwd xmm5, xmm3 + movdqa xmm1, xmm4 + movdqa xmm3, xmm5 + pmaddwd xmm4, [rel PW_F130_F054] ; xmm4=tmp3L + pmaddwd xmm5, [rel PW_F130_F054] ; xmm5=tmp3H + pmaddwd xmm1, [rel PW_F054_MF130] ; xmm1=tmp2L + pmaddwd xmm3, [rel PW_F054_MF130] ; xmm3=tmp2H + + movdqa xmm6, xmm0 + paddw xmm0, xmm2 ; xmm0=in0+in4 + psubw xmm6, xmm2 ; xmm6=in0-in4 + + pxor xmm7, xmm7 + pxor xmm2, xmm2 + punpcklwd xmm7, xmm0 ; xmm7=tmp0L + punpckhwd xmm2, xmm0 ; xmm2=tmp0H + psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS + psrad xmm2, (16-CONST_BITS) ; psrad xmm2,16 & pslld xmm2,CONST_BITS + + movdqa xmm0, xmm7 + paddd xmm7, xmm4 ; xmm7=tmp10L + psubd xmm0, xmm4 ; xmm0=tmp13L + movdqa xmm4, xmm2 + paddd xmm2, xmm5 ; xmm2=tmp10H + psubd xmm4, xmm5 ; xmm4=tmp13H + + movdqa XMMWORD [wk(0)], xmm7 ; wk(0)=tmp10L + movdqa XMMWORD [wk(1)], xmm2 ; wk(1)=tmp10H + movdqa XMMWORD [wk(2)], xmm0 ; wk(2)=tmp13L + movdqa XMMWORD [wk(3)], xmm4 ; wk(3)=tmp13H + + pxor xmm5, xmm5 + pxor xmm7, xmm7 + punpcklwd xmm5, xmm6 ; xmm5=tmp1L + punpckhwd xmm7, xmm6 ; xmm7=tmp1H + psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS + psrad xmm7, (16-CONST_BITS) ; psrad xmm7,16 & pslld xmm7,CONST_BITS + + movdqa xmm2, xmm5 + paddd xmm5, xmm1 ; xmm5=tmp11L + psubd xmm2, xmm1 ; xmm2=tmp12L + movdqa xmm0, xmm7 + paddd xmm7, xmm3 ; xmm7=tmp11H + psubd xmm0, xmm3 ; xmm0=tmp12H + + movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L + movdqa XMMWORD [wk(5)], xmm7 ; wk(5)=tmp11H + movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=tmp12L + movdqa XMMWORD [wk(7)], xmm0 ; wk(7)=tmp12H + + ; -- Odd part + + movdqa xmm4, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movdqa xmm6, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + pmullw xmm4, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm6, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + pmullw xmm1, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + movdqa xmm5, xmm6 + movdqa xmm7, xmm4 + paddw xmm5, xmm3 ; xmm5=z3 + paddw xmm7, xmm1 ; xmm7=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm2, xmm5 + movdqa xmm0, xmm5 + punpcklwd xmm2, xmm7 + punpckhwd xmm0, xmm7 + movdqa xmm5, xmm2 + movdqa xmm7, xmm0 + pmaddwd xmm2, [rel PW_MF078_F117] ; xmm2=z3L + pmaddwd xmm0, [rel PW_MF078_F117] ; xmm0=z3H + pmaddwd xmm5, [rel PW_F117_F078] ; xmm5=z4L + pmaddwd xmm7, [rel PW_F117_F078] ; xmm7=z4H + + movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=z3L + movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=z3H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + movdqa xmm2, xmm3 + movdqa xmm0, xmm3 + punpcklwd xmm2, xmm4 + punpckhwd xmm0, xmm4 + movdqa xmm3, xmm2 + movdqa xmm4, xmm0 + pmaddwd xmm2, [rel PW_MF060_MF089] ; xmm2=tmp0L + pmaddwd xmm0, [rel PW_MF060_MF089] ; xmm0=tmp0H + pmaddwd xmm3, [rel PW_MF089_F060] ; xmm3=tmp3L + pmaddwd xmm4, [rel PW_MF089_F060] ; xmm4=tmp3H + + paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp0L + paddd xmm0, XMMWORD [wk(11)] ; xmm0=tmp0H + paddd xmm3, xmm5 ; xmm3=tmp3L + paddd xmm4, xmm7 ; xmm4=tmp3H + + movdqa XMMWORD [wk(8)], xmm2 ; wk(8)=tmp0L + movdqa XMMWORD [wk(9)], xmm0 ; wk(9)=tmp0H + + movdqa xmm2, xmm1 + movdqa xmm0, xmm1 + punpcklwd xmm2, xmm6 + punpckhwd xmm0, xmm6 + movdqa xmm1, xmm2 + movdqa xmm6, xmm0 + pmaddwd xmm2, [rel PW_MF050_MF256] ; xmm2=tmp1L + pmaddwd xmm0, [rel PW_MF050_MF256] ; xmm0=tmp1H + pmaddwd xmm1, [rel PW_MF256_F050] ; xmm1=tmp2L + pmaddwd xmm6, [rel PW_MF256_F050] ; xmm6=tmp2H + + paddd xmm2, xmm5 ; xmm2=tmp1L + paddd xmm0, xmm7 ; xmm0=tmp1H + paddd xmm1, XMMWORD [wk(10)] ; xmm1=tmp2L + paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H + + movdqa XMMWORD [wk(10)], xmm2 ; wk(10)=tmp1L + movdqa XMMWORD [wk(11)], xmm0 ; wk(11)=tmp1H + + ; -- Final output stage + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L + movdqa xmm7, XMMWORD [wk(1)] ; xmm7=tmp10H + + movdqa xmm2, xmm5 + movdqa xmm0, xmm7 + paddd xmm5, xmm3 ; xmm5=data0L + paddd xmm7, xmm4 ; xmm7=data0H + psubd xmm2, xmm3 ; xmm2=data7L + psubd xmm0, xmm4 ; xmm0=data7H + + movdqa xmm3, [rel PD_DESCALE_P1] ; xmm3=[rel PD_DESCALE_P1] + + paddd xmm5, xmm3 + paddd xmm7, xmm3 + psrad xmm5, DESCALE_P1 + psrad xmm7, DESCALE_P1 + paddd xmm2, xmm3 + paddd xmm0, xmm3 + psrad xmm2, DESCALE_P1 + psrad xmm0, DESCALE_P1 + + packssdw xmm5, xmm7 ; xmm5=data0=(00 01 02 03 04 05 06 07) + packssdw xmm2, xmm0 ; xmm2=data7=(70 71 72 73 74 75 76 77) + + movdqa xmm4, XMMWORD [wk(4)] ; xmm4=tmp11L + movdqa xmm3, XMMWORD [wk(5)] ; xmm3=tmp11H + + movdqa xmm7, xmm4 + movdqa xmm0, xmm3 + paddd xmm4, xmm1 ; xmm4=data1L + paddd xmm3, xmm6 ; xmm3=data1H + psubd xmm7, xmm1 ; xmm7=data6L + psubd xmm0, xmm6 ; xmm0=data6H + + movdqa xmm1, [rel PD_DESCALE_P1] ; xmm1=[rel PD_DESCALE_P1] + + paddd xmm4, xmm1 + paddd xmm3, xmm1 + psrad xmm4, DESCALE_P1 + psrad xmm3, DESCALE_P1 + paddd xmm7, xmm1 + paddd xmm0, xmm1 + psrad xmm7, DESCALE_P1 + psrad xmm0, DESCALE_P1 + + packssdw xmm4, xmm3 ; xmm4=data1=(10 11 12 13 14 15 16 17) + packssdw xmm7, xmm0 ; xmm7=data6=(60 61 62 63 64 65 66 67) + + movdqa xmm6, xmm5 ; transpose coefficients(phase 1) + punpcklwd xmm5, xmm4 ; xmm5=(00 10 01 11 02 12 03 13) + punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17) + movdqa xmm1, xmm7 ; transpose coefficients(phase 1) + punpcklwd xmm7, xmm2 ; xmm7=(60 70 61 71 62 72 63 73) + punpckhwd xmm1, xmm2 ; xmm1=(64 74 65 75 66 76 67 77) + + movdqa xmm3, XMMWORD [wk(6)] ; xmm3=tmp12L + movdqa xmm0, XMMWORD [wk(7)] ; xmm0=tmp12H + movdqa xmm4, XMMWORD [wk(10)] ; xmm4=tmp1L + movdqa xmm2, XMMWORD [wk(11)] ; xmm2=tmp1H + + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 01 11 02 12 03 13) + movdqa XMMWORD [wk(1)], xmm6 ; wk(1)=(04 14 05 15 06 16 07 17) + movdqa XMMWORD [wk(4)], xmm7 ; wk(4)=(60 70 61 71 62 72 63 73) + movdqa XMMWORD [wk(5)], xmm1 ; wk(5)=(64 74 65 75 66 76 67 77) + + movdqa xmm5, xmm3 + movdqa xmm6, xmm0 + paddd xmm3, xmm4 ; xmm3=data2L + paddd xmm0, xmm2 ; xmm0=data2H + psubd xmm5, xmm4 ; xmm5=data5L + psubd xmm6, xmm2 ; xmm6=data5H + + movdqa xmm7, [rel PD_DESCALE_P1] ; xmm7=[rel PD_DESCALE_P1] + + paddd xmm3, xmm7 + paddd xmm0, xmm7 + psrad xmm3, DESCALE_P1 + psrad xmm0, DESCALE_P1 + paddd xmm5, xmm7 + paddd xmm6, xmm7 + psrad xmm5, DESCALE_P1 + psrad xmm6, DESCALE_P1 + + packssdw xmm3, xmm0 ; xmm3=data2=(20 21 22 23 24 25 26 27) + packssdw xmm5, xmm6 ; xmm5=data5=(50 51 52 53 54 55 56 57) + + movdqa xmm1, XMMWORD [wk(2)] ; xmm1=tmp13L + movdqa xmm4, XMMWORD [wk(3)] ; xmm4=tmp13H + movdqa xmm2, XMMWORD [wk(8)] ; xmm2=tmp0L + movdqa xmm7, XMMWORD [wk(9)] ; xmm7=tmp0H + + movdqa xmm0, xmm1 + movdqa xmm6, xmm4 + paddd xmm1, xmm2 ; xmm1=data3L + paddd xmm4, xmm7 ; xmm4=data3H + psubd xmm0, xmm2 ; xmm0=data4L + psubd xmm6, xmm7 ; xmm6=data4H + + movdqa xmm2, [rel PD_DESCALE_P1] ; xmm2=[rel PD_DESCALE_P1] + + paddd xmm1, xmm2 + paddd xmm4, xmm2 + psrad xmm1, DESCALE_P1 + psrad xmm4, DESCALE_P1 + paddd xmm0, xmm2 + paddd xmm6, xmm2 + psrad xmm0, DESCALE_P1 + psrad xmm6, DESCALE_P1 + + packssdw xmm1, xmm4 ; xmm1=data3=(30 31 32 33 34 35 36 37) + packssdw xmm0, xmm6 ; xmm0=data4=(40 41 42 43 44 45 46 47) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 01 11 02 12 03 13) + movdqa xmm2, XMMWORD [wk(1)] ; xmm2=(04 14 05 15 06 16 07 17) + + movdqa xmm4, xmm3 ; transpose coefficients(phase 1) + punpcklwd xmm3, xmm1 ; xmm3=(20 30 21 31 22 32 23 33) + punpckhwd xmm4, xmm1 ; xmm4=(24 34 25 35 26 36 27 37) + movdqa xmm6, xmm0 ; transpose coefficients(phase 1) + punpcklwd xmm0, xmm5 ; xmm0=(40 50 41 51 42 52 43 53) + punpckhwd xmm6, xmm5 ; xmm6=(44 54 45 55 46 56 47 57) + + movdqa xmm1, xmm7 ; transpose coefficients(phase 2) + punpckldq xmm7, xmm3 ; xmm7=(00 10 20 30 01 11 21 31) + punpckhdq xmm1, xmm3 ; xmm1=(02 12 22 32 03 13 23 33) + movdqa xmm5, xmm2 ; transpose coefficients(phase 2) + punpckldq xmm2, xmm4 ; xmm2=(04 14 24 34 05 15 25 35) + punpckhdq xmm5, xmm4 ; xmm5=(06 16 26 36 07 17 27 37) + + movdqa xmm3, XMMWORD [wk(4)] ; xmm3=(60 70 61 71 62 72 63 73) + movdqa xmm4, XMMWORD [wk(5)] ; xmm4=(64 74 65 75 66 76 67 77) + + movdqa XMMWORD [wk(6)], xmm2 ; wk(6)=(04 14 24 34 05 15 25 35) + movdqa XMMWORD [wk(7)], xmm5 ; wk(7)=(06 16 26 36 07 17 27 37) + + movdqa xmm2, xmm0 ; transpose coefficients(phase 2) + punpckldq xmm0, xmm3 ; xmm0=(40 50 60 70 41 51 61 71) + punpckhdq xmm2, xmm3 ; xmm2=(42 52 62 72 43 53 63 73) + movdqa xmm5, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm4 ; xmm6=(44 54 64 74 45 55 65 75) + punpckhdq xmm5, xmm4 ; xmm5=(46 56 66 76 47 57 67 77) + + movdqa xmm3, xmm7 ; transpose coefficients(phase 3) + punpcklqdq xmm7, xmm0 ; xmm7=col0=(00 10 20 30 40 50 60 70) + punpckhqdq xmm3, xmm0 ; xmm3=col1=(01 11 21 31 41 51 61 71) + movdqa xmm4, xmm1 ; transpose coefficients(phase 3) + punpcklqdq xmm1, xmm2 ; xmm1=col2=(02 12 22 32 42 52 62 72) + punpckhqdq xmm4, xmm2 ; xmm4=col3=(03 13 23 33 43 53 63 73) + + movdqa xmm0, XMMWORD [wk(6)] ; xmm0=(04 14 24 34 05 15 25 35) + movdqa xmm2, XMMWORD [wk(7)] ; xmm2=(06 16 26 36 07 17 27 37) + + movdqa XMMWORD [wk(8)], xmm3 ; wk(8)=col1 + movdqa XMMWORD [wk(9)], xmm4 ; wk(9)=col3 + + movdqa xmm3, xmm0 ; transpose coefficients(phase 3) + punpcklqdq xmm0, xmm6 ; xmm0=col4=(04 14 24 34 44 54 64 74) + punpckhqdq xmm3, xmm6 ; xmm3=col5=(05 15 25 35 45 55 65 75) + movdqa xmm4, xmm2 ; transpose coefficients(phase 3) + punpcklqdq xmm2, xmm5 ; xmm2=col6=(06 16 26 36 46 56 66 76) + punpckhqdq xmm4, xmm5 ; xmm4=col7=(07 17 27 37 47 57 67 77) + + movdqa XMMWORD [wk(10)], xmm3 ; wk(10)=col5 + movdqa XMMWORD [wk(11)], xmm4 ; wk(11)=col7 +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows from work array, store into output array. + + mov rdi, r12 ; (JSAMPROW *) + mov eax, r13d + + ; -- Even part + + ; xmm7=col0, xmm1=col2, xmm0=col4, xmm2=col6 + + ; (Original) + ; z1 = (z2 + z3) * 0.541196100; + ; tmp2 = z1 + z3 * -1.847759065; + ; tmp3 = z1 + z2 * 0.765366865; + ; + ; (This implementation) + ; tmp2 = z2 * 0.541196100 + z3 * (0.541196100 - 1.847759065); + ; tmp3 = z2 * (0.541196100 + 0.765366865) + z3 * 0.541196100; + + movdqa xmm6, xmm1 ; xmm1=in2=z2 + movdqa xmm5, xmm1 + punpcklwd xmm6, xmm2 ; xmm2=in6=z3 + punpckhwd xmm5, xmm2 + movdqa xmm1, xmm6 + movdqa xmm2, xmm5 + pmaddwd xmm6, [rel PW_F130_F054] ; xmm6=tmp3L + pmaddwd xmm5, [rel PW_F130_F054] ; xmm5=tmp3H + pmaddwd xmm1, [rel PW_F054_MF130] ; xmm1=tmp2L + pmaddwd xmm2, [rel PW_F054_MF130] ; xmm2=tmp2H + + movdqa xmm3, xmm7 + paddw xmm7, xmm0 ; xmm7=in0+in4 + psubw xmm3, xmm0 ; xmm3=in0-in4 + + pxor xmm4, xmm4 + pxor xmm0, xmm0 + punpcklwd xmm4, xmm7 ; xmm4=tmp0L + punpckhwd xmm0, xmm7 ; xmm0=tmp0H + psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS + psrad xmm0, (16-CONST_BITS) ; psrad xmm0,16 & pslld xmm0,CONST_BITS + + movdqa xmm7, xmm4 + paddd xmm4, xmm6 ; xmm4=tmp10L + psubd xmm7, xmm6 ; xmm7=tmp13L + movdqa xmm6, xmm0 + paddd xmm0, xmm5 ; xmm0=tmp10H + psubd xmm6, xmm5 ; xmm6=tmp13H + + movdqa XMMWORD [wk(0)], xmm4 ; wk(0)=tmp10L + movdqa XMMWORD [wk(1)], xmm0 ; wk(1)=tmp10H + movdqa XMMWORD [wk(2)], xmm7 ; wk(2)=tmp13L + movdqa XMMWORD [wk(3)], xmm6 ; wk(3)=tmp13H + + pxor xmm5, xmm5 + pxor xmm4, xmm4 + punpcklwd xmm5, xmm3 ; xmm5=tmp1L + punpckhwd xmm4, xmm3 ; xmm4=tmp1H + psrad xmm5, (16-CONST_BITS) ; psrad xmm5,16 & pslld xmm5,CONST_BITS + psrad xmm4, (16-CONST_BITS) ; psrad xmm4,16 & pslld xmm4,CONST_BITS + + movdqa xmm0, xmm5 + paddd xmm5, xmm1 ; xmm5=tmp11L + psubd xmm0, xmm1 ; xmm0=tmp12L + movdqa xmm7, xmm4 + paddd xmm4, xmm2 ; xmm4=tmp11H + psubd xmm7, xmm2 ; xmm7=tmp12H + + movdqa XMMWORD [wk(4)], xmm5 ; wk(4)=tmp11L + movdqa XMMWORD [wk(5)], xmm4 ; wk(5)=tmp11H + movdqa XMMWORD [wk(6)], xmm0 ; wk(6)=tmp12L + movdqa XMMWORD [wk(7)], xmm7 ; wk(7)=tmp12H + + ; -- Odd part + + movdqa xmm6, XMMWORD [wk(9)] ; xmm6=col3 + movdqa xmm3, XMMWORD [wk(8)] ; xmm3=col1 + movdqa xmm1, XMMWORD [wk(11)] ; xmm1=col7 + movdqa xmm2, XMMWORD [wk(10)] ; xmm2=col5 + + movdqa xmm5, xmm6 + movdqa xmm4, xmm3 + paddw xmm5, xmm1 ; xmm5=z3 + paddw xmm4, xmm2 ; xmm4=z4 + + ; (Original) + ; z5 = (z3 + z4) * 1.175875602; + ; z3 = z3 * -1.961570560; z4 = z4 * -0.390180644; + ; z3 += z5; z4 += z5; + ; + ; (This implementation) + ; z3 = z3 * (1.175875602 - 1.961570560) + z4 * 1.175875602; + ; z4 = z3 * 1.175875602 + z4 * (1.175875602 - 0.390180644); + + movdqa xmm0, xmm5 + movdqa xmm7, xmm5 + punpcklwd xmm0, xmm4 + punpckhwd xmm7, xmm4 + movdqa xmm5, xmm0 + movdqa xmm4, xmm7 + pmaddwd xmm0, [rel PW_MF078_F117] ; xmm0=z3L + pmaddwd xmm7, [rel PW_MF078_F117] ; xmm7=z3H + pmaddwd xmm5, [rel PW_F117_F078] ; xmm5=z4L + pmaddwd xmm4, [rel PW_F117_F078] ; xmm4=z4H + + movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=z3L + movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=z3H + + ; (Original) + ; z1 = tmp0 + tmp3; z2 = tmp1 + tmp2; + ; tmp0 = tmp0 * 0.298631336; tmp1 = tmp1 * 2.053119869; + ; tmp2 = tmp2 * 3.072711026; tmp3 = tmp3 * 1.501321110; + ; z1 = z1 * -0.899976223; z2 = z2 * -2.562915447; + ; tmp0 += z1 + z3; tmp1 += z2 + z4; + ; tmp2 += z2 + z3; tmp3 += z1 + z4; + ; + ; (This implementation) + ; tmp0 = tmp0 * (0.298631336 - 0.899976223) + tmp3 * -0.899976223; + ; tmp1 = tmp1 * (2.053119869 - 2.562915447) + tmp2 * -2.562915447; + ; tmp2 = tmp1 * -2.562915447 + tmp2 * (3.072711026 - 2.562915447); + ; tmp3 = tmp0 * -0.899976223 + tmp3 * (1.501321110 - 0.899976223); + ; tmp0 += z3; tmp1 += z4; + ; tmp2 += z3; tmp3 += z4; + + movdqa xmm0, xmm1 + movdqa xmm7, xmm1 + punpcklwd xmm0, xmm3 + punpckhwd xmm7, xmm3 + movdqa xmm1, xmm0 + movdqa xmm3, xmm7 + pmaddwd xmm0, [rel PW_MF060_MF089] ; xmm0=tmp0L + pmaddwd xmm7, [rel PW_MF060_MF089] ; xmm7=tmp0H + pmaddwd xmm1, [rel PW_MF089_F060] ; xmm1=tmp3L + pmaddwd xmm3, [rel PW_MF089_F060] ; xmm3=tmp3H + + paddd xmm0, XMMWORD [wk(10)] ; xmm0=tmp0L + paddd xmm7, XMMWORD [wk(11)] ; xmm7=tmp0H + paddd xmm1, xmm5 ; xmm1=tmp3L + paddd xmm3, xmm4 ; xmm3=tmp3H + + movdqa XMMWORD [wk(8)], xmm0 ; wk(8)=tmp0L + movdqa XMMWORD [wk(9)], xmm7 ; wk(9)=tmp0H + + movdqa xmm0, xmm2 + movdqa xmm7, xmm2 + punpcklwd xmm0, xmm6 + punpckhwd xmm7, xmm6 + movdqa xmm2, xmm0 + movdqa xmm6, xmm7 + pmaddwd xmm0, [rel PW_MF050_MF256] ; xmm0=tmp1L + pmaddwd xmm7, [rel PW_MF050_MF256] ; xmm7=tmp1H + pmaddwd xmm2, [rel PW_MF256_F050] ; xmm2=tmp2L + pmaddwd xmm6, [rel PW_MF256_F050] ; xmm6=tmp2H + + paddd xmm0, xmm5 ; xmm0=tmp1L + paddd xmm7, xmm4 ; xmm7=tmp1H + paddd xmm2, XMMWORD [wk(10)] ; xmm2=tmp2L + paddd xmm6, XMMWORD [wk(11)] ; xmm6=tmp2H + + movdqa XMMWORD [wk(10)], xmm0 ; wk(10)=tmp1L + movdqa XMMWORD [wk(11)], xmm7 ; wk(11)=tmp1H + + ; -- Final output stage + + movdqa xmm5, XMMWORD [wk(0)] ; xmm5=tmp10L + movdqa xmm4, XMMWORD [wk(1)] ; xmm4=tmp10H + + movdqa xmm0, xmm5 + movdqa xmm7, xmm4 + paddd xmm5, xmm1 ; xmm5=data0L + paddd xmm4, xmm3 ; xmm4=data0H + psubd xmm0, xmm1 ; xmm0=data7L + psubd xmm7, xmm3 ; xmm7=data7H + + movdqa xmm1, [rel PD_DESCALE_P2] ; xmm1=[rel PD_DESCALE_P2] + + paddd xmm5, xmm1 + paddd xmm4, xmm1 + psrad xmm5, DESCALE_P2 + psrad xmm4, DESCALE_P2 + paddd xmm0, xmm1 + paddd xmm7, xmm1 + psrad xmm0, DESCALE_P2 + psrad xmm7, DESCALE_P2 + + packssdw xmm5, xmm4 ; xmm5=data0=(00 10 20 30 40 50 60 70) + packssdw xmm0, xmm7 ; xmm0=data7=(07 17 27 37 47 57 67 77) + + movdqa xmm3, XMMWORD [wk(4)] ; xmm3=tmp11L + movdqa xmm1, XMMWORD [wk(5)] ; xmm1=tmp11H + + movdqa xmm4, xmm3 + movdqa xmm7, xmm1 + paddd xmm3, xmm2 ; xmm3=data1L + paddd xmm1, xmm6 ; xmm1=data1H + psubd xmm4, xmm2 ; xmm4=data6L + psubd xmm7, xmm6 ; xmm7=data6H + + movdqa xmm2, [rel PD_DESCALE_P2] ; xmm2=[rel PD_DESCALE_P2] + + paddd xmm3, xmm2 + paddd xmm1, xmm2 + psrad xmm3, DESCALE_P2 + psrad xmm1, DESCALE_P2 + paddd xmm4, xmm2 + paddd xmm7, xmm2 + psrad xmm4, DESCALE_P2 + psrad xmm7, DESCALE_P2 + + packssdw xmm3, xmm1 ; xmm3=data1=(01 11 21 31 41 51 61 71) + packssdw xmm4, xmm7 ; xmm4=data6=(06 16 26 36 46 56 66 76) + + packsswb xmm5, xmm4 ; xmm5=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + packsswb xmm3, xmm0 ; xmm3=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + movdqa xmm6, XMMWORD [wk(6)] ; xmm6=tmp12L + movdqa xmm2, XMMWORD [wk(7)] ; xmm2=tmp12H + movdqa xmm1, XMMWORD [wk(10)] ; xmm1=tmp1L + movdqa xmm7, XMMWORD [wk(11)] ; xmm7=tmp1H + + movdqa XMMWORD [wk(0)], xmm5 ; wk(0)=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + movdqa xmm4, xmm6 + movdqa xmm0, xmm2 + paddd xmm6, xmm1 ; xmm6=data2L + paddd xmm2, xmm7 ; xmm2=data2H + psubd xmm4, xmm1 ; xmm4=data5L + psubd xmm0, xmm7 ; xmm0=data5H + + movdqa xmm5, [rel PD_DESCALE_P2] ; xmm5=[rel PD_DESCALE_P2] + + paddd xmm6, xmm5 + paddd xmm2, xmm5 + psrad xmm6, DESCALE_P2 + psrad xmm2, DESCALE_P2 + paddd xmm4, xmm5 + paddd xmm0, xmm5 + psrad xmm4, DESCALE_P2 + psrad xmm0, DESCALE_P2 + + packssdw xmm6, xmm2 ; xmm6=data2=(02 12 22 32 42 52 62 72) + packssdw xmm4, xmm0 ; xmm4=data5=(05 15 25 35 45 55 65 75) + + movdqa xmm3, XMMWORD [wk(2)] ; xmm3=tmp13L + movdqa xmm1, XMMWORD [wk(3)] ; xmm1=tmp13H + movdqa xmm7, XMMWORD [wk(8)] ; xmm7=tmp0L + movdqa xmm5, XMMWORD [wk(9)] ; xmm5=tmp0H + + movdqa xmm2, xmm3 + movdqa xmm0, xmm1 + paddd xmm3, xmm7 ; xmm3=data3L + paddd xmm1, xmm5 ; xmm1=data3H + psubd xmm2, xmm7 ; xmm2=data4L + psubd xmm0, xmm5 ; xmm0=data4H + + movdqa xmm7, [rel PD_DESCALE_P2] ; xmm7=[rel PD_DESCALE_P2] + + paddd xmm3, xmm7 + paddd xmm1, xmm7 + psrad xmm3, DESCALE_P2 + psrad xmm1, DESCALE_P2 + paddd xmm2, xmm7 + paddd xmm0, xmm7 + psrad xmm2, DESCALE_P2 + psrad xmm0, DESCALE_P2 + + movdqa xmm5, [rel PB_CENTERJSAMP] ; xmm5=[rel PB_CENTERJSAMP] + + packssdw xmm3, xmm1 ; xmm3=data3=(03 13 23 33 43 53 63 73) + packssdw xmm2, xmm0 ; xmm2=data4=(04 14 24 34 44 54 64 74) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=(00 10 20 30 40 50 60 70 06 16 26 36 46 56 66 76) + movdqa xmm1, XMMWORD [wk(1)] ; xmm1=(01 11 21 31 41 51 61 71 07 17 27 37 47 57 67 77) + + packsswb xmm6, xmm2 ; xmm6=(02 12 22 32 42 52 62 72 04 14 24 34 44 54 64 74) + packsswb xmm3, xmm4 ; xmm3=(03 13 23 33 43 53 63 73 05 15 25 35 45 55 65 75) + + paddb xmm7, xmm5 + paddb xmm1, xmm5 + paddb xmm6, xmm5 + paddb xmm3, xmm5 + + movdqa xmm0, xmm7 ; transpose coefficients(phase 1) + punpcklbw xmm7, xmm1 ; xmm7=(00 01 10 11 20 21 30 31 40 41 50 51 60 61 70 71) + punpckhbw xmm0, xmm1 ; xmm0=(06 07 16 17 26 27 36 37 46 47 56 57 66 67 76 77) + movdqa xmm2, xmm6 ; transpose coefficients(phase 1) + punpcklbw xmm6, xmm3 ; xmm6=(02 03 12 13 22 23 32 33 42 43 52 53 62 63 72 73) + punpckhbw xmm2, xmm3 ; xmm2=(04 05 14 15 24 25 34 35 44 45 54 55 64 65 74 75) + + movdqa xmm4, xmm7 ; transpose coefficients(phase 2) + punpcklwd xmm7, xmm6 ; xmm7=(00 01 02 03 10 11 12 13 20 21 22 23 30 31 32 33) + punpckhwd xmm4, xmm6 ; xmm4=(40 41 42 43 50 51 52 53 60 61 62 63 70 71 72 73) + movdqa xmm5, xmm2 ; transpose coefficients(phase 2) + punpcklwd xmm2, xmm0 ; xmm2=(04 05 06 07 14 15 16 17 24 25 26 27 34 35 36 37) + punpckhwd xmm5, xmm0 ; xmm5=(44 45 46 47 54 55 56 57 64 65 66 67 74 75 76 77) + + movdqa xmm1, xmm7 ; transpose coefficients(phase 3) + punpckldq xmm7, xmm2 ; xmm7=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + punpckhdq xmm1, xmm2 ; xmm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + movdqa xmm3, xmm4 ; transpose coefficients(phase 3) + punpckldq xmm4, xmm5 ; xmm4=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) + punpckhdq xmm3, xmm5 ; xmm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) + + pshufd xmm6, xmm7, 0x4E ; xmm6=(10 11 12 13 14 15 16 17 00 01 02 03 04 05 06 07) + pshufd xmm0, xmm1, 0x4E ; xmm0=(30 31 32 33 34 35 36 37 20 21 22 23 24 25 26 27) + pshufd xmm2, xmm4, 0x4E ; xmm2=(50 51 52 53 54 55 56 57 40 41 42 43 44 45 46 47) + pshufd xmm5, xmm3, 0x4E ; xmm5=(70 71 72 73 74 75 76 77 60 61 62 63 64 65 66 67) + + mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+2*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm7 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm1 + mov rdxp, JSAMPROW [rdi+4*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+6*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3 + + mov rdxp, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+3*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm6 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm0 + mov rdxp, JSAMPROW [rdi+5*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+7*SIZEOF_JSAMPROW] + movq XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE], xmm2 + movq XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE], xmm5 + + UNCOLLECT_ARGS 4 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctred-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctred-sse2.asm new file mode 100644 index 0000000..6fb7095 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jidctred-sse2.asm @@ -0,0 +1,570 @@ +; +; jidctred.asm - reduced-size IDCT (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. +; +; This file contains inverse-DCT routines that produce reduced-size +; output: either 4x4 or 2x2 pixels from an 8x8 DCT block. +; The following code is based directly on the IJG's original jidctred.c; +; see the jidctred.c for more details. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + +%define CONST_BITS 13 +%define PASS1_BITS 2 + +%define DESCALE_P1_4 (CONST_BITS - PASS1_BITS + 1) +%define DESCALE_P2_4 (CONST_BITS + PASS1_BITS + 3 + 1) +%define DESCALE_P1_2 (CONST_BITS - PASS1_BITS + 2) +%define DESCALE_P2_2 (CONST_BITS + PASS1_BITS + 3 + 2) + +%if CONST_BITS == 13 +F_0_211 equ 1730 ; FIX(0.211164243) +F_0_509 equ 4176 ; FIX(0.509795579) +F_0_601 equ 4926 ; FIX(0.601344887) +F_0_720 equ 5906 ; FIX(0.720959822) +F_0_765 equ 6270 ; FIX(0.765366865) +F_0_850 equ 6967 ; FIX(0.850430095) +F_0_899 equ 7373 ; FIX(0.899976223) +F_1_061 equ 8697 ; FIX(1.061594337) +F_1_272 equ 10426 ; FIX(1.272758580) +F_1_451 equ 11893 ; FIX(1.451774981) +F_1_847 equ 15137 ; FIX(1.847759065) +F_2_172 equ 17799 ; FIX(2.172734803) +F_2_562 equ 20995 ; FIX(2.562915447) +F_3_624 equ 29692 ; FIX(3.624509785) +%else +; NASM cannot do compile-time arithmetic on floating-point constants. +%define DESCALE(x, n) (((x) + (1 << ((n) - 1))) >> (n)) +F_0_211 equ DESCALE( 226735879, 30 - CONST_BITS) ; FIX(0.211164243) +F_0_509 equ DESCALE( 547388834, 30 - CONST_BITS) ; FIX(0.509795579) +F_0_601 equ DESCALE( 645689155, 30 - CONST_BITS) ; FIX(0.601344887) +F_0_720 equ DESCALE( 774124714, 30 - CONST_BITS) ; FIX(0.720959822) +F_0_765 equ DESCALE( 821806413, 30 - CONST_BITS) ; FIX(0.765366865) +F_0_850 equ DESCALE( 913142361, 30 - CONST_BITS) ; FIX(0.850430095) +F_0_899 equ DESCALE( 966342111, 30 - CONST_BITS) ; FIX(0.899976223) +F_1_061 equ DESCALE(1139878239, 30 - CONST_BITS) ; FIX(1.061594337) +F_1_272 equ DESCALE(1366614119, 30 - CONST_BITS) ; FIX(1.272758580) +F_1_451 equ DESCALE(1558831516, 30 - CONST_BITS) ; FIX(1.451774981) +F_1_847 equ DESCALE(1984016188, 30 - CONST_BITS) ; FIX(1.847759065) +F_2_172 equ DESCALE(2332956230, 30 - CONST_BITS) ; FIX(2.172734803) +F_2_562 equ DESCALE(2751909506, 30 - CONST_BITS) ; FIX(2.562915447) +F_3_624 equ DESCALE(3891787747, 30 - CONST_BITS) ; FIX(3.624509785) +%endif + +; -------------------------------------------------------------------------- + SECTION SEG_CONST + + ALIGNZ 32 + GLOBAL_DATA(jconst_idct_red_sse2) + +EXTN(jconst_idct_red_sse2): + +PW_F184_MF076 times 4 dw F_1_847, -F_0_765 +PW_F256_F089 times 4 dw F_2_562, F_0_899 +PW_F106_MF217 times 4 dw F_1_061, -F_2_172 +PW_MF060_MF050 times 4 dw -F_0_601, -F_0_509 +PW_F145_MF021 times 4 dw F_1_451, -F_0_211 +PW_F362_MF127 times 4 dw F_3_624, -F_1_272 +PW_F085_MF072 times 4 dw F_0_850, -F_0_720 +PD_DESCALE_P1_4 times 4 dd 1 << (DESCALE_P1_4 - 1) +PD_DESCALE_P2_4 times 4 dd 1 << (DESCALE_P2_4 - 1) +PD_DESCALE_P1_2 times 4 dd 1 << (DESCALE_P1_2 - 1) +PD_DESCALE_P2_2 times 4 dd 1 << (DESCALE_P2_2 - 1) +PB_CENTERJSAMP times 16 db CENTERJSAMPLE + + ALIGNZ 32 + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Perform dequantization and inverse DCT on one block of coefficients, +; producing a reduced-size 4x4 output block. +; +; GLOBAL(void) +; jsimd_idct_4x4_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +; r10 = void *dct_table +; r11 = JCOEFPTR coef_block +; r12 = JSAMPARRAY output_buf +; r13d = JDIMENSION output_col + +%define wk(i) r15 - (WK_NUM - (i)) * SIZEOF_XMMWORD + ; xmmword wk[WK_NUM] +%define WK_NUM 2 + + align 32 + GLOBAL_FUNCTION(jsimd_idct_4x4_sse2) + +EXTN(jsimd_idct_4x4_sse2): + ENDBR64 + push rbp + mov rbp, rsp + push r15 + and rsp, byte (-SIZEOF_XMMWORD) ; align to 128 bits + ; Allocate stack space for wk array. r15 is used to access it. + mov r15, rsp + sub rsp, byte (SIZEOF_XMMWORD * WK_NUM) + COLLECT_ARGS 4 + + ; ---- Pass 1: process columns from input. + + mov rdx, r10 ; quantptr + mov rsi, r11 ; inptr + +%ifndef NO_ZERO_COLUMN_TEST_4X4_SSE2 + mov eax, dword [DWBLOCK(1,0,rsi,SIZEOF_JCOEF)] + or eax, dword [DWBLOCK(2,0,rsi,SIZEOF_JCOEF)] + jnz short .columnDCT + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + por xmm0, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + por xmm1, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + por xmm0, xmm1 + packsswb xmm0, xmm0 + packsswb xmm0, xmm0 + movd eax, xmm0 + test rax, rax + jnz short .columnDCT + + ; -- AC terms all zero + + movdqa xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + psllw xmm0, PASS1_BITS + + movdqa xmm3, xmm0 ; xmm0=in0=(00 01 02 03 04 05 06 07) + punpcklwd xmm0, xmm0 ; xmm0=(00 00 01 01 02 02 03 03) + punpckhwd xmm3, xmm3 ; xmm3=(04 04 05 05 06 06 07 07) + + pshufd xmm1, xmm0, 0x50 ; xmm1=[col0 col1]=(00 00 00 00 01 01 01 01) + pshufd xmm0, xmm0, 0xFA ; xmm0=[col2 col3]=(02 02 02 02 03 03 03 03) + pshufd xmm6, xmm3, 0x50 ; xmm6=[col4 col5]=(04 04 04 04 05 05 05 05) + pshufd xmm3, xmm3, 0xFA ; xmm3=[col6 col7]=(06 06 06 06 07 07 07 07) + + jmp near .column_end +%endif +.columnDCT: + + ; -- Odd part + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + movdqa xmm4, xmm0 + movdqa xmm5, xmm0 + punpcklwd xmm4, xmm1 + punpckhwd xmm5, xmm1 + movdqa xmm0, xmm4 + movdqa xmm1, xmm5 + pmaddwd xmm4, [rel PW_F256_F089] ; xmm4=(tmp2L) + pmaddwd xmm5, [rel PW_F256_F089] ; xmm5=(tmp2H) + pmaddwd xmm0, [rel PW_F106_MF217] ; xmm0=(tmp0L) + pmaddwd xmm1, [rel PW_F106_MF217] ; xmm1=(tmp0H) + + movdqa xmm6, xmm2 + movdqa xmm7, xmm2 + punpcklwd xmm6, xmm3 + punpckhwd xmm7, xmm3 + movdqa xmm2, xmm6 + movdqa xmm3, xmm7 + pmaddwd xmm6, [rel PW_MF060_MF050] ; xmm6=(tmp2L) + pmaddwd xmm7, [rel PW_MF060_MF050] ; xmm7=(tmp2H) + pmaddwd xmm2, [rel PW_F145_MF021] ; xmm2=(tmp0L) + pmaddwd xmm3, [rel PW_F145_MF021] ; xmm3=(tmp0H) + + paddd xmm6, xmm4 ; xmm6=tmp2L + paddd xmm7, xmm5 ; xmm7=tmp2H + paddd xmm2, xmm0 ; xmm2=tmp0L + paddd xmm3, xmm1 ; xmm3=tmp0H + + movdqa XMMWORD [wk(0)], xmm2 ; wk(0)=tmp0L + movdqa XMMWORD [wk(1)], xmm3 ; wk(1)=tmp0H + + ; -- Even part + + movdqa xmm4, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + movdqa xmm5, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_JCOEF)] + movdqa xmm0, XMMWORD [XMMBLOCK(6,0,rsi,SIZEOF_JCOEF)] + pmullw xmm4, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm5, XMMWORD [XMMBLOCK(2,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm0, XMMWORD [XMMBLOCK(6,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + pxor xmm1, xmm1 + pxor xmm2, xmm2 + punpcklwd xmm1, xmm4 ; xmm1=tmp0L + punpckhwd xmm2, xmm4 ; xmm2=tmp0H + psrad xmm1, (16-CONST_BITS-1) ; psrad xmm1,16 & pslld xmm1,CONST_BITS+1 + psrad xmm2, (16-CONST_BITS-1) ; psrad xmm2,16 & pslld xmm2,CONST_BITS+1 + + movdqa xmm3, xmm5 ; xmm5=in2=z2 + punpcklwd xmm5, xmm0 ; xmm0=in6=z3 + punpckhwd xmm3, xmm0 + pmaddwd xmm5, [rel PW_F184_MF076] ; xmm5=tmp2L + pmaddwd xmm3, [rel PW_F184_MF076] ; xmm3=tmp2H + + movdqa xmm4, xmm1 + movdqa xmm0, xmm2 + paddd xmm1, xmm5 ; xmm1=tmp10L + paddd xmm2, xmm3 ; xmm2=tmp10H + psubd xmm4, xmm5 ; xmm4=tmp12L + psubd xmm0, xmm3 ; xmm0=tmp12H + + ; -- Final output stage + + movdqa xmm5, xmm1 + movdqa xmm3, xmm2 + paddd xmm1, xmm6 ; xmm1=data0L + paddd xmm2, xmm7 ; xmm2=data0H + psubd xmm5, xmm6 ; xmm5=data3L + psubd xmm3, xmm7 ; xmm3=data3H + + movdqa xmm6, [rel PD_DESCALE_P1_4] ; xmm6=[rel PD_DESCALE_P1_4] + + paddd xmm1, xmm6 + paddd xmm2, xmm6 + psrad xmm1, DESCALE_P1_4 + psrad xmm2, DESCALE_P1_4 + paddd xmm5, xmm6 + paddd xmm3, xmm6 + psrad xmm5, DESCALE_P1_4 + psrad xmm3, DESCALE_P1_4 + + packssdw xmm1, xmm2 ; xmm1=data0=(00 01 02 03 04 05 06 07) + packssdw xmm5, xmm3 ; xmm5=data3=(30 31 32 33 34 35 36 37) + + movdqa xmm7, XMMWORD [wk(0)] ; xmm7=tmp0L + movdqa xmm6, XMMWORD [wk(1)] ; xmm6=tmp0H + + movdqa xmm2, xmm4 + movdqa xmm3, xmm0 + paddd xmm4, xmm7 ; xmm4=data1L + paddd xmm0, xmm6 ; xmm0=data1H + psubd xmm2, xmm7 ; xmm2=data2L + psubd xmm3, xmm6 ; xmm3=data2H + + movdqa xmm7, [rel PD_DESCALE_P1_4] ; xmm7=[rel PD_DESCALE_P1_4] + + paddd xmm4, xmm7 + paddd xmm0, xmm7 + psrad xmm4, DESCALE_P1_4 + psrad xmm0, DESCALE_P1_4 + paddd xmm2, xmm7 + paddd xmm3, xmm7 + psrad xmm2, DESCALE_P1_4 + psrad xmm3, DESCALE_P1_4 + + packssdw xmm4, xmm0 ; xmm4=data1=(10 11 12 13 14 15 16 17) + packssdw xmm2, xmm3 ; xmm2=data2=(20 21 22 23 24 25 26 27) + + movdqa xmm6, xmm1 ; transpose coefficients(phase 1) + punpcklwd xmm1, xmm4 ; xmm1=(00 10 01 11 02 12 03 13) + punpckhwd xmm6, xmm4 ; xmm6=(04 14 05 15 06 16 07 17) + movdqa xmm7, xmm2 ; transpose coefficients(phase 1) + punpcklwd xmm2, xmm5 ; xmm2=(20 30 21 31 22 32 23 33) + punpckhwd xmm7, xmm5 ; xmm7=(24 34 25 35 26 36 27 37) + + movdqa xmm0, xmm1 ; transpose coefficients(phase 2) + punpckldq xmm1, xmm2 ; xmm1=[col0 col1]=(00 10 20 30 01 11 21 31) + punpckhdq xmm0, xmm2 ; xmm0=[col2 col3]=(02 12 22 32 03 13 23 33) + movdqa xmm3, xmm6 ; transpose coefficients(phase 2) + punpckldq xmm6, xmm7 ; xmm6=[col4 col5]=(04 14 24 34 05 15 25 35) + punpckhdq xmm3, xmm7 ; xmm3=[col6 col7]=(06 16 26 36 07 17 27 37) +.column_end: + + ; -- Prefetch the next coefficient block + + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows, store into output array. + + mov rdi, r12 ; (JSAMPROW *) + mov eax, r13d + + ; -- Even part + + pxor xmm4, xmm4 + punpcklwd xmm4, xmm1 ; xmm4=tmp0 + psrad xmm4, (16-CONST_BITS-1) ; psrad xmm4,16 & pslld xmm4,CONST_BITS+1 + + ; -- Odd part + + punpckhwd xmm1, xmm0 + punpckhwd xmm6, xmm3 + movdqa xmm5, xmm1 + movdqa xmm2, xmm6 + pmaddwd xmm1, [rel PW_F256_F089] ; xmm1=(tmp2) + pmaddwd xmm6, [rel PW_MF060_MF050] ; xmm6=(tmp2) + pmaddwd xmm5, [rel PW_F106_MF217] ; xmm5=(tmp0) + pmaddwd xmm2, [rel PW_F145_MF021] ; xmm2=(tmp0) + + paddd xmm6, xmm1 ; xmm6=tmp2 + paddd xmm2, xmm5 ; xmm2=tmp0 + + ; -- Even part + + punpcklwd xmm0, xmm3 + pmaddwd xmm0, [rel PW_F184_MF076] ; xmm0=tmp2 + + movdqa xmm7, xmm4 + paddd xmm4, xmm0 ; xmm4=tmp10 + psubd xmm7, xmm0 ; xmm7=tmp12 + + ; -- Final output stage + + movdqa xmm1, [rel PD_DESCALE_P2_4] ; xmm1=[rel PD_DESCALE_P2_4] + + movdqa xmm5, xmm4 + movdqa xmm3, xmm7 + paddd xmm4, xmm6 ; xmm4=data0=(00 10 20 30) + paddd xmm7, xmm2 ; xmm7=data1=(01 11 21 31) + psubd xmm5, xmm6 ; xmm5=data3=(03 13 23 33) + psubd xmm3, xmm2 ; xmm3=data2=(02 12 22 32) + + paddd xmm4, xmm1 + paddd xmm7, xmm1 + psrad xmm4, DESCALE_P2_4 + psrad xmm7, DESCALE_P2_4 + paddd xmm5, xmm1 + paddd xmm3, xmm1 + psrad xmm5, DESCALE_P2_4 + psrad xmm3, DESCALE_P2_4 + + packssdw xmm4, xmm3 ; xmm4=(00 10 20 30 02 12 22 32) + packssdw xmm7, xmm5 ; xmm7=(01 11 21 31 03 13 23 33) + + movdqa xmm0, xmm4 ; transpose coefficients(phase 1) + punpcklwd xmm4, xmm7 ; xmm4=(00 01 10 11 20 21 30 31) + punpckhwd xmm0, xmm7 ; xmm0=(02 03 12 13 22 23 32 33) + + movdqa xmm6, xmm4 ; transpose coefficients(phase 2) + punpckldq xmm4, xmm0 ; xmm4=(00 01 02 03 10 11 12 13) + punpckhdq xmm6, xmm0 ; xmm6=(20 21 22 23 30 31 32 33) + + packsswb xmm4, xmm6 ; xmm4=(00 01 02 03 10 11 12 13 20 ..) + paddb xmm4, [rel PB_CENTERJSAMP] + + pshufd xmm2, xmm4, 0x39 ; xmm2=(10 11 12 13 20 21 22 23 30 ..) + pshufd xmm1, xmm4, 0x4E ; xmm1=(20 21 22 23 30 31 32 33 00 ..) + pshufd xmm3, xmm4, 0x93 ; xmm3=(30 31 32 33 00 01 02 03 10 ..) + + mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] + movd XMM_DWORD [rdx+rax*SIZEOF_JSAMPLE], xmm4 + movd XMM_DWORD [rsi+rax*SIZEOF_JSAMPLE], xmm2 + mov rdxp, JSAMPROW [rdi+2*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+3*SIZEOF_JSAMPROW] + movd XMM_DWORD [rdx+rax*SIZEOF_JSAMPLE], xmm1 + movd XMM_DWORD [rsi+rax*SIZEOF_JSAMPLE], xmm3 + + UNCOLLECT_ARGS 4 + lea rsp, [rbp-8] + pop r15 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Perform dequantization and inverse DCT on one block of coefficients, +; producing a reduced-size 2x2 output block. +; +; GLOBAL(void) +; jsimd_idct_2x2_sse2(void *dct_table, JCOEFPTR coef_block, +; JSAMPARRAY output_buf, JDIMENSION output_col) +; + +; r10 = void *dct_table +; r11 = JCOEFPTR coef_block +; r12 = JSAMPARRAY output_buf +; r13d = JDIMENSION output_col + + align 32 + GLOBAL_FUNCTION(jsimd_idct_2x2_sse2) + +EXTN(jsimd_idct_2x2_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 4 + push rbx + + ; ---- Pass 1: process columns from input. + + mov rdx, r10 ; quantptr + mov rsi, r11 ; inptr + + ; | input: | result: | + ; | 00 01 ** 03 ** 05 ** 07 | | + ; | 10 11 ** 13 ** 15 ** 17 | | + ; | ** ** ** ** ** ** ** ** | | + ; | 30 31 ** 33 ** 35 ** 37 | A0 A1 A3 A5 A7 | + ; | ** ** ** ** ** ** ** ** | B0 B1 B3 B5 B7 | + ; | 50 51 ** 53 ** 55 ** 57 | | + ; | ** ** ** ** ** ** ** ** | | + ; | 70 71 ** 73 ** 75 ** 77 | | + + ; -- Odd part + + movdqa xmm0, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_JCOEF)] + movdqa xmm1, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_JCOEF)] + pmullw xmm0, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm1, XMMWORD [XMMBLOCK(3,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + movdqa xmm2, XMMWORD [XMMBLOCK(5,0,rsi,SIZEOF_JCOEF)] + movdqa xmm3, XMMWORD [XMMBLOCK(7,0,rsi,SIZEOF_JCOEF)] + pmullw xmm2, XMMWORD [XMMBLOCK(5,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + pmullw xmm3, XMMWORD [XMMBLOCK(7,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + ; xmm0=(10 11 ** 13 ** 15 ** 17), xmm1=(30 31 ** 33 ** 35 ** 37) + ; xmm2=(50 51 ** 53 ** 55 ** 57), xmm3=(70 71 ** 73 ** 75 ** 77) + + pcmpeqd xmm7, xmm7 + pslld xmm7, WORD_BIT ; xmm7={0x0000 0xFFFF 0x0000 0xFFFF ..} + + movdqa xmm4, xmm0 ; xmm4=(10 11 ** 13 ** 15 ** 17) + movdqa xmm5, xmm2 ; xmm5=(50 51 ** 53 ** 55 ** 57) + punpcklwd xmm4, xmm1 ; xmm4=(10 30 11 31 ** ** 13 33) + punpcklwd xmm5, xmm3 ; xmm5=(50 70 51 71 ** ** 53 73) + pmaddwd xmm4, [rel PW_F362_MF127] + pmaddwd xmm5, [rel PW_F085_MF072] + + psrld xmm0, WORD_BIT ; xmm0=(11 -- 13 -- 15 -- 17 --) + pand xmm1, xmm7 ; xmm1=(-- 31 -- 33 -- 35 -- 37) + psrld xmm2, WORD_BIT ; xmm2=(51 -- 53 -- 55 -- 57 --) + pand xmm3, xmm7 ; xmm3=(-- 71 -- 73 -- 75 -- 77) + por xmm0, xmm1 ; xmm0=(11 31 13 33 15 35 17 37) + por xmm2, xmm3 ; xmm2=(51 71 53 73 55 75 57 77) + pmaddwd xmm0, [rel PW_F362_MF127] + pmaddwd xmm2, [rel PW_F085_MF072] + + paddd xmm4, xmm5 ; xmm4=tmp0[col0 col1 **** col3] + paddd xmm0, xmm2 ; xmm0=tmp0[col1 col3 col5 col7] + + ; -- Even part + + movdqa xmm6, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_JCOEF)] + pmullw xmm6, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_ISLOW_MULT_TYPE)] + + ; xmm6=(00 01 ** 03 ** 05 ** 07) + + movdqa xmm1, xmm6 ; xmm1=(00 01 ** 03 ** 05 ** 07) + pslld xmm6, WORD_BIT ; xmm6=(-- 00 -- ** -- ** -- **) + pand xmm1, xmm7 ; xmm1=(-- 01 -- 03 -- 05 -- 07) + psrad xmm6, (WORD_BIT-CONST_BITS-2) ; xmm6=tmp10[col0 **** **** ****] + psrad xmm1, (WORD_BIT-CONST_BITS-2) ; xmm1=tmp10[col1 col3 col5 col7] + + ; -- Final output stage + + movdqa xmm3, xmm6 + movdqa xmm5, xmm1 + paddd xmm6, xmm4 ; xmm6=data0[col0 **** **** ****]=(A0 ** ** **) + paddd xmm1, xmm0 ; xmm1=data0[col1 col3 col5 col7]=(A1 A3 A5 A7) + psubd xmm3, xmm4 ; xmm3=data1[col0 **** **** ****]=(B0 ** ** **) + psubd xmm5, xmm0 ; xmm5=data1[col1 col3 col5 col7]=(B1 B3 B5 B7) + + movdqa xmm2, [rel PD_DESCALE_P1_2] ; xmm2=[rel PD_DESCALE_P1_2] + + punpckldq xmm6, xmm3 ; xmm6=(A0 B0 ** **) + + movdqa xmm7, xmm1 + punpcklqdq xmm1, xmm5 ; xmm1=(A1 A3 B1 B3) + punpckhqdq xmm7, xmm5 ; xmm7=(A5 A7 B5 B7) + + paddd xmm6, xmm2 + psrad xmm6, DESCALE_P1_2 + + paddd xmm1, xmm2 + paddd xmm7, xmm2 + psrad xmm1, DESCALE_P1_2 + psrad xmm7, DESCALE_P1_2 + + ; -- Prefetch the next coefficient block + + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 0*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 1*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 2*32] + prefetchnta [rsi + DCTSIZE2*SIZEOF_JCOEF + 3*32] + + ; ---- Pass 2: process rows, store into output array. + + mov rdi, r12 ; (JSAMPROW *) + mov eax, r13d + + ; | input:| result:| + ; | A0 B0 | | + ; | A1 B1 | C0 C1 | + ; | A3 B3 | D0 D1 | + ; | A5 B5 | | + ; | A7 B7 | | + + ; -- Odd part + + packssdw xmm1, xmm1 ; xmm1=(A1 A3 B1 B3 A1 A3 B1 B3) + packssdw xmm7, xmm7 ; xmm7=(A5 A7 B5 B7 A5 A7 B5 B7) + pmaddwd xmm1, [rel PW_F362_MF127] + pmaddwd xmm7, [rel PW_F085_MF072] + + paddd xmm1, xmm7 ; xmm1=tmp0[row0 row1 row0 row1] + + ; -- Even part + + pslld xmm6, (CONST_BITS+2) ; xmm6=tmp10[row0 row1 **** ****] + + ; -- Final output stage + + movdqa xmm4, xmm6 + paddd xmm6, xmm1 ; xmm6=data0[row0 row1 **** ****]=(C0 C1 ** **) + psubd xmm4, xmm1 ; xmm4=data1[row0 row1 **** ****]=(D0 D1 ** **) + + punpckldq xmm6, xmm4 ; xmm6=(C0 D0 C1 D1) + + paddd xmm6, [rel PD_DESCALE_P2_2] + psrad xmm6, DESCALE_P2_2 + + packssdw xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1) + packsswb xmm6, xmm6 ; xmm6=(C0 D0 C1 D1 C0 D0 C1 D1 ..) + paddb xmm6, [rel PB_CENTERJSAMP] + + pextrw ebx, xmm6, 0x00 ; ebx=(C0 D0 -- --) + pextrw ecx, xmm6, 0x01 ; ecx=(C1 D1 -- --) + + mov rdxp, JSAMPROW [rdi+0*SIZEOF_JSAMPROW] + mov rsip, JSAMPROW [rdi+1*SIZEOF_JSAMPROW] + mov word [rdx+rax*SIZEOF_JSAMPLE], bx + mov word [rsi+rax*SIZEOF_JSAMPLE], cx + + pop rbx + UNCOLLECT_ARGS 4 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquantf-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquantf-sse2.asm new file mode 100644 index 0000000..6476333 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquantf-sse2.asm @@ -0,0 +1,151 @@ +; +; jquantf.asm - sample data conversion and quantization (64-bit SSE & SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_float_sse2(JSAMPARRAY sample_data, JDIMENSION start_col, +; FAST_FLOAT *workspace); +; + +; r10 = JSAMPARRAY sample_data +; r11d = JDIMENSION start_col +; r12 = FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_float_sse2) + +EXTN(jsimd_convsamp_float_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 3 + push rbx + + pcmpeqw xmm7, xmm7 + psllw xmm7, 7 + packsswb xmm7, xmm7 ; xmm7 = PB_CENTERJSAMPLE (0x808080..) + + mov rsi, r10 + mov eax, r11d + mov rdi, r12 + mov rcx, DCTSIZE/2 +.convloop: + mov rbxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rdxp, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] + movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] + + psubb xmm0, xmm7 ; xmm0=(01234567) + psubb xmm1, xmm7 ; xmm1=(89ABCDEF) + + punpcklbw xmm0, xmm0 ; xmm0=(*0*1*2*3*4*5*6*7) + punpcklbw xmm1, xmm1 ; xmm1=(*8*9*A*B*C*D*E*F) + + punpcklwd xmm2, xmm0 ; xmm2=(***0***1***2***3) + punpckhwd xmm0, xmm0 ; xmm0=(***4***5***6***7) + punpcklwd xmm3, xmm1 ; xmm3=(***8***9***A***B) + punpckhwd xmm1, xmm1 ; xmm1=(***C***D***E***F) + + psrad xmm2, (DWORD_BIT-BYTE_BIT) ; xmm2=(0123) + psrad xmm0, (DWORD_BIT-BYTE_BIT) ; xmm0=(4567) + cvtdq2ps xmm2, xmm2 ; xmm2=(0123) + cvtdq2ps xmm0, xmm0 ; xmm0=(4567) + psrad xmm3, (DWORD_BIT-BYTE_BIT) ; xmm3=(89AB) + psrad xmm1, (DWORD_BIT-BYTE_BIT) ; xmm1=(CDEF) + cvtdq2ps xmm3, xmm3 ; xmm3=(89AB) + cvtdq2ps xmm1, xmm1 ; xmm1=(CDEF) + + movaps XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_FAST_FLOAT)], xmm2 + movaps XMMWORD [XMMBLOCK(0,1,rdi,SIZEOF_FAST_FLOAT)], xmm0 + movaps XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_FAST_FLOAT)], xmm3 + movaps XMMWORD [XMMBLOCK(1,1,rdi,SIZEOF_FAST_FLOAT)], xmm1 + + add rsi, byte 2*SIZEOF_JSAMPROW + add rdi, byte 2*DCTSIZE*SIZEOF_FAST_FLOAT + dec rcx + jnz short .convloop + + pop rbx + UNCOLLECT_ARGS 3 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; GLOBAL(void) +; jsimd_quantize_float_sse2(JCOEFPTR coef_block, FAST_FLOAT *divisors, +; FAST_FLOAT *workspace); +; + +; r10 = JCOEFPTR coef_block +; r11 = FAST_FLOAT *divisors +; r12 = FAST_FLOAT *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_float_sse2) + +EXTN(jsimd_quantize_float_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 3 + + mov rsi, r12 + mov rdx, r11 + mov rdi, r10 + mov rax, DCTSIZE2/16 +.quantloop: + movaps xmm0, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm1, XMMWORD [XMMBLOCK(0,1,rsi,SIZEOF_FAST_FLOAT)] + mulps xmm0, XMMWORD [XMMBLOCK(0,0,rdx,SIZEOF_FAST_FLOAT)] + mulps xmm1, XMMWORD [XMMBLOCK(0,1,rdx,SIZEOF_FAST_FLOAT)] + movaps xmm2, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_FAST_FLOAT)] + movaps xmm3, XMMWORD [XMMBLOCK(1,1,rsi,SIZEOF_FAST_FLOAT)] + mulps xmm2, XMMWORD [XMMBLOCK(1,0,rdx,SIZEOF_FAST_FLOAT)] + mulps xmm3, XMMWORD [XMMBLOCK(1,1,rdx,SIZEOF_FAST_FLOAT)] + + cvtps2dq xmm0, xmm0 + cvtps2dq xmm1, xmm1 + cvtps2dq xmm2, xmm2 + cvtps2dq xmm3, xmm3 + + packssdw xmm0, xmm1 + packssdw xmm2, xmm3 + + movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_JCOEF)], xmm0 + movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_JCOEF)], xmm2 + + add rsi, byte 16*SIZEOF_FAST_FLOAT + add rdx, byte 16*SIZEOF_FAST_FLOAT + add rdi, byte 16*SIZEOF_JCOEF + dec rax + jnz short .quantloop + + UNCOLLECT_ARGS 3 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-avx2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-avx2.asm new file mode 100644 index 0000000..7e126e8 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-avx2.asm @@ -0,0 +1,159 @@ +; +; jquanti.asm - sample data conversion and quantization (64-bit AVX2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2018, 2024, D. R. Commander. +; Copyright (C) 2016, Matthieu Darbois. +; Copyright (C) 2018, Matthias Räncker. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_avx2(JSAMPARRAY sample_data, JDIMENSION start_col, +; DCTELEM *workspace); +; + +; r10 = JSAMPARRAY sample_data +; r11d = JDIMENSION start_col +; r12 = DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_avx2) + +EXTN(jsimd_convsamp_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 3 + + mov eax, r11d + + mov rsip, JSAMPROW [r10+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rdip, JSAMPROW [r10+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm0, XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE] + pinsrq xmm0, XMM_MMWORD [rdi+rax*SIZEOF_JSAMPLE], 1 + + mov rsip, JSAMPROW [r10+2*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rdip, JSAMPROW [r10+3*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm1, XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE] + pinsrq xmm1, XMM_MMWORD [rdi+rax*SIZEOF_JSAMPLE], 1 + + mov rsip, JSAMPROW [r10+4*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rdip, JSAMPROW [r10+5*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm2, XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE] + pinsrq xmm2, XMM_MMWORD [rdi+rax*SIZEOF_JSAMPLE], 1 + + mov rsip, JSAMPROW [r10+6*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rdip, JSAMPROW [r10+7*SIZEOF_JSAMPROW] ; (JSAMPLE *) + movq xmm3, XMM_MMWORD [rsi+rax*SIZEOF_JSAMPLE] + pinsrq xmm3, XMM_MMWORD [rdi+rax*SIZEOF_JSAMPLE], 1 + + vpmovzxbw ymm0, xmm0 ; ymm0=(00 01 02 03 04 05 06 07 10 11 12 13 14 15 16 17) + vpmovzxbw ymm1, xmm1 ; ymm1=(20 21 22 23 24 25 26 27 30 31 32 33 34 35 36 37) + vpmovzxbw ymm2, xmm2 ; ymm2=(40 41 42 43 44 45 46 47 50 51 52 53 54 55 56 57) + vpmovzxbw ymm3, xmm3 ; ymm3=(60 61 62 63 64 65 66 67 70 71 72 73 74 75 76 77) + + vpcmpeqw ymm7, ymm7, ymm7 + vpsllw ymm7, ymm7, 7 ; ymm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + vpaddw ymm0, ymm0, ymm7 + vpaddw ymm1, ymm1, ymm7 + vpaddw ymm2, ymm2, ymm7 + vpaddw ymm3, ymm3, ymm7 + + vmovdqu YMMWORD [YMMBLOCK(0,0,r12,SIZEOF_DCTELEM)], ymm0 + vmovdqu YMMWORD [YMMBLOCK(2,0,r12,SIZEOF_DCTELEM)], ymm1 + vmovdqu YMMWORD [YMMBLOCK(4,0,r12,SIZEOF_DCTELEM)], ymm2 + vmovdqu YMMWORD [YMMBLOCK(6,0,r12,SIZEOF_DCTELEM)], ymm3 + + vzeroupper + UNCOLLECT_ARGS 3 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; This implementation is based on an algorithm described in +; "Optimizing subroutines in assembly language: +; An optimization guide for x86 platforms" (https://agner.org/optimize). +; +; GLOBAL(void) +; jsimd_quantize_avx2(JCOEFPTR coef_block, DCTELEM *divisors, +; DCTELEM *workspace); +; + +%define RECIPROCAL(m, n, b) \ + YMMBLOCK(DCTSIZE * 0 + (m), (n), (b), SIZEOF_DCTELEM) +%define CORRECTION(m, n, b) \ + YMMBLOCK(DCTSIZE * 1 + (m), (n), (b), SIZEOF_DCTELEM) +%define SCALE(m, n, b) \ + YMMBLOCK(DCTSIZE * 2 + (m), (n), (b), SIZEOF_DCTELEM) + +; r10 = JCOEFPTR coef_block +; r11 = DCTELEM *divisors +; r12 = DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_avx2) + +EXTN(jsimd_quantize_avx2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 3 + + vmovdqu ymm4, [YMMBLOCK(0,0,r12,SIZEOF_DCTELEM)] + vmovdqu ymm5, [YMMBLOCK(2,0,r12,SIZEOF_DCTELEM)] + vmovdqu ymm6, [YMMBLOCK(4,0,r12,SIZEOF_DCTELEM)] + vmovdqu ymm7, [YMMBLOCK(6,0,r12,SIZEOF_DCTELEM)] + vpabsw ymm0, ymm4 + vpabsw ymm1, ymm5 + vpabsw ymm2, ymm6 + vpabsw ymm3, ymm7 + + vpaddw ymm0, YMMWORD [CORRECTION(0,0,r11)] ; correction + roundfactor + vpaddw ymm1, YMMWORD [CORRECTION(2,0,r11)] + vpaddw ymm2, YMMWORD [CORRECTION(4,0,r11)] + vpaddw ymm3, YMMWORD [CORRECTION(6,0,r11)] + vpmulhuw ymm0, YMMWORD [RECIPROCAL(0,0,r11)] ; reciprocal + vpmulhuw ymm1, YMMWORD [RECIPROCAL(2,0,r11)] + vpmulhuw ymm2, YMMWORD [RECIPROCAL(4,0,r11)] + vpmulhuw ymm3, YMMWORD [RECIPROCAL(6,0,r11)] + vpmulhuw ymm0, YMMWORD [SCALE(0,0,r11)] ; scale + vpmulhuw ymm1, YMMWORD [SCALE(2,0,r11)] + vpmulhuw ymm2, YMMWORD [SCALE(4,0,r11)] + vpmulhuw ymm3, YMMWORD [SCALE(6,0,r11)] + + vpsignw ymm0, ymm0, ymm4 + vpsignw ymm1, ymm1, ymm5 + vpsignw ymm2, ymm2, ymm6 + vpsignw ymm3, ymm3, ymm7 + + vmovdqu [YMMBLOCK(0,0,r10,SIZEOF_DCTELEM)], ymm0 + vmovdqu [YMMBLOCK(2,0,r10,SIZEOF_DCTELEM)], ymm1 + vmovdqu [YMMBLOCK(4,0,r10,SIZEOF_DCTELEM)], ymm2 + vmovdqu [YMMBLOCK(6,0,r10,SIZEOF_DCTELEM)], ymm3 + + vzeroupper + UNCOLLECT_ARGS 3 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-sse2.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-sse2.asm new file mode 100644 index 0000000..284b9fe --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jquanti-sse2.asm @@ -0,0 +1,184 @@ +; +; jquanti.asm - sample data conversion and quantization (64-bit SSE2) +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2009, 2016, 2024, D. R. Commander. +; Copyright (C) 2018, Matthias Räncker. +; +; Based on the x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" +%include "jdct.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Load data into workspace, applying unsigned->signed conversion +; +; GLOBAL(void) +; jsimd_convsamp_sse2(JSAMPARRAY sample_data, JDIMENSION start_col, +; DCTELEM *workspace); +; + +; r10 = JSAMPARRAY sample_data +; r11d = JDIMENSION start_col +; r12 = DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_convsamp_sse2) + +EXTN(jsimd_convsamp_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 3 + push rbx + + pxor xmm6, xmm6 ; xmm6=(all 0's) + pcmpeqw xmm7, xmm7 + psllw xmm7, 7 ; xmm7={0xFF80 0xFF80 0xFF80 0xFF80 ..} + + mov rsi, r10 + mov eax, r11d + mov rdi, r12 + mov rcx, DCTSIZE/4 +.convloop: + mov rbxp, JSAMPROW [rsi+0*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rdxp, JSAMPROW [rsi+1*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq xmm0, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] ; xmm0=(01234567) + movq xmm1, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] ; xmm1=(89ABCDEF) + + mov rbxp, JSAMPROW [rsi+2*SIZEOF_JSAMPROW] ; (JSAMPLE *) + mov rdxp, JSAMPROW [rsi+3*SIZEOF_JSAMPROW] ; (JSAMPLE *) + + movq xmm2, XMM_MMWORD [rbx+rax*SIZEOF_JSAMPLE] ; xmm2=(GHIJKLMN) + movq xmm3, XMM_MMWORD [rdx+rax*SIZEOF_JSAMPLE] ; xmm3=(OPQRSTUV) + + punpcklbw xmm0, xmm6 ; xmm0=(01234567) + punpcklbw xmm1, xmm6 ; xmm1=(89ABCDEF) + paddw xmm0, xmm7 + paddw xmm1, xmm7 + punpcklbw xmm2, xmm6 ; xmm2=(GHIJKLMN) + punpcklbw xmm3, xmm6 ; xmm3=(OPQRSTUV) + paddw xmm2, xmm7 + paddw xmm3, xmm7 + + movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0 + movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1 + movdqa XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2 + movdqa XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3 + + add rsi, byte 4*SIZEOF_JSAMPROW + add rdi, byte 4*DCTSIZE*SIZEOF_DCTELEM + dec rcx + jnz short .convloop + + pop rbx + UNCOLLECT_ARGS 3 + pop rbp + ret + +; -------------------------------------------------------------------------- +; +; Quantize/descale the coefficients, and store into coef_block +; +; This implementation is based on an algorithm described in +; "Optimizing subroutines in assembly language: +; An optimization guide for x86 platforms" (https://agner.org/optimize). +; +; GLOBAL(void) +; jsimd_quantize_sse2(JCOEFPTR coef_block, DCTELEM *divisors, +; DCTELEM *workspace); +; + +%define RECIPROCAL(m, n, b) \ + XMMBLOCK(DCTSIZE * 0 + (m), (n), (b), SIZEOF_DCTELEM) +%define CORRECTION(m, n, b) \ + XMMBLOCK(DCTSIZE * 1 + (m), (n), (b), SIZEOF_DCTELEM) +%define SCALE(m, n, b) \ + XMMBLOCK(DCTSIZE * 2 + (m), (n), (b), SIZEOF_DCTELEM) + +; r10 = JCOEFPTR coef_block +; r11 = DCTELEM *divisors +; r12 = DCTELEM *workspace + + align 32 + GLOBAL_FUNCTION(jsimd_quantize_sse2) + +EXTN(jsimd_quantize_sse2): + ENDBR64 + push rbp + mov rbp, rsp + COLLECT_ARGS 3 + + mov rsi, r12 + mov rdx, r11 + mov rdi, r10 + mov rax, DCTSIZE2/32 +.quantloop: + movdqa xmm4, XMMWORD [XMMBLOCK(0,0,rsi,SIZEOF_DCTELEM)] + movdqa xmm5, XMMWORD [XMMBLOCK(1,0,rsi,SIZEOF_DCTELEM)] + movdqa xmm6, XMMWORD [XMMBLOCK(2,0,rsi,SIZEOF_DCTELEM)] + movdqa xmm7, XMMWORD [XMMBLOCK(3,0,rsi,SIZEOF_DCTELEM)] + movdqa xmm0, xmm4 + movdqa xmm1, xmm5 + movdqa xmm2, xmm6 + movdqa xmm3, xmm7 + psraw xmm4, (WORD_BIT-1) + psraw xmm5, (WORD_BIT-1) + psraw xmm6, (WORD_BIT-1) + psraw xmm7, (WORD_BIT-1) + pxor xmm0, xmm4 + pxor xmm1, xmm5 + pxor xmm2, xmm6 + pxor xmm3, xmm7 + psubw xmm0, xmm4 ; if (xmm0 < 0) xmm0 = -xmm0; + psubw xmm1, xmm5 ; if (xmm1 < 0) xmm1 = -xmm1; + psubw xmm2, xmm6 ; if (xmm2 < 0) xmm2 = -xmm2; + psubw xmm3, xmm7 ; if (xmm3 < 0) xmm3 = -xmm3; + + paddw xmm0, XMMWORD [CORRECTION(0,0,rdx)] ; correction + roundfactor + paddw xmm1, XMMWORD [CORRECTION(1,0,rdx)] + paddw xmm2, XMMWORD [CORRECTION(2,0,rdx)] + paddw xmm3, XMMWORD [CORRECTION(3,0,rdx)] + pmulhuw xmm0, XMMWORD [RECIPROCAL(0,0,rdx)] ; reciprocal + pmulhuw xmm1, XMMWORD [RECIPROCAL(1,0,rdx)] + pmulhuw xmm2, XMMWORD [RECIPROCAL(2,0,rdx)] + pmulhuw xmm3, XMMWORD [RECIPROCAL(3,0,rdx)] + pmulhuw xmm0, XMMWORD [SCALE(0,0,rdx)] ; scale + pmulhuw xmm1, XMMWORD [SCALE(1,0,rdx)] + pmulhuw xmm2, XMMWORD [SCALE(2,0,rdx)] + pmulhuw xmm3, XMMWORD [SCALE(3,0,rdx)] + + pxor xmm0, xmm4 + pxor xmm1, xmm5 + pxor xmm2, xmm6 + pxor xmm3, xmm7 + psubw xmm0, xmm4 + psubw xmm1, xmm5 + psubw xmm2, xmm6 + psubw xmm3, xmm7 + movdqa XMMWORD [XMMBLOCK(0,0,rdi,SIZEOF_DCTELEM)], xmm0 + movdqa XMMWORD [XMMBLOCK(1,0,rdi,SIZEOF_DCTELEM)], xmm1 + movdqa XMMWORD [XMMBLOCK(2,0,rdi,SIZEOF_DCTELEM)], xmm2 + movdqa XMMWORD [XMMBLOCK(3,0,rdi,SIZEOF_DCTELEM)], xmm3 + + add rsi, byte 32*SIZEOF_DCTELEM + add rdx, byte 32*SIZEOF_DCTELEM + add rdi, byte 32*SIZEOF_JCOEF + dec rax + jnz near .quantloop + + UNCOLLECT_ARGS 3 + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimd.c b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimd.c new file mode 100644 index 0000000..3f5ee77 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimd.c @@ -0,0 +1,1110 @@ +/* + * jsimd_x86_64.c + * + * Copyright 2009 Pierre Ossman for Cendio AB + * Copyright (C) 2009-2011, 2014, 2016, 2018, 2022-2023, D. R. Commander. + * Copyright (C) 2015-2016, 2018, 2022, Matthieu Darbois. + * + * Based on the x86 SIMD extension for IJG JPEG library, + * Copyright (C) 1999-2006, MIYASAKA Masaru. + * For conditions of distribution and use, see copyright notice in jsimdext.inc + * + * This file contains the interface between the "normal" portions + * of the library and the SIMD implementations when running on a + * 64-bit x86 architecture. + */ + +#define JPEG_INTERNALS +#include "../../jinclude.h" +#include "../../jpeglib.h" +#include "../../jsimd.h" +#include "../../jdct.h" +#include "../../jsimddct.h" +#include "../jsimd.h" + +/* + * In the PIC cases, we have no guarantee that constants will keep + * their alignment. This macro allows us to verify it at runtime. + */ +#define IS_ALIGNED(ptr, order) (((size_t)ptr & ((1 << order) - 1)) == 0) + +#define IS_ALIGNED_SSE(ptr) (IS_ALIGNED(ptr, 4)) /* 16 byte alignment */ +#define IS_ALIGNED_AVX(ptr) (IS_ALIGNED(ptr, 5)) /* 32 byte alignment */ + +static THREAD_LOCAL unsigned int simd_support = (unsigned int)(~0); +static THREAD_LOCAL unsigned int simd_huffman = 1; + +/* + * Check what SIMD accelerations are supported. + */ +LOCAL(void) +init_simd(void) +{ +#ifndef NO_GETENV + char env[2] = { 0 }; +#endif + + if (simd_support != ~0U) + return; + + simd_support = jpeg_simd_cpu_support(); + +#ifndef NO_GETENV + /* Force different settings through environment variables */ + if (!GETENV_S(env, 2, "JSIMD_FORCESSE2") && !strcmp(env, "1")) + simd_support &= JSIMD_SSE2; + if (!GETENV_S(env, 2, "JSIMD_FORCEAVX2") && !strcmp(env, "1")) + simd_support &= JSIMD_AVX2; + if (!GETENV_S(env, 2, "JSIMD_FORCENONE") && !strcmp(env, "1")) + simd_support = 0; + if (!GETENV_S(env, 2, "JSIMD_NOHUFFENC") && !strcmp(env, "1")) + simd_huffman = 0; +#endif +} + +GLOBAL(int) +jsimd_can_rgb_ycc(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_rgb_ycc_convert_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_rgb_ycc_convert_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_rgb_gray(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_rgb_gray_convert_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_rgb_gray_convert_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_ycc_rgb(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if ((RGB_PIXELSIZE != 3) && (RGB_PIXELSIZE != 4)) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_ycc_rgb_convert_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_ycc_rgb_convert_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_ycc_rgb565(void) +{ + return 0; +} + +GLOBAL(void) +jsimd_rgb_ycc_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows) +{ + void (*avx2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + void (*sse2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->in_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_extrgb_ycc_convert_avx2; + sse2fct = jsimd_extrgb_ycc_convert_sse2; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_extrgbx_ycc_convert_avx2; + sse2fct = jsimd_extrgbx_ycc_convert_sse2; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_extbgr_ycc_convert_avx2; + sse2fct = jsimd_extbgr_ycc_convert_sse2; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_extbgrx_ycc_convert_avx2; + sse2fct = jsimd_extbgrx_ycc_convert_sse2; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_extxbgr_ycc_convert_avx2; + sse2fct = jsimd_extxbgr_ycc_convert_sse2; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_extxrgb_ycc_convert_avx2; + sse2fct = jsimd_extxrgb_ycc_convert_sse2; + break; + default: + avx2fct = jsimd_rgb_ycc_convert_avx2; + sse2fct = jsimd_rgb_ycc_convert_sse2; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); + else + sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); +} + +GLOBAL(void) +jsimd_rgb_gray_convert(j_compress_ptr cinfo, JSAMPARRAY input_buf, + JSAMPIMAGE output_buf, JDIMENSION output_row, + int num_rows) +{ + void (*avx2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + void (*sse2fct) (JDIMENSION, JSAMPARRAY, JSAMPIMAGE, JDIMENSION, int); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->in_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_extrgb_gray_convert_avx2; + sse2fct = jsimd_extrgb_gray_convert_sse2; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_extrgbx_gray_convert_avx2; + sse2fct = jsimd_extrgbx_gray_convert_sse2; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_extbgr_gray_convert_avx2; + sse2fct = jsimd_extbgr_gray_convert_sse2; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_extbgrx_gray_convert_avx2; + sse2fct = jsimd_extbgrx_gray_convert_sse2; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_extxbgr_gray_convert_avx2; + sse2fct = jsimd_extxbgr_gray_convert_sse2; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_extxrgb_gray_convert_avx2; + sse2fct = jsimd_extxrgb_gray_convert_sse2; + break; + default: + avx2fct = jsimd_rgb_gray_convert_avx2; + sse2fct = jsimd_rgb_gray_convert_sse2; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); + else + sse2fct(cinfo->image_width, input_buf, output_buf, output_row, num_rows); +} + +GLOBAL(void) +jsimd_ycc_rgb_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION input_row, JSAMPARRAY output_buf, + int num_rows) +{ + void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int); + void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY, int); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_ycc_extrgb_convert_avx2; + sse2fct = jsimd_ycc_extrgb_convert_sse2; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_ycc_extrgbx_convert_avx2; + sse2fct = jsimd_ycc_extrgbx_convert_sse2; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_ycc_extbgr_convert_avx2; + sse2fct = jsimd_ycc_extbgr_convert_sse2; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_ycc_extbgrx_convert_avx2; + sse2fct = jsimd_ycc_extbgrx_convert_sse2; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_ycc_extxbgr_convert_avx2; + sse2fct = jsimd_ycc_extxbgr_convert_sse2; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_ycc_extxrgb_convert_avx2; + sse2fct = jsimd_ycc_extxrgb_convert_sse2; + break; + default: + avx2fct = jsimd_ycc_rgb_convert_avx2; + sse2fct = jsimd_ycc_rgb_convert_sse2; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows); + else + sse2fct(cinfo->output_width, input_buf, input_row, output_buf, num_rows); +} + +GLOBAL(void) +jsimd_ycc_rgb565_convert(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION input_row, JSAMPARRAY output_buf, + int num_rows) +{ +} + +GLOBAL(int) +jsimd_can_h2v2_downsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_downsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY output_data) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v2_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); + else + jsimd_h2v2_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); +} + +GLOBAL(void) +jsimd_h2v1_downsample(j_compress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY output_data) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v1_downsample_avx2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); + else + jsimd_h2v1_downsample_sse2(cinfo->image_width, cinfo->max_v_samp_factor, + compptr->v_samp_factor, + compptr->width_in_blocks, input_data, + output_data); +} + +GLOBAL(int) +jsimd_can_h2v2_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v2_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); + else + jsimd_h2v2_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); +} + +GLOBAL(void) +jsimd_h2v1_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v1_upsample_avx2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); + else + jsimd_h2v1_upsample_sse2(cinfo->max_v_samp_factor, cinfo->output_width, + input_data, output_data_ptr); +} + +GLOBAL(int) +jsimd_can_h2v2_fancy_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_fancy_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_fancy_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_fancy_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_fancy_upsample_sse2)) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v2_fancy_upsample_avx2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); + else + jsimd_h2v2_fancy_upsample_sse2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); +} + +GLOBAL(void) +jsimd_h2v1_fancy_upsample(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JSAMPARRAY input_data, JSAMPARRAY *output_data_ptr) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_h2v1_fancy_upsample_avx2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); + else + jsimd_h2v1_fancy_upsample_sse2(cinfo->max_v_samp_factor, + compptr->downsampled_width, input_data, + output_data_ptr); +} + +GLOBAL(int) +jsimd_can_h2v2_merged_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_merged_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_h2v1_merged_upsample(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + + if ((simd_support & JSIMD_AVX2) && + IS_ALIGNED_AVX(jconst_merged_upsample_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && + IS_ALIGNED_SSE(jconst_merged_upsample_sse2)) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_h2v2_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf) +{ + void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_h2v2_extrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extrgb_merged_upsample_sse2; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_h2v2_extrgbx_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extrgbx_merged_upsample_sse2; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_h2v2_extbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extbgr_merged_upsample_sse2; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_h2v2_extbgrx_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extbgrx_merged_upsample_sse2; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_h2v2_extxbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extxbgr_merged_upsample_sse2; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_h2v2_extxrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v2_extxrgb_merged_upsample_sse2; + break; + default: + avx2fct = jsimd_h2v2_merged_upsample_avx2; + sse2fct = jsimd_h2v2_merged_upsample_sse2; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); + else + sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); +} + +GLOBAL(void) +jsimd_h2v1_merged_upsample(j_decompress_ptr cinfo, JSAMPIMAGE input_buf, + JDIMENSION in_row_group_ctr, JSAMPARRAY output_buf) +{ + void (*avx2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + void (*sse2fct) (JDIMENSION, JSAMPIMAGE, JDIMENSION, JSAMPARRAY); + + if (simd_support == ~0U) + init_simd(); + + switch (cinfo->out_color_space) { + case JCS_EXT_RGB: + avx2fct = jsimd_h2v1_extrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extrgb_merged_upsample_sse2; + break; + case JCS_EXT_RGBX: + case JCS_EXT_RGBA: + avx2fct = jsimd_h2v1_extrgbx_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extrgbx_merged_upsample_sse2; + break; + case JCS_EXT_BGR: + avx2fct = jsimd_h2v1_extbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extbgr_merged_upsample_sse2; + break; + case JCS_EXT_BGRX: + case JCS_EXT_BGRA: + avx2fct = jsimd_h2v1_extbgrx_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extbgrx_merged_upsample_sse2; + break; + case JCS_EXT_XBGR: + case JCS_EXT_ABGR: + avx2fct = jsimd_h2v1_extxbgr_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extxbgr_merged_upsample_sse2; + break; + case JCS_EXT_XRGB: + case JCS_EXT_ARGB: + avx2fct = jsimd_h2v1_extxrgb_merged_upsample_avx2; + sse2fct = jsimd_h2v1_extxrgb_merged_upsample_sse2; + break; + default: + avx2fct = jsimd_h2v1_merged_upsample_avx2; + sse2fct = jsimd_h2v1_merged_upsample_sse2; + break; + } + + if (simd_support & JSIMD_AVX2) + avx2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); + else + sse2fct(cinfo->output_width, input_buf, in_row_group_ctr, output_buf); +} + +GLOBAL(int) +jsimd_can_convsamp(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_convsamp_float(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_convsamp(JSAMPARRAY sample_data, JDIMENSION start_col, + DCTELEM *workspace) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_convsamp_avx2(sample_data, start_col, workspace); + else + jsimd_convsamp_sse2(sample_data, start_col, workspace); +} + +GLOBAL(void) +jsimd_convsamp_float(JSAMPARRAY sample_data, JDIMENSION start_col, + FAST_FLOAT *workspace) +{ + jsimd_convsamp_float_sse2(sample_data, start_col, workspace); +} + +GLOBAL(int) +jsimd_can_fdct_islow(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if ((simd_support & JSIMD_AVX2) && IS_ALIGNED_AVX(jconst_fdct_islow_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_islow_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_fdct_ifast(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_fdct_ifast_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_fdct_float(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + + if ((simd_support & JSIMD_SSE) && IS_ALIGNED_SSE(jconst_fdct_float_sse)) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_fdct_islow(DCTELEM *data) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_fdct_islow_avx2(data); + else + jsimd_fdct_islow_sse2(data); +} + +GLOBAL(void) +jsimd_fdct_ifast(DCTELEM *data) +{ + jsimd_fdct_ifast_sse2(data); +} + +GLOBAL(void) +jsimd_fdct_float(FAST_FLOAT *data) +{ + jsimd_fdct_float_sse(data); +} + +GLOBAL(int) +jsimd_can_quantize(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (sizeof(DCTELEM) != 2) + return 0; + + if (simd_support & JSIMD_AVX2) + return 1; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_quantize_float(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_quantize(JCOEFPTR coef_block, DCTELEM *divisors, DCTELEM *workspace) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_quantize_avx2(coef_block, divisors, workspace); + else + jsimd_quantize_sse2(coef_block, divisors, workspace); +} + +GLOBAL(void) +jsimd_quantize_float(JCOEFPTR coef_block, FAST_FLOAT *divisors, + FAST_FLOAT *workspace) +{ + jsimd_quantize_float_sse2(coef_block, divisors, workspace); +} + +GLOBAL(int) +jsimd_can_idct_2x2(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(ISLOW_MULT_TYPE) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_idct_4x4(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(ISLOW_MULT_TYPE) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_red_sse2)) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_idct_2x2(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + jsimd_idct_2x2_sse2(compptr->dct_table, coef_block, output_buf, output_col); +} + +GLOBAL(void) +jsimd_idct_4x4(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + jsimd_idct_4x4_sse2(compptr->dct_table, coef_block, output_buf, output_col); +} + +GLOBAL(int) +jsimd_can_idct_islow(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(ISLOW_MULT_TYPE) != 2) + return 0; + + if ((simd_support & JSIMD_AVX2) && IS_ALIGNED_AVX(jconst_idct_islow_avx2)) + return 1; + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_islow_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_idct_ifast(void) +{ + init_simd(); + + /* The code is optimised for these values only */ + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(IFAST_MULT_TYPE) != 2) + return 0; + if (IFAST_SCALE_BITS != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_ifast_sse2)) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_can_idct_float(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (BITS_IN_JSAMPLE != 8) + return 0; + if (sizeof(JDIMENSION) != 4) + return 0; + if (sizeof(FAST_FLOAT) != 4) + return 0; + if (sizeof(FLOAT_MULT_TYPE) != 4) + return 0; + + if ((simd_support & JSIMD_SSE2) && IS_ALIGNED_SSE(jconst_idct_float_sse2)) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_idct_islow(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + if (simd_support == ~0U) + init_simd(); + + if (simd_support & JSIMD_AVX2) + jsimd_idct_islow_avx2(compptr->dct_table, coef_block, output_buf, + output_col); + else + jsimd_idct_islow_sse2(compptr->dct_table, coef_block, output_buf, + output_col); +} + +GLOBAL(void) +jsimd_idct_ifast(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + jsimd_idct_ifast_sse2(compptr->dct_table, coef_block, output_buf, + output_col); +} + +GLOBAL(void) +jsimd_idct_float(j_decompress_ptr cinfo, jpeg_component_info *compptr, + JCOEFPTR coef_block, JSAMPARRAY output_buf, + JDIMENSION output_col) +{ + jsimd_idct_float_sse2(compptr->dct_table, coef_block, output_buf, + output_col); +} + +GLOBAL(int) +jsimd_can_huff_encode_one_block(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + + if ((simd_support & JSIMD_SSE2) && simd_huffman && + IS_ALIGNED_SSE(jconst_huff_encode_one_block)) + return 1; + + return 0; +} + +GLOBAL(JOCTET *) +jsimd_huff_encode_one_block(void *state, JOCTET *buffer, JCOEFPTR block, + int last_dc_val, c_derived_tbl *dctbl, + c_derived_tbl *actbl) +{ + return jsimd_huff_encode_one_block_sse2(state, buffer, block, last_dc_val, + dctbl, actbl); +} + +GLOBAL(int) +jsimd_can_encode_mcu_AC_first_prepare(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(void) +jsimd_encode_mcu_AC_first_prepare(const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *values, size_t *zerobits) +{ + jsimd_encode_mcu_AC_first_prepare_sse2(block, jpeg_natural_order_start, + Sl, Al, values, zerobits); +} + +GLOBAL(int) +jsimd_can_encode_mcu_AC_refine_prepare(void) +{ + init_simd(); + + if (DCTSIZE != 8) + return 0; + if (sizeof(JCOEF) != 2) + return 0; + if (simd_support & JSIMD_SSE2) + return 1; + + return 0; +} + +GLOBAL(int) +jsimd_encode_mcu_AC_refine_prepare(const JCOEF *block, + const int *jpeg_natural_order_start, int Sl, + int Al, UJCOEF *absvalues, size_t *bits) +{ + return jsimd_encode_mcu_AC_refine_prepare_sse2(block, + jpeg_natural_order_start, + Sl, Al, absvalues, bits); +} diff --git a/src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimdcpu.asm b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimdcpu.asm new file mode 100644 index 0000000..b72f3b0 --- /dev/null +++ b/src/spice2x/external/libjpeg-turbo/simd/x86_64/jsimdcpu.asm @@ -0,0 +1,86 @@ +; +; jsimdcpu.asm - SIMD instruction support check +; +; Copyright 2009 Pierre Ossman for Cendio AB +; Copyright (C) 2016, D. R. Commander. +; Copyright (C) 2023, Aliaksiej Kandracienka. +; +; Based on +; x86 SIMD extension for IJG JPEG library +; Copyright (C) 1999-2006, MIYASAKA Masaru. +; For conditions of distribution and use, see copyright notice in jsimdext.inc +; +; This file should be assembled with NASM (Netwide Assembler) or Yasm. + +%include "jsimdext.inc" + +; -------------------------------------------------------------------------- + SECTION SEG_TEXT + BITS 64 +; +; Check if the CPU supports SIMD instructions +; +; GLOBAL(unsigned int) +; jpeg_simd_cpu_support(void) +; + + align 32 + GLOBAL_FUNCTION(jpeg_simd_cpu_support) + +EXTN(jpeg_simd_cpu_support): + push rbp + mov rbp, rsp + push rbx + push rdi + + xor rdi, rdi ; simd support flag + + ; Assume that all x86-64 processors support SSE & SSE2 instructions + or rdi, JSIMD_SSE2 + or rdi, JSIMD_SSE + + ; Check whether CPUID leaf 07H is supported + ; (leaf 07H is used to check for AVX2 instruction support) + mov rax, 0 + cpuid + cmp rax, 7 + jl short .return ; Maximum leaf < 07H + + ; Check for AVX2 instruction support + mov rax, 7 + xor rcx, rcx + cpuid + mov rax, rbx ; rax = Extended feature flags + + test rax, 1<<5 ; bit5:AVX2 + jz short .return + + ; Check for AVX2 O/S support + mov rax, 1 + xor rcx, rcx + cpuid + test rcx, 1<<27 + jz short .return ; O/S does not support XSAVE + test rcx, 1<<28 + jz short .return ; CPU does not support AVX2 + + xor rcx, rcx + xgetbv + and rax, 6 + cmp rax, 6 ; O/S does not manage XMM/YMM state + ; using XSAVE + jnz short .return + + or rdi, JSIMD_AVX2 + +.return: + mov rax, rdi + + pop rdi + pop rbx + pop rbp + ret + +; For some reason, the OS X linker does not honor the request to align the +; segment unless we do this. + align 32 diff --git a/src/spice2x/external/toojpeg/LICENSE b/src/spice2x/external/toojpeg/LICENSE deleted file mode 100644 index 76d0362..0000000 --- a/src/spice2x/external/toojpeg/LICENSE +++ /dev/null @@ -1,10 +0,0 @@ -zlib License - -Copyright (c) 2011-2016 Stephan Brumme - -This software is provided 'as-is', without any express or implied warranty. In no event will the authors be held liable for any damages arising from the use of this software. -Permission is granted to anyone to use this software for any purpose, including commercial applications, and to alter it and redistribute it freely, subject to the following restrictions: -1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. - If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required. -2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software. -3. This notice may not be removed or altered from any source distribution. diff --git a/src/spice2x/external/toojpeg/toojpeg.cpp b/src/spice2x/external/toojpeg/toojpeg.cpp deleted file mode 100644 index 866072c..0000000 --- a/src/spice2x/external/toojpeg/toojpeg.cpp +++ /dev/null @@ -1,665 +0,0 @@ -// ////////////////////////////////////////////////////////// -// toojpeg.cpp -// written by Stephan Brumme, 2018-2019 -// see https://create.stephan-brumme.com/toojpeg/ -// - -#include "toojpeg.h" - -// - the "official" specifications: https://www.w3.org/Graphics/JPEG/itu-t81.pdf and https://www.w3.org/Graphics/JPEG/jfif3.pdf -// - Wikipedia has a short description of the JFIF/JPEG file format: https://en.wikipedia.org/wiki/JPEG_File_Interchange_Format -// - the popular STB Image library includes Jon's JPEG encoder as well: https://github.com/nothings/stb/blob/master/stb_image_write.h -// - the most readable JPEG book (from a developer's perspective) is Miano's "Compressed Image File Formats" (1999, ISBN 0-201-60443-4), -// used copies are really cheap nowadays and include a CD with C++ sources as well (plus great format descriptions of GIF & PNG) -// - much more detailled is Mitchell/Pennebaker's "JPEG: Still Image Data Compression Standard" (1993, ISBN 0-442-01272-1) -// which contains the official JPEG standard, too - fun fact: I bought a signed copy in a second-hand store without noticing - -namespace // anonymous namespace to hide local functions / constants / etc. -{ -// //////////////////////////////////////// -// data types -using uint8_t = unsigned char; -using uint16_t = unsigned short; -using int16_t = short; -using int32_t = int; // at least four bytes - -// //////////////////////////////////////// -// constants - -// quantization tables from JPEG Standard, Annex K -const uint8_t DefaultQuantLuminance[8*8] = - { 16, 11, 10, 16, 24, 40, 51, 61, // there are a few experts proposing slightly more efficient values, - 12, 12, 14, 19, 26, 58, 60, 55, // e.g. https://www.imagemagick.org/discourse-server/viewtopic.php?t=20333 - 14, 13, 16, 24, 40, 57, 69, 56, // btw: Google's Guetzli project optimizes the quantization tables per image - 14, 17, 22, 29, 51, 87, 80, 62, - 18, 22, 37, 56, 68,109,103, 77, - 24, 35, 55, 64, 81,104,113, 92, - 49, 64, 78, 87,103,121,120,101, - 72, 92, 95, 98,112,100,103, 99 }; -const uint8_t DefaultQuantChrominance[8*8] = - { 17, 18, 24, 47, 99, 99, 99, 99, - 18, 21, 26, 66, 99, 99, 99, 99, - 24, 26, 56, 99, 99, 99, 99, 99, - 47, 66, 99, 99, 99, 99, 99, 99, - 99, 99, 99, 99, 99, 99, 99, 99, - 99, 99, 99, 99, 99, 99, 99, 99, - 99, 99, 99, 99, 99, 99, 99, 99, - 99, 99, 99, 99, 99, 99, 99, 99 }; - -// 8x8 blocks are processed in zig-zag order -// most encoders use a zig-zag "forward" table, I switched to its inverse for performance reasons -// note: ZigZagInv[ZigZag[i]] = i -const uint8_t ZigZagInv[8*8] = - { 0, 1, 8,16, 9, 2, 3,10, // ZigZag[] = 0, 1, 5, 6,14,15,27,28, - 17,24,32,25,18,11, 4, 5, // 2, 4, 7,13,16,26,29,42, - 12,19,26,33,40,48,41,34, // 3, 8,12,17,25,30,41,43, - 27,20,13, 6, 7,14,21,28, // 9,11,18,24,31,40,44,53, - 35,42,49,56,57,50,43,36, // 10,19,23,32,39,45,52,54, - 29,22,15,23,30,37,44,51, // 20,22,33,38,46,51,55,60, - 58,59,52,45,38,31,39,46, // 21,34,37,47,50,56,59,61, - 53,60,61,54,47,55,62,63 }; // 35,36,48,49,57,58,62,63 - -// static Huffman code tables from JPEG standard Annex K -// - CodesPerBitsize tables define how many Huffman codes will have a certain bitsize (plus 1 because there nothing with zero bits), -// e.g. DcLuminanceCodesPerBitsize[2] = 5 because there are 5 Huffman codes being 2+1=3 bits long -// - Values tables are a list of values ordered by their Huffman code bitsize, -// e.g. AcLuminanceValues => Huffman(0x01,0x02 and 0x03) will have 2 bits, Huffman(0x00) will have 3 bits, Huffman(0x04,0x11 and 0x05) will have 4 bits, ... - -// Huffman definitions for first DC/AC tables (luminance / Y channel) -const uint8_t DcLuminanceCodesPerBitsize[16] = { 0,1,5,1,1,1,1,1,1,0,0,0,0,0,0,0 }; // sum = 12 -const uint8_t DcLuminanceValues [12] = { 0,1,2,3,4,5,6,7,8,9,10,11 }; // => 12 codes -const uint8_t AcLuminanceCodesPerBitsize[16] = { 0,2,1,3,3,2,4,3,5,5,4,4,0,0,1,125 }; // sum = 162 -const uint8_t AcLuminanceValues [162] = // => 162 codes - { 0x01,0x02,0x03,0x00,0x04,0x11,0x05,0x12,0x21,0x31,0x41,0x06,0x13,0x51,0x61,0x07,0x22,0x71,0x14,0x32,0x81,0x91,0xA1,0x08, // 16*10+2 symbols because - 0x23,0x42,0xB1,0xC1,0x15,0x52,0xD1,0xF0,0x24,0x33,0x62,0x72,0x82,0x09,0x0A,0x16,0x17,0x18,0x19,0x1A,0x25,0x26,0x27,0x28, // upper 4 bits can be 0..F - 0x29,0x2A,0x34,0x35,0x36,0x37,0x38,0x39,0x3A,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4A,0x53,0x54,0x55,0x56,0x57,0x58,0x59, // while lower 4 bits can be 1..A - 0x5A,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6A,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7A,0x83,0x84,0x85,0x86,0x87,0x88,0x89, // plus two special codes 0x00 and 0xF0 - 0x8A,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9A,0xA2,0xA3,0xA4,0xA5,0xA6,0xA7,0xA8,0xA9,0xAA,0xB2,0xB3,0xB4,0xB5,0xB6, // order of these symbols was determined empirically by JPEG committee - 0xB7,0xB8,0xB9,0xBA,0xC2,0xC3,0xC4,0xC5,0xC6,0xC7,0xC8,0xC9,0xCA,0xD2,0xD3,0xD4,0xD5,0xD6,0xD7,0xD8,0xD9,0xDA,0xE1,0xE2, - 0xE3,0xE4,0xE5,0xE6,0xE7,0xE8,0xE9,0xEA,0xF1,0xF2,0xF3,0xF4,0xF5,0xF6,0xF7,0xF8,0xF9,0xFA }; -// Huffman definitions for second DC/AC tables (chrominance / Cb and Cr channels) -const uint8_t DcChrominanceCodesPerBitsize[16] = { 0,3,1,1,1,1,1,1,1,1,1,0,0,0,0,0 }; // sum = 12 -const uint8_t DcChrominanceValues [12] = { 0,1,2,3,4,5,6,7,8,9,10,11 }; // => 12 codes (identical to DcLuminanceValues) -const uint8_t AcChrominanceCodesPerBitsize[16] = { 0,2,1,2,4,4,3,4,7,5,4,4,0,1,2,119 }; // sum = 162 -const uint8_t AcChrominanceValues [162] = // => 162 codes - { 0x00,0x01,0x02,0x03,0x11,0x04,0x05,0x21,0x31,0x06,0x12,0x41,0x51,0x07,0x61,0x71,0x13,0x22,0x32,0x81,0x08,0x14,0x42,0x91, // same number of symbol, just different order - 0xA1,0xB1,0xC1,0x09,0x23,0x33,0x52,0xF0,0x15,0x62,0x72,0xD1,0x0A,0x16,0x24,0x34,0xE1,0x25,0xF1,0x17,0x18,0x19,0x1A,0x26, // (which is more efficient for AC coding) - 0x27,0x28,0x29,0x2A,0x35,0x36,0x37,0x38,0x39,0x3A,0x43,0x44,0x45,0x46,0x47,0x48,0x49,0x4A,0x53,0x54,0x55,0x56,0x57,0x58, - 0x59,0x5A,0x63,0x64,0x65,0x66,0x67,0x68,0x69,0x6A,0x73,0x74,0x75,0x76,0x77,0x78,0x79,0x7A,0x82,0x83,0x84,0x85,0x86,0x87, - 0x88,0x89,0x8A,0x92,0x93,0x94,0x95,0x96,0x97,0x98,0x99,0x9A,0xA2,0xA3,0xA4,0xA5,0xA6,0xA7,0xA8,0xA9,0xAA,0xB2,0xB3,0xB4, - 0xB5,0xB6,0xB7,0xB8,0xB9,0xBA,0xC2,0xC3,0xC4,0xC5,0xC6,0xC7,0xC8,0xC9,0xCA,0xD2,0xD3,0xD4,0xD5,0xD6,0xD7,0xD8,0xD9,0xDA, - 0xE2,0xE3,0xE4,0xE5,0xE6,0xE7,0xE8,0xE9,0xEA,0xF2,0xF3,0xF4,0xF5,0xF6,0xF7,0xF8,0xF9,0xFA }; -const int16_t CodeWordLimit = 2048; // +/-2^11, maximum value after DCT - -// //////////////////////////////////////// -// structs - -// represent a single Huffman code -struct BitCode -{ - BitCode() = default; // undefined state, must be initialized at a later time - BitCode(uint16_t code_, uint8_t numBits_) - : code(code_), numBits(numBits_) {} - uint16_t code; // JPEG's Huffman codes are limited to 16 bits - uint8_t numBits; // number of valid bits -}; - -// wrapper for bit output operations -struct BitWriter -{ - // user-supplied callback that writes/stores one byte - TooJpeg::WRITE_ONE_BYTE output; - // initialize writer - explicit BitWriter(TooJpeg::WRITE_ONE_BYTE output_) : output(output_) {} - - // store the most recently encoded bits that are not written yet - struct BitBuffer - { - int32_t data = 0; // actually only at most 24 bits are used - uint8_t numBits = 0; // number of valid bits (the right-most bits) - } buffer; - - // write Huffman bits stored in BitCode, keep excess bits in BitBuffer - BitWriter& operator<<(const BitCode& data) - { - // append the new bits to those bits leftover from previous call(s) - buffer.numBits += data.numBits; - buffer.data <<= data.numBits; - buffer.data |= data.code; - - // write all "full" bytes - while (buffer.numBits >= 8) - { - // extract highest 8 bits - buffer.numBits -= 8; - auto oneByte = uint8_t(buffer.data >> buffer.numBits); - output(oneByte); - - if (oneByte == 0xFF) // 0xFF has a special meaning for JPEGs (it's a block marker) - output(0); // therefore pad a zero to indicate "nope, this one ain't a marker, it's just a coincidence" - - // note: I don't clear those written bits, therefore buffer.bits may contain garbage in the high bits - // if you really want to "clean up" (e.g. for debugging purposes) then uncomment the following line - //buffer.bits &= (1 << buffer.numBits) - 1; - } - return *this; - } - - // write all non-yet-written bits, fill gaps with 1s (that's a strange JPEG thing) - void flush() - { - // at most seven set bits needed to "fill" the last byte: 0x7F = binary 0111 1111 - *this << BitCode(0x7F, 7); // I should set buffer.numBits = 0 but since there are no single bits written after flush() I can safely ignore it - } - - // NOTE: all the following BitWriter functions IGNORE the BitBuffer and write straight to output ! - // write a single byte - BitWriter& operator<<(uint8_t oneByte) - { - output(oneByte); - return *this; - } - - // write an array of bytes - template - BitWriter& operator<<(T (&manyBytes)[Size]) - { - for (auto c : manyBytes) - output(c); - return *this; - } - - // start a new JFIF block - void addMarker(uint8_t id, uint16_t length) - { - output(0xFF); output(id); // ID, always preceded by 0xFF - output(uint8_t(length >> 8)); // length of the block (big-endian, includes the 2 length bytes as well) - output(uint8_t(length & 0xFF)); - } -}; - -// //////////////////////////////////////// -// functions / templates - -// same as std::min() -template -Number minimum(Number value, Number maximum) -{ - return value <= maximum ? value : maximum; -} - -// restrict a value to the interval [minimum, maximum] -template -Number clamp(Number value, Limit minValue, Limit maxValue) -{ - if (value <= minValue) return minValue; // never smaller than the minimum - if (value >= maxValue) return maxValue; // never bigger than the maximum - return value; // value was inside interval, keep it -} - -// convert from RGB to YCbCr, constants are similar to ITU-R, see https://en.wikipedia.org/wiki/YCbCr#JPEG_conversion -float rgb2y (float r, float g, float b) { return +0.299f * r +0.587f * g +0.114f * b; } -float rgb2cb(float r, float g, float b) { return -0.16874f * r -0.33126f * g +0.5f * b; } -float rgb2cr(float r, float g, float b) { return +0.5f * r -0.41869f * g -0.08131f * b; } - -// forward DCT computation "in one dimension" (fast AAN algorithm by Arai, Agui and Nakajima: "A fast DCT-SQ scheme for images") -void DCT(float block[8*8], uint8_t stride) // stride must be 1 (=horizontal) or 8 (=vertical) -{ - const auto SqrtHalfSqrt = 1.306562965f; // sqrt((2 + sqrt(2)) / 2) = cos(pi * 1 / 8) * sqrt(2) - const auto InvSqrt = 0.707106781f; // 1 / sqrt(2) = cos(pi * 2 / 8) - const auto HalfSqrtSqrt = 0.382683432f; // sqrt(2 - sqrt(2)) / 2 = cos(pi * 3 / 8) - const auto InvSqrtSqrt = 0.541196100f; // 1 / sqrt(2 - sqrt(2)) = cos(pi * 3 / 8) * sqrt(2) - - // modify in-place - auto& block0 = block[0 ]; - auto& block1 = block[1 * stride]; - auto& block2 = block[2 * stride]; - auto& block3 = block[3 * stride]; - auto& block4 = block[4 * stride]; - auto& block5 = block[5 * stride]; - auto& block6 = block[6 * stride]; - auto& block7 = block[7 * stride]; - - // based on https://dev.w3.org/Amaya/libjpeg/jfdctflt.c , the original variable names can be found in my comments - auto add07 = block0 + block7; auto sub07 = block0 - block7; // tmp0, tmp7 - auto add16 = block1 + block6; auto sub16 = block1 - block6; // tmp1, tmp6 - auto add25 = block2 + block5; auto sub25 = block2 - block5; // tmp2, tmp5 - auto add34 = block3 + block4; auto sub34 = block3 - block4; // tmp3, tmp4 - - auto add0347 = add07 + add34; auto sub07_34 = add07 - add34; // tmp10, tmp13 ("even part" / "phase 2") - auto add1256 = add16 + add25; auto sub16_25 = add16 - add25; // tmp11, tmp12 - - block0 = add0347 + add1256; block4 = add0347 - add1256; // "phase 3" - - auto z1 = (sub16_25 + sub07_34) * InvSqrt; // all temporary z-variables kept their original names - block2 = sub07_34 + z1; block6 = sub07_34 - z1; // "phase 5" - - auto sub23_45 = sub25 + sub34; // tmp10 ("odd part" / "phase 2") - auto sub12_56 = sub16 + sub25; // tmp11 - auto sub01_67 = sub16 + sub07; // tmp12 - - auto z5 = (sub23_45 - sub01_67) * HalfSqrtSqrt; - auto z2 = sub23_45 * InvSqrtSqrt + z5; - auto z3 = sub12_56 * InvSqrt; - auto z4 = sub01_67 * SqrtHalfSqrt + z5; - auto z6 = sub07 + z3; // z11 ("phase 5") - auto z7 = sub07 - z3; // z13 - block1 = z6 + z4; block7 = z6 - z4; // "phase 6" - block5 = z7 + z2; block3 = z7 - z2; -} - -// run DCT, quantize and write Huffman bit codes -int16_t encodeBlock(BitWriter& writer, float block[8][8], const float scaled[8*8], int16_t lastDC, - const BitCode huffmanDC[256], const BitCode huffmanAC[256], const BitCode* codewords) -{ - // "linearize" the 8x8 block, treat it as a flat array of 64 floats - auto block64 = (float*) block; - - // DCT: rows - for (auto offset = 0; offset < 8; offset++) - DCT(block64 + offset*8, 1); - // DCT: columns - for (auto offset = 0; offset < 8; offset++) - DCT(block64 + offset*1, 8); - - // scale - for (auto i = 0; i < 8*8; i++) - block64[i] *= scaled[i]; - - // encode DC (the first coefficient is the "average color" of the 8x8 block) - auto DC = int(block64[0] + (block64[0] >= 0 ? +0.5f : -0.5f)); // C++11's nearbyint() achieves a similar effect - - // quantize and zigzag the other 63 coefficients - auto posNonZero = 0; // find last coefficient which is not zero (because trailing zeros are encoded differently) - int16_t quantized[8*8]; - for (auto i = 1; i < 8*8; i++) // start at 1 because block64[0]=DC was already processed - { - auto value = block64[ZigZagInv[i]]; - // round to nearest integer - quantized[i] = int(value + (value >= 0 ? +0.5f : -0.5f)); // C++11's nearbyint() achieves a similar effect - // remember offset of last non-zero coefficient - if (quantized[i] != 0) - posNonZero = i; - } - - // same "average color" as previous block ? - auto diff = DC - lastDC; - if (diff == 0) - writer << huffmanDC[0x00]; // yes, write a special short symbol - else - { - auto bits = codewords[diff]; // nope, encode the difference to previous block's average color - writer << huffmanDC[bits.numBits] << bits; - } - - // encode ACs (quantized[1..63]) - auto offset = 0; // upper 4 bits count the number of consecutive zeros - for (auto i = 1; i <= posNonZero; i++) // quantized[0] was already written, skip all trailing zeros, too - { - // zeros are encoded in a special way - while (quantized[i] == 0) // found another zero ? - { - offset += 0x10; // add 1 to the upper 4 bits - // split into blocks of at most 16 consecutive zeros - if (offset > 0xF0) // remember, the counter is in the upper 4 bits, 0xF = 15 - { - writer << huffmanAC[0xF0]; // 0xF0 is a special code for "16 zeros" - offset = 0; - } - i++; - } - - auto encoded = codewords[quantized[i]]; - // combine number of zeros with the number of bits of the next non-zero value - writer << huffmanAC[offset + encoded.numBits] << encoded; // and the value itself - offset = 0; - } - - // send end-of-block code (0x00), only needed if there are trailing zeros - if (posNonZero < 8*8 - 1) // = 63 - writer << huffmanAC[0x00]; - - return DC; -} - -// Jon's code includes the pre-generated Huffman codes -// I don't like these "magic constants" and compute them on my own :-) -void generateHuffmanTable(const uint8_t numCodes[16], const uint8_t* values, BitCode result[256]) -{ - // process all bitsizes 1 thru 16, no JPEG Huffman code is allowed to exceed 16 bits - auto huffmanCode = 0; - for (auto numBits = 1; numBits <= 16; numBits++) - { - // ... and each code of these bitsizes - for (auto i = 0; i < numCodes[numBits - 1]; i++) // note: numCodes array starts at zero, but smallest bitsize is 1 - result[*values++] = BitCode(huffmanCode++, numBits); - - // next Huffman code needs to be one bit wider - huffmanCode <<= 1; - } -} - -} // end of anonymous namespace - -// -------------------- externally visible code -------------------- - -namespace TooJpeg -{ -// the only exported function ... -bool writeJpeg(WRITE_ONE_BYTE output, const void* pixels_, unsigned short width, unsigned short height, - bool isRGB, unsigned char quality_, bool downsample, const char* comment) -{ - // reject invalid pointers - if (output == nullptr || pixels_ == nullptr) - return false; - // check image format - if (width == 0 || height == 0) - return false; - - // number of components - const auto numComponents = isRGB ? 3 : 1; - // note: if there is just one component (=grayscale), then only luminance needs to be stored in the file - // thus everything related to chrominance need not to be written to the JPEG - // I still compute a few things, like quantization tables to avoid a complete code mess - - // grayscale images can't be downsampled (because there are no Cb + Cr channels) - if (!isRGB) - downsample = false; - - // wrapper for all output operations - BitWriter bitWriter(output); - - // //////////////////////////////////////// - // JFIF headers - const uint8_t HeaderJfif[2+2+16] = - { 0xFF,0xD8, // SOI marker (start of image) - 0xFF,0xE0, // JFIF APP0 tag - 0,16, // length: 16 bytes (14 bytes payload + 2 bytes for this length field) - 'J','F','I','F',0, // JFIF identifier, zero-terminated - 1,1, // JFIF version 1.1 - 0, // no density units specified - 0,1,0,1, // density: 1 pixel "per pixel" horizontally and vertically - 0,0 }; // no thumbnail (size 0 x 0) - bitWriter << HeaderJfif; - - // //////////////////////////////////////// - // comment (optional) - if (comment != nullptr) - { - // look for zero terminator - auto length = 0; // = strlen(comment); - while (comment[length] != 0) - length++; - - // write COM marker - bitWriter.addMarker(0xFE, 2+length); // block size is number of bytes (without zero terminator) + 2 bytes for this length field - // ... and write the comment itself - for (auto i = 0; i < length; i++) - bitWriter << comment[i]; - } - - // //////////////////////////////////////// - // adjust quantization tables to desired quality - - // quality level must be in 1 ... 100 - auto quality = clamp(quality_, 1, 100); - // convert to an internal JPEG quality factor, formula taken from libjpeg - quality = quality < 50 ? 5000 / quality : 200 - quality * 2; - - uint8_t quantLuminance [8*8]; - uint8_t quantChrominance[8*8]; - for (auto i = 0; i < 8*8; i++) - { - int luminance = (DefaultQuantLuminance [ZigZagInv[i]] * quality + 50) / 100; - int chrominance = (DefaultQuantChrominance[ZigZagInv[i]] * quality + 50) / 100; - - // clamp to 1..255 - quantLuminance [i] = clamp(luminance, 1, 255); - quantChrominance[i] = clamp(chrominance, 1, 255); - } - - // write quantization tables - bitWriter.addMarker(0xDB, 2 + (isRGB ? 2 : 1) * (1 + 8*8)); // length: 65 bytes per table + 2 bytes for this length field - // each table has 64 entries and is preceded by an ID byte - - bitWriter << 0x00 << quantLuminance; // first quantization table - if (isRGB) - bitWriter << 0x01 << quantChrominance; // second quantization table, only relevant for color images - - // //////////////////////////////////////// - // write image infos (SOF0 - start of frame) - bitWriter.addMarker(0xC0, 2+6+3*numComponents); // length: 6 bytes general info + 3 per channel + 2 bytes for this length field - - // 8 bits per channel - bitWriter << 0x08 - // image dimensions (big-endian) - << (height >> 8) << (height & 0xFF) - << (width >> 8) << (width & 0xFF); - - // sampling and quantization tables for each component - bitWriter << numComponents; // 1 component (grayscale, Y only) or 3 components (Y,Cb,Cr) - for (auto id = 1; id <= numComponents; id++) - bitWriter << id // component ID (Y=1, Cb=2, Cr=3) - // bitmasks for sampling: highest 4 bits: horizontal, lowest 4 bits: vertical - << (id == 1 && downsample ? 0x22 : 0x11) // 0x11 is default YCbCr 4:4:4 and 0x22 stands for YCbCr 4:2:0 - << (id == 1 ? 0 : 1); // use quantization table 0 for Y, table 1 for Cb and Cr - - // //////////////////////////////////////// - // Huffman tables - // DHT marker - define Huffman tables - bitWriter.addMarker(0xC4, isRGB ? (2+208+208) : (2+208)); - // 2 bytes for the length field, store chrominance only if needed - // 1+16+12 for the DC luminance - // 1+16+162 for the AC luminance (208 = 1+16+12 + 1+16+162) - // 1+16+12 for the DC chrominance - // 1+16+162 for the AC chrominance (208 = 1+16+12 + 1+16+162, same as above) - - // store luminance's DC+AC Huffman table definitions - bitWriter << 0x00 // highest 4 bits: 0 => DC, lowest 4 bits: 0 => Y (baseline) - << DcLuminanceCodesPerBitsize - << DcLuminanceValues; - bitWriter << 0x10 // highest 4 bits: 1 => AC, lowest 4 bits: 0 => Y (baseline) - << AcLuminanceCodesPerBitsize - << AcLuminanceValues; - - // compute actual Huffman code tables (see Jon's code for precalculated tables) - BitCode huffmanLuminanceDC[256]; - BitCode huffmanLuminanceAC[256]; - generateHuffmanTable(DcLuminanceCodesPerBitsize, DcLuminanceValues, huffmanLuminanceDC); - generateHuffmanTable(AcLuminanceCodesPerBitsize, AcLuminanceValues, huffmanLuminanceAC); - - // chrominance is only relevant for color images - BitCode huffmanChrominanceDC[256]; - BitCode huffmanChrominanceAC[256]; - if (isRGB) - { - // store luminance's DC+AC Huffman table definitions - bitWriter << 0x01 // highest 4 bits: 0 => DC, lowest 4 bits: 1 => Cr,Cb (baseline) - << DcChrominanceCodesPerBitsize - << DcChrominanceValues; - bitWriter << 0x11 // highest 4 bits: 1 => AC, lowest 4 bits: 1 => Cr,Cb (baseline) - << AcChrominanceCodesPerBitsize - << AcChrominanceValues; - - // compute actual Huffman code tables (see Jon's code for precalculated tables) - generateHuffmanTable(DcChrominanceCodesPerBitsize, DcChrominanceValues, huffmanChrominanceDC); - generateHuffmanTable(AcChrominanceCodesPerBitsize, AcChrominanceValues, huffmanChrominanceAC); - } - - // //////////////////////////////////////// - // start of scan (there is only a single scan for baseline JPEGs) - bitWriter.addMarker(0xDA, 2+1+2*numComponents+3); // 2 bytes for the length field, 1 byte for number of components, - // then 2 bytes for each component and 3 bytes for spectral selection - - // assign Huffman tables to each component - bitWriter << numComponents; - for (auto id = 1; id <= numComponents; id++) - // highest 4 bits: DC Huffman table, lowest 4 bits: AC Huffman table - bitWriter << id << (id == 1 ? 0x00 : 0x11); // Y: tables 0 for DC and AC; Cb + Cr: tables 1 for DC and AC - - // constant values for our baseline JPEGs (which have a single sequential scan) - static const uint8_t Spectral[3] = { 0, 63, 0 }; // spectral selection: must be from 0 to 63; successive approximation must be 0 - bitWriter << Spectral; - - // //////////////////////////////////////// - // adjust quantization tables with AAN scaling factors to simplify DCT - float scaledLuminance [8*8]; - float scaledChrominance[8*8]; - for (auto i = 0; i < 8*8; i++) - { - auto row = ZigZagInv[i] / 8; // same as ZigZagInv[i] >> 3 - auto column = ZigZagInv[i] % 8; // same as ZigZagInv[i] & 7 - - // scaling constants for AAN DCT algorithm: AanScaleFactors[0] = 1, AanScaleFactors[k=1..7] = cos(k*PI/16) * sqrt(2) - static const float AanScaleFactors[8] = { 1, 1.387039845f, 1.306562965f, 1.175875602f, 1, 0.785694958f, 0.541196100f, 0.275899379f }; - auto factor = 1 / (AanScaleFactors[row] * AanScaleFactors[column] * 8); - scaledLuminance [ZigZagInv[i]] = factor / quantLuminance [i]; - scaledChrominance[ZigZagInv[i]] = factor / quantChrominance[i]; - // if you really want JPEGs that are bitwise identical to Jon Olick's code then you need slightly different formulas (note: sqrt(8) = 2.828427125f) - //static const float aasf[] = { 1.0f * 2.828427125f, 1.387039845f * 2.828427125f, 1.306562965f * 2.828427125f, 1.175875602f * 2.828427125f, 1.0f * 2.828427125f, 0.785694958f * 2.828427125f, 0.541196100f * 2.828427125f, 0.275899379f * 2.828427125f }; // line 240 of jo_jpeg.cpp - //scaledLuminance [ZigZagInv[i]] = 1 / (quantLuminance [i] * aasf[row] * aasf[column]); // lines 266-267 of jo_jpeg.cpp - //scaledChrominance[ZigZagInv[i]] = 1 / (quantChrominance[i] * aasf[row] * aasf[column]); - } - - // //////////////////////////////////////// - // precompute JPEG codewords for quantized DCT - BitCode codewordsArray[2 * CodeWordLimit]; // note: quantized[i] is found at codewordsArray[quantized[i] + CodeWordLimit] - BitCode* codewords = &codewordsArray[CodeWordLimit]; // allow negative indices, so quantized[i] is at codewords[quantized[i]] - uint8_t numBits = 1; // each codeword has at least one bit (value == 0 is undefined) - int32_t mask = 1; // mask is always 2^numBits - 1, initial value 2^1-1 = 2-1 = 1 - for (int16_t value = 1; value < CodeWordLimit; value++) - { - // numBits = position of highest set bit (ignoring the sign) - // mask = (2^numBits) - 1 - if (value > mask) // one more bit ? - { - numBits++; - mask = (mask << 1) | 1; // append a set bit - } - codewords[-value] = BitCode(mask - value, numBits); // note that I use a negative index => codewords[-value] = codewordsArray[CodeWordLimit value] - codewords[+value] = BitCode( value, numBits); - } - - // just convert image data from void* - auto pixels = (const uint8_t*)pixels_; - - // the next two variables are frequently used when checking for image borders - const auto maxWidth = width - 1; // "last row" - const auto maxHeight = height - 1; // "bottom line" - - // process MCUs (minimum codes units) => image is subdivided into a grid of 8x8 or 16x16 tiles - const auto sampling = downsample ? 2 : 1; // 1x1 or 2x2 sampling - const auto mcuSize = 8 * sampling; - - // average color of the previous MCU - int16_t lastYDC = 0, lastCbDC = 0, lastCrDC = 0; - // convert from RGB to YCbCr - float Y[8][8], Cb[8][8], Cr[8][8]; - - for (auto mcuY = 0; mcuY < height; mcuY += mcuSize) // each step is either 8 or 16 (=mcuSize) - for (auto mcuX = 0; mcuX < width; mcuX += mcuSize) - { - // YCbCr 4:4:4 format: each MCU is a 8x8 block - the same applies to grayscale images, too - // YCbCr 4:2:0 format: each MCU represents a 16x16 block, stored as 4x 8x8 Y-blocks plus 1x 8x8 Cb and 1x 8x8 Cr block) - for (auto blockY = 0; blockY < mcuSize; blockY += 8) // iterate once (YCbCr444 and grayscale) or twice (YCbCr420) - for (auto blockX = 0; blockX < mcuSize; blockX += 8) - { - // now we finally have an 8x8 block ... - for (auto deltaY = 0; deltaY < 8; deltaY++) - { - auto column = minimum(mcuX + blockX , maxWidth); // must not exceed image borders, replicate last row/column if needed - auto row = minimum(mcuY + blockY + deltaY, maxHeight); - for (auto deltaX = 0; deltaX < 8; deltaX++) - { - // find actual pixel position within the current image - auto pixelPos = row * int(width) + column; // the cast ensures that we don't run into multiplication overflows - if (column < maxWidth) - column++; - - // grayscale images have solely a Y channel which can be easily derived from the input pixel by shifting it by 128 - if (!isRGB) - { - Y[deltaY][deltaX] = pixels[pixelPos] - 128.f; - continue; - } - - // RGB: 3 bytes per pixel (whereas grayscale images have only 1 byte per pixel) - auto r = pixels[3 * pixelPos ]; - auto g = pixels[3 * pixelPos + 1]; - auto b = pixels[3 * pixelPos + 2]; - - Y [deltaY][deltaX] = rgb2y (r, g, b) - 128; // again, the JPEG standard requires Y to be shifted by 128 - // YCbCr444 is easy - the more complex YCbCr420 has to be computed about 20 lines below in a second pass - if (!downsample) - { - Cb[deltaY][deltaX] = rgb2cb(r, g, b); // standard RGB-to-YCbCr conversion - Cr[deltaY][deltaX] = rgb2cr(r, g, b); - } - } - } - - // encode Y channel - lastYDC = encodeBlock(bitWriter, Y, scaledLuminance, lastYDC, huffmanLuminanceDC, huffmanLuminanceAC, codewords); - // Cb and Cr are encoded about 50 lines below - } - - // grayscale images don't need any Cb and Cr information - if (!isRGB) - continue; - - // //////////////////////////////////////// - // the following lines are only relevant for YCbCr420: - // average/downsample chrominance of four pixels while respecting the image borders - if (downsample) - for (short deltaY = 7; downsample && deltaY >= 0; deltaY--) // iterating loop in reverse increases cache read efficiency - { - auto row = minimum(mcuY + 2*deltaY, maxHeight); // each deltaX/Y step covers a 2x2 area - auto column = mcuX; // column is updated inside next loop - auto pixelPos = (row * int(width) + column) * 3; // numComponents = 3 - - // deltas (in bytes) to next row / column, must not exceed image borders - auto rowStep = (row < maxHeight) ? 3 * int(width) : 0; // always numComponents*width except for bottom line - auto columnStep = (column < maxWidth ) ? 3 : 0; // always numComponents except for rightmost pixel - - for (short deltaX = 0; deltaX < 8; deltaX++) - { - // let's add all four samples (2x2 area) - auto right = pixelPos + columnStep; - auto down = pixelPos + rowStep; - auto downRight = pixelPos + columnStep + rowStep; - - // note: cast from 8 bits to >8 bits to avoid overflows when adding - auto r = short(pixels[pixelPos ]) + pixels[right ] + pixels[down ] + pixels[downRight ]; - auto g = short(pixels[pixelPos + 1]) + pixels[right + 1] + pixels[down + 1] + pixels[downRight + 1]; - auto b = short(pixels[pixelPos + 2]) + pixels[right + 2] + pixels[down + 2] + pixels[downRight + 2]; - - // convert to Cb and Cr - Cb[deltaY][deltaX] = rgb2cb(r, g, b) / 4; // I still have to divide r,g,b by 4 to get their average values - Cr[deltaY][deltaX] = rgb2cr(r, g, b) / 4; // it's a bit faster if done AFTER CbCr conversion - - // step forward to next 2x2 area - pixelPos += 2*3; // 2 pixels => 6 bytes (2*numComponents) - column += 2; - - // reached right border ? - if (column >= maxWidth) - { - columnStep = 0; - pixelPos = ((row + 1) * int(width) - 1) * 3; // same as (row * width + maxWidth) * numComponents => current's row last pixel - } - } - } // end of YCbCr420 code for Cb and Cr - - // encode Cb and Cr - lastCbDC = encodeBlock(bitWriter, Cb, scaledChrominance, lastCbDC, huffmanChrominanceDC, huffmanChrominanceAC, codewords); - lastCrDC = encodeBlock(bitWriter, Cr, scaledChrominance, lastCrDC, huffmanChrominanceDC, huffmanChrominanceAC, codewords); - } - - bitWriter.flush(); // now image is completely encoded, write any bits still left in the buffer - - // /////////////////////////// - // EOI marker - bitWriter << 0xFF << 0xD9; // this marker has no length, therefore I can't use addMarker() - return true; -} // writeJpeg() -} // namespace TooJpeg diff --git a/src/spice2x/external/toojpeg/toojpeg.h b/src/spice2x/external/toojpeg/toojpeg.h deleted file mode 100644 index aa16b0c..0000000 --- a/src/spice2x/external/toojpeg/toojpeg.h +++ /dev/null @@ -1,62 +0,0 @@ -// ////////////////////////////////////////////////////////// -// toojpeg.h -// written by Stephan Brumme, 2018-2019 -// see https://create.stephan-brumme.com/toojpeg/ -// - -// This is a compact baseline JPEG/JFIF writer, written in C++ (but looks like C for the most part). -// Its interface has only one function: writeJpeg() - and that's it ! -// -// basic example: -// => create an image with any content you like, e.g. 1024x768, RGB = 3 bytes per pixel -// auto pixels = new unsigned char[1024*768*3]; -// => you need to define a callback that receives the compressed data byte-by-byte from my JPEG writer -// void myOutput(unsigned char oneByte) { fputc(oneByte, myFileHandle); } // save byte to file -// => let's go ! -// TooJpeg::writeJpeg(myOutput, mypixels, 1024, 768); - -#pragma once - -namespace TooJpeg -{ - // write one byte (to disk, memory, ...) - typedef void (*WRITE_ONE_BYTE)(unsigned char); - // this callback is called for every byte generated by the encoder and behaves similar to fputc - // if you prefer stylish C++11 syntax then it can be a lambda, too: - // auto myOutput = [](unsigned char oneByte) { fputc(oneByte, output); }; - - // output - callback that stores a single byte (writes to disk, memory, ...) - // pixels - stored in RGB format or grayscale, stored from upper-left to lower-right - // width,height - image size - // isRGB - true if RGB format (3 bytes per pixel); false if grayscale (1 byte per pixel) - // quality - between 1 (worst) and 100 (best) - // downsample - if true then YCbCr 4:2:0 format is used (smaller size, minor quality loss) instead of 4:4:4, not relevant for grayscale - // comment - optional JPEG comment (0/NULL if no comment), must not contain ASCII code 0xFF - bool writeJpeg(WRITE_ONE_BYTE output, const void* pixels, unsigned short width, unsigned short height, - bool isRGB = true, unsigned char quality = 90, bool downsample = false, const char* comment = nullptr); -} // namespace TooJpeg - -// My main inspiration was Jon Olick's Minimalistic JPEG writer -// ( https://www.jonolick.com/code.html => direct link is https://www.jonolick.com/uploads/7/9/2/1/7921194/jo_jpeg.cpp ). -// However, his code documentation is quite sparse - probably because it wasn't written from scratch and is (quote:) "based on a javascript jpeg writer", -// most likely Andreas Ritter's code: https://github.com/eugeneware/jpeg-js/blob/master/lib/encoder.js -// -// Therefore I wrote the whole lib from scratch and tried hard to add tons of comments to my code, especially describing where all those magic numbers come from. -// And I managed to remove the need for any external includes ... -// yes, that's right: my library has no (!) includes at all, not even #include -// Depending on your callback WRITE_ONE_BYTE, the library writes either to disk, or in-memory, or wherever you wish. -// Moreover, no dynamic memory allocations are performed, just a few bytes on the stack. -// -// In contrast to Jon's code, compression can be significantly improved in many use cases: -// a) grayscale JPEG images need just a single Y channel, no need to save the superfluous Cb + Cr channels -// b) YCbCr 4:2:0 downsampling is often about 20% more efficient (=smaller) than the default YCbCr 4:4:4 with only little visual loss -// -// TooJpeg 1.2+ compresses about twice as fast as jo_jpeg (and about half as fast as libjpeg-turbo). -// A few benchmark numbers can be found on my website https://create.stephan-brumme.com/toojpeg/#benchmark -// -// Last but not least you can optionally add a JPEG comment. -// -// Your C++ compiler needs to support a reasonable subset of C++11 (g++ 4.7 or Visual C++ 2013 are sufficient). -// I haven't tested the code on big-endian systems or anything that smells like an apple. -// -// USE AT YOUR OWN RISK. Because you are a brave soul :-) diff --git a/src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.cpp b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.cpp index a8b5365..26bb855 100644 --- a/src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.cpp +++ b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_device.cpp @@ -17,6 +17,7 @@ #include "d3d9_backend.h" #include "d3d9_live2d.h" +#include "d3d9_readback.h" #include "d3d9_texture.h" #ifndef SPICE64 @@ -156,6 +157,8 @@ ULONG STDMETHODCALLTYPE WrappedIDirect3DDevice9::Release() { } } + d3d9_readback::release_device_resources(this->pReal); + if (overlay::ENABLED) { const std::lock_guard lock(overlay::OVERLAY_MUTEX); diff --git a/src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.cpp b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.cpp new file mode 100644 index 0000000..1f36d77 --- /dev/null +++ b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.cpp @@ -0,0 +1,241 @@ +#include "d3d9_readback.h" + +#include +#include + +#include "hooks/graphics/graphics.h" +#include "util/logging.h" + +namespace d3d9_readback { + +namespace { + +SurfacePtr create_readback_surface(IDirect3DDevice9 *device, const D3DSURFACE_DESC &desc) { + IDirect3DSurface9 *surface = nullptr; + const HRESULT hr = device->CreateOffscreenPlainSurface( + desc.Width, desc.Height, desc.Format, D3DPOOL_SYSTEMMEM, &surface, nullptr); + + if (FAILED(hr) || surface == nullptr) { + log_warning("graphics::d3d9", + "failed to create readback surface, hr={}", + FMT_HRESULT(hr)); + return nullptr; + } + + return SurfacePtr(surface); +} + +size_t surface_bytes(const D3DSURFACE_DESC &desc) { + size_t bytes_per_pixel = 4; + switch (desc.Format) { + case D3DFMT_R5G6B5: + case D3DFMT_X1R5G5B5: + case D3DFMT_A1R5G5B5: + bytes_per_pixel = 2; + break; + default: + break; + } + + return static_cast(desc.Width) * desc.Height * bytes_per_pixel; +} + +// idle surfaces are kept between captures, bucketed by layout so that screens of +// differing resolution do not evict each other. a new device drops everything, +// since system memory surfaces outlive Reset but not the device itself +class ReadbackPool { +public: + SurfacePtr acquire(IDirect3DDevice9 *device, const D3DSURFACE_DESC &desc) { + { + std::lock_guard lock(this->mutex); + + if (this->device != device) { + this->drop(); + this->device = device; + } + + auto *bucket = this->find(desc); + if (bucket && !bucket->idle.empty()) { + auto surface = std::move(bucket->idle.back()); + bucket->idle.pop_back(); + + const size_t bytes = surface_bytes(desc); + this->idle_bytes = this->idle_bytes > bytes ? this->idle_bytes - bytes : 0; + return surface; + } + } + + return create_readback_surface(device, desc); + } + + void release(IDirect3DDevice9 *device, SurfacePtr surface) { + if (!surface) { + return; + } + + D3DSURFACE_DESC desc {}; + if (FAILED(surface->GetDesc(&desc))) { + return; + } + + const size_t bytes = surface_bytes(desc); + + std::lock_guard lock(this->mutex); + if (this->device != device || this->idle_bytes + bytes > MAX_IDLE_BYTES) { + return; + } + + auto *bucket = this->find(desc); + if (bucket == nullptr) { + if (this->buckets.size() >= MAX_BUCKETS) { + return; + } + + this->buckets.push_back(Bucket { desc.Width, desc.Height, desc.Format, {} }); + bucket = &this->buckets.back(); + } + + if (bucket->idle.size() < MAX_IDLE_PER_BUCKET) { + bucket->idle.push_back(std::move(surface)); + this->idle_bytes += bytes; + } + } + + // every cached surface holds a reference on the device, so they have to go + // before it does or the device never reaches a zero reference count + void clear_device(IDirect3DDevice9 *device) { + std::lock_guard lock(this->mutex); + if (this->device != device) { + return; + } + + this->drop(); + this->device = nullptr; + } + +private: + struct Bucket { + UINT width; + UINT height; + D3DFORMAT format; + std::vector idle; + }; + + static constexpr size_t MAX_BUCKETS = GRAPHICS_CAPTURE_SCREEN_NO; + + // one returning surface plus one for the next capture; a full screen surface + // is several megabytes, so the cap matters + static constexpr size_t MAX_IDLE_PER_BUCKET = 2; + + // a 4K surface is 33MB, so the per bucket count alone does not bound this + static constexpr size_t MAX_IDLE_BYTES = 64u * 1024 * 1024; + + void drop() { + this->buckets.clear(); + this->idle_bytes = 0; + } + + Bucket *find(const D3DSURFACE_DESC &desc) { + for (auto &bucket : this->buckets) { + if (bucket.width == desc.Width + && bucket.height == desc.Height + && bucket.format == desc.Format) { + return &bucket; + } + } + + return nullptr; + } + + std::mutex mutex; + std::vector buckets; + IDirect3DDevice9 *device = nullptr; + size_t idle_bytes = 0; +}; + +// deliberately never destroyed: releasing D3D surfaces during static destruction +// would run after d3d9 may already be unloaded +ReadbackPool &pool() { + static ReadbackPool *instance = new ReadbackPool(); + return *instance; +} + +} // namespace + +void release_device_resources(IDirect3DDevice9 *device) { + pool().clear_device(device); +} + +BackbufferCopy::~BackbufferCopy() { + if (this->pooled && this->surface) { + pool().release(this->device, std::move(this->surface)); + } +} + +std::optional acquire_backbuffer_copy( + IDirect3DDevice9 *device, IDirect3DSwapChain9 *swap_chain, int screen, bool pooled) { + + IDirect3DSurface9 *buffer = nullptr; + HRESULT hr = swap_chain->GetBackBuffer(0, D3DBACKBUFFER_TYPE_MONO, &buffer); + if (FAILED(hr) || buffer == nullptr) { + log_warning("graphics::d3d9", + "failed to get back buffer for screen {}, hr={}", + screen, + FMT_HRESULT(hr)); + return std::nullopt; + } + + D3DSURFACE_DESC desc {}; + hr = buffer->GetDesc(&desc); + if (FAILED(hr)) { + log_warning("graphics::d3d9", + "failed to acquire back buffer descriptor, hr={}", + FMT_HRESULT(hr)); + buffer->Release(); + return std::nullopt; + } + + // GetRenderTargetData rejects multisampled sources. no supported game has been + // seen presenting one, so resolving is left unimplemented rather than untested + if (desc.MultiSampleType != D3DMULTISAMPLE_NONE) { + static std::once_flag warned; + std::call_once(warned, [&desc] { + log_warning("graphics::d3d9", + "back buffer is multisampled ({}), screenshots and capture are unsupported", + static_cast(desc.MultiSampleType)); + }); + buffer->Release(); + return std::nullopt; + } + + auto destination = pooled + ? pool().acquire(device, desc) + : create_readback_surface(device, desc); + if (!destination) { + buffer->Release(); + return std::nullopt; + } + + hr = device->GetRenderTargetData(buffer, destination.get()); + buffer->Release(); + + if (FAILED(hr)) { + log_warning("graphics::d3d9", + "failed to copy back buffer contents, hr={}", + FMT_HRESULT(hr)); + if (pooled) { + pool().release(device, std::move(destination)); + } + return std::nullopt; + } + + BackbufferCopy copy; + copy.screen = screen; + copy.desc = desc; + copy.device = device; + copy.surface = std::move(destination); + copy.pooled = pooled; + + return copy; +} +} diff --git a/src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.h b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.h new file mode 100644 index 0000000..d49e595 --- /dev/null +++ b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_readback.h @@ -0,0 +1,44 @@ +#pragma once + +#include +#include + +#include + +namespace d3d9_readback { + + struct SurfaceReleaser { + void operator()(IDirect3DSurface9 *surface) const { + surface->Release(); + } + }; + + using SurfacePtr = std::unique_ptr; + + // system memory copy of a back buffer; locking it neither stalls the GPU nor reads over PCIe + struct BackbufferCopy { + int screen {}; + D3DSURFACE_DESC desc {}; + IDirect3DDevice9 *device = nullptr; + SurfacePtr surface; + bool pooled = false; + + BackbufferCopy() = default; + BackbufferCopy(BackbufferCopy &&) noexcept = default; + BackbufferCopy &operator=(BackbufferCopy &&) noexcept = default; + BackbufferCopy(const BackbufferCopy &) = delete; + BackbufferCopy &operator=(const BackbufferCopy &) = delete; + ~BackbufferCopy(); + }; + + // pooled copies reuse surfaces across calls and return them once the copy is destroyed, + // so the caller must keep it alive for as long as the pixels are being read + std::optional acquire_backbuffer_copy( + IDirect3DDevice9 *device, + IDirect3DSwapChain9 *swap_chain, + int screen, + bool pooled); + + // pooled surfaces hold references on the device; call this before releasing it + void release_device_resources(IDirect3DDevice9 *device); +} diff --git a/src/spice2x/hooks/graphics/backends/d3d9/d3d9_screenshot.cpp b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_screenshot.cpp index 0197e31..e23f09c 100644 --- a/src/spice2x/hooks/graphics/backends/d3d9/d3d9_screenshot.cpp +++ b/src/spice2x/hooks/graphics/backends/d3d9/d3d9_screenshot.cpp @@ -1,7 +1,10 @@ #include "d3d9_screenshot.h" #include +#include #include +#include +#include #include #include #include @@ -10,38 +13,21 @@ #include #include - -#ifdef __GNUC__ -#include -#endif +#include #include "avs/game.h" #include "hooks/graphics/graphics.h" #include "misc/clipboard.h" #include "overlay/notifications.h" #include "util/fileutils.h" -#include "util/libutils.h" #include "util/logging.h" #include "util/threadpool.h" #include "d3d9_device.h" +#include "d3d9_readback.h" -#ifdef __GNUC__ -typedef decltype(D3DXSaveSurfaceToFileA) *D3DXSaveSurfaceToFileA_t; -#else -#define D3DXIFF_PNG ((DWORD) 3) - -typedef HRESULT (WINAPI *D3DXSaveSurfaceToFileA_t)( - LPCSTR pDestFile, - DWORD DestFormat, - LPDIRECT3DSURFACE9 pSrcSurface, - CONST PALETTEENTRY *pSrcPalette, - CONST RECT *pSrcRect); -#endif - -static bool ATTEMPTED_D3DX9_LOAD_LIBRARY = false; - -// genpath picks free filenames by probing the disk, so only one save may run at a time +// genpath picks filenames by probing the disk, so the whole save has to be serialised: +// a name is only taken once its file exists, not when genpath hands it out static std::mutex SCREENSHOT_SAVE_M; namespace { @@ -56,56 +42,146 @@ struct ImageRequest { int screen; }; -struct SurfaceReleaser { - void operator()(IDirect3DSurface9 *surface) const { - surface->Release(); - } -}; - -using SurfacePtr = std::unique_ptr; - -struct BackbufferCopy { +// a screen already read out of its surface, so nothing here touches D3D. the bytes +// are still in the surface's format; converting them is left to the encode +struct PendingWrite { int screen {}; - D3DSURFACE_DESC desc {}; - SurfacePtr surface; + D3DFORMAT format {}; + UINT width {}; + UINT height {}; + size_t pitch {}; + std::vector data; + std::string path; + bool saved = false; }; -} // namespace +struct PendingCapture { + int screen {}; + D3DFORMAT format {}; + UINT width {}; + UINT height {}; + size_t pitch {}; + std::vector data; +}; -static void save_capture( - int screen, +// packed 24bpp RGB, what both the png encoder and the api capture consume +constexpr size_t RGB_PIXEL_SIZE = 3; + +// the formats surface_to_rgb knows how to convert; the two must stay in sync +static std::optional surface_pixel_size(D3DFORMAT format) { + switch (format) { + // what back buffers are actually created as in practice + case D3DFMT_X8R8G8B8: + case D3DFMT_A8R8G8B8: + + // a valid display format, but no supported game has been seen presenting one + case D3DFMT_A2R10G10B10: + return 4; + + // valid display formats, but no supported game has been seen presenting one + case D3DFMT_R5G6B5: + case D3DFMT_X1R5G5B5: + case D3DFMT_A1R5G5B5: + return 2; + + default: + return std::nullopt; + } +} + +struct ImageSize { + size_t row_size {}; + size_t total_size {}; +}; + +static std::optional compute_image_size( + UINT width, + UINT height, + size_t bytes_per_pixel) { + + if (width == 0 || height == 0 + || width > std::numeric_limits::max() / bytes_per_pixel) { + return std::nullopt; + } + + const size_t row_size = static_cast(width) * bytes_per_pixel; + if (height > std::numeric_limits::max() / row_size) { + return std::nullopt; + } + + return ImageSize { row_size, static_cast(height) * row_size }; +} + +static bool resize_pixels(std::vector &pixels, size_t size) { + try { + pixels.resize(size); + return true; + } catch (const std::exception &error) { + log_warning("graphics::d3d9", "failed to allocate image buffer: {}", error.what()); + return false; + } +} + +// the api capture stages a whole back buffer every frame, so the staging buffer +// is recycled rather than reallocated. returned buffers keep their size, which +// leaves the reuse free of a zero fill +class CaptureBuffers { +public: + std::vector take() { + std::lock_guard lock(this->mutex); + if (this->idle.empty()) { + return {}; + } + + auto buffer = std::move(this->idle.back()); + this->idle.pop_back(); + return buffer; + } + + void give(std::vector buffer) { + std::lock_guard lock(this->mutex); + if (this->idle.size() < MAX_IDLE) { + this->idle.push_back(std::move(buffer)); + } + } + +private: + // one per save in flight plus one for the next capture; a full screen is + // several megabytes, so the cap matters + static constexpr size_t MAX_IDLE = 2; + + std::mutex mutex; + std::vector> idle; +}; + +// deliberately never destroyed, so a save still running at process exit cannot +// hand a buffer back to a dead free list +CaptureBuffers &capture_buffers() { + static CaptureBuffers *instance = new CaptureBuffers(); + return *instance; +} + +// encodes get their own pool: the dispatch below already occupies a worker on its +// pool, so queueing onto that one and waiting could starve itself. never destroyed +// for the same reason as the buffers above +ThreadPool &encode_pool() { + static auto *instance = new ThreadPool(2); + return *instance; +} + +// normalize the supported D3D formats to packed 24bpp RGB. callers screen the +// format through surface_pixel_size first, so the black fill below is a fallback +void surface_to_rgb( D3DFORMAT format, UINT width, UINT height, - IDirect3DSurface9 *surface) { - HRESULT hr; + const uint8_t *data, + size_t pitch, + uint8_t *pixels) { - // lock surface to be able to access the data - D3DLOCKED_RECT finished_copy {}; - hr = surface->LockRect(&finished_copy, nullptr, 0); - if (FAILED(hr)) { - log_warning("graphics::d3d9", "failed to lock screenshot surface, hr={}", FMT_HRESULT(hr)); - graphics_capture_skip(screen); - return; - } - - // normalize supported D3D formats to packed RGB for API capture - size_t pitch = finished_copy.Pitch; - auto data = reinterpret_cast(finished_copy.pBits); - auto pixels = std::unique_ptr(new uint8_t[width * height * 3]); for (size_t row = 0; row < height; row++) { size_t offset_row = row * width * 3; switch (format) { - case D3DFMT_R8G8B8: { - for (size_t column = 0; column < width; column++) { - auto cell = data + row * pitch + column * 3; - auto pixel = &pixels[offset_row + column * 3]; - pixel[0] = cell[0]; - pixel[1] = cell[1]; - pixel[2] = cell[2]; - } - break; - } case D3DFMT_X8R8G8B8: case D3DFMT_A8R8G8B8: { for (size_t column = 0; column < width; column++) { @@ -117,14 +193,45 @@ static void save_capture( } break; } - case D3DFMT_X8B8G8R8: - case D3DFMT_A8B8G8R8: { + // the 5 and 6 bit channels are widened by bit replication so that + // full scale stays full scale + case D3DFMT_R5G6B5: { + auto cells = reinterpret_cast(data + row * pitch); for (size_t column = 0; column < width; column++) { - auto cell = data + row * pitch + column * 4; + const uint16_t cell = cells[column]; + const uint8_t red = (cell >> 11) & 0x1F; + const uint8_t green = (cell >> 5) & 0x3F; + const uint8_t blue = cell & 0x1F; auto pixel = &pixels[offset_row + column * 3]; - pixel[0] = cell[0]; - pixel[1] = cell[1]; - pixel[2] = cell[2]; + pixel[0] = (red << 3) | (red >> 2); + pixel[1] = (green << 2) | (green >> 4); + pixel[2] = (blue << 3) | (blue >> 2); + } + break; + } + case D3DFMT_X1R5G5B5: + case D3DFMT_A1R5G5B5: { + auto cells = reinterpret_cast(data + row * pitch); + for (size_t column = 0; column < width; column++) { + const uint16_t cell = cells[column]; + const uint8_t red = (cell >> 10) & 0x1F; + const uint8_t green = (cell >> 5) & 0x1F; + const uint8_t blue = cell & 0x1F; + auto pixel = &pixels[offset_row + column * 3]; + pixel[0] = (red << 3) | (red >> 2); + pixel[1] = (green << 3) | (green >> 2); + pixel[2] = (blue << 3) | (blue >> 2); + } + break; + } + case D3DFMT_A2R10G10B10: { + auto cells = reinterpret_cast(data + row * pitch); + for (size_t column = 0; column < width; column++) { + const uint32_t cell = cells[column]; + auto pixel = &pixels[offset_row + column * 3]; + pixel[0] = static_cast((cell >> 22) & 0xFF); + pixel[1] = static_cast((cell >> 12) & 0xFF); + pixel[2] = static_cast((cell >> 2) & 0xFF); } break; } @@ -138,99 +245,148 @@ static void save_capture( } } } +} - // unlock surface - hr = surface->UnlockRect(); - if (FAILED(hr)) { - log_warning("graphics::d3d9", "failed to unlock screenshot surface, hr={}", FMT_HRESULT(hr)); - graphics_capture_skip(screen); +} // namespace + +using d3d9_readback::BackbufferCopy; + +static void save_capture(PendingCapture capture) { + const auto size = compute_image_size(capture.width, capture.height, RGB_PIXEL_SIZE); + if (!size.has_value()) { + capture_buffers().give(std::move(capture.data)); + graphics_capture_skip(capture.screen); return; } - // enqueue - graphics_capture_enqueue(screen, pixels.release(), width, height); + auto pixels = std::unique_ptr(new (std::nothrow) uint8_t[size->total_size]); + if (!pixels) { + log_warning("graphics::d3d9", "failed to allocate capture image buffer"); + capture_buffers().give(std::move(capture.data)); + graphics_capture_skip(capture.screen); + return; + } + + // a format we cannot read still has to produce a frame, or api clients stall + if (capture.data.empty()) { + std::memset(pixels.get(), 0, size->total_size); + } else { + surface_to_rgb( + capture.format, + capture.width, + capture.height, + capture.data.data(), + capture.pitch, + pixels.get()); + + capture_buffers().give(std::move(capture.data)); + } + + graphics_capture_enqueue(capture.screen, pixels.release(), capture.width, capture.height); } -static bool save_screenshot( +enum class SurfaceRead { + Ok, + Unsupported, + Failed, +}; + +// copying the surface touches D3D, so it stays on the caller's thread. the bytes come +// out in the surface's own format; converting them is plain memory work for later +static SurfaceRead read_surface_raw( + const BackbufferCopy ©, + size_t &row_size, + std::vector &out) { + + const auto bytes_per_pixel = surface_pixel_size(copy.desc.Format); + if (!bytes_per_pixel.has_value()) { + static std::once_flag warned; + std::call_once(warned, [©] { + log_warning("graphics::d3d9", + "unsupported surface format {}", + static_cast(copy.desc.Format)); + }); + return SurfaceRead::Unsupported; + } + + const auto size = compute_image_size(copy.desc.Width, copy.desc.Height, *bytes_per_pixel); + if (!size.has_value() || !resize_pixels(out, size->total_size)) { + return SurfaceRead::Failed; + } + + D3DLOCKED_RECT locked {}; + HRESULT hr = copy.surface->LockRect(&locked, nullptr, D3DLOCK_READONLY); + if (FAILED(hr)) { + log_warning("graphics::d3d9", "failed to lock capture surface, hr={}", FMT_HRESULT(hr)); + return SurfaceRead::Failed; + } + + if (locked.Pitch < 0 || static_cast(locked.Pitch) < size->row_size) { + log_warning("graphics::d3d9", "capture surface has invalid pitch {}", locked.Pitch); + copy.surface->UnlockRect(); + return SurfaceRead::Failed; + } + + auto data = reinterpret_cast(locked.pBits); + for (size_t row = 0; row < copy.desc.Height; row++) { + std::memcpy( + out.data() + row * size->row_size, + data + row * locked.Pitch, + size->row_size); + } + + hr = copy.surface->UnlockRect(); + if (FAILED(hr)) { + log_warning("graphics::d3d9", "failed to unlock capture surface, hr={}", FMT_HRESULT(hr)); + return SurfaceRead::Failed; + } + + row_size = size->row_size; + return SurfaceRead::Ok; +} + +static bool read_capture_surface( + const BackbufferCopy ©, + PendingCapture &capture) { + + capture.screen = copy.screen; + capture.format = copy.desc.Format; + capture.width = copy.desc.Width; + capture.height = copy.desc.Height; + + capture.data = capture_buffers().take(); + const auto result = read_surface_raw(copy, capture.pitch, capture.data); + if (result == SurfaceRead::Ok) { + return true; + } + + capture_buffers().give(std::move(capture.data)); + capture.data.clear(); + + // a format we cannot read is reported as a black frame rather than nothing, + // so a client polling the api keeps getting responses + return result == SurfaceRead::Unsupported; +} + +static bool write_screenshot_png( const std::string &file_path, - D3DFORMAT format, UINT width, UINT height, - IDirect3DSurface9 *surface) { - // 32-bit XRGB and ARGB surfaces use byte 3 as alpha; force opaque PNG output - if (format == D3DFMT_X8R8G8B8 || format == D3DFMT_A8R8G8B8 || - format == D3DFMT_X8B8G8R8 || format == D3DFMT_A8B8G8R8) { + const std::vector &pixels) { - D3DLOCKED_RECT finished_copy {}; - HRESULT hr = surface->LockRect(&finished_copy, nullptr, 0); - if (FAILED(hr)) { - log_warning("graphics::d3d9", "failed to lock screenshot surface, hr={}", FMT_HRESULT(hr)); - return false; - } + // a no-op while FPNG_NO_SSE is set, but fpng requires it before any encode + static std::once_flag fpng_ready; + std::call_once(fpng_ready, [] { fpng::fpng_init(); }); - const size_t pitch = finished_copy.Pitch; - auto data = reinterpret_cast(finished_copy.pBits); - for (size_t row = 0; row < height; row++) { - for (size_t column = 0; column < width; column++) { - data[row * pitch + column * 4 + 3] = 255; - } - } - - hr = surface->UnlockRect(); - if (FAILED(hr)) { - log_warning("graphics::d3d9", "failed to unlock screenshot surface, hr={}", FMT_HRESULT(hr)); - return false; - } - } - - // lazy load function - static D3DXSaveSurfaceToFileA_t D3DXSaveSurfaceToFileA_ptr = nullptr; - if (D3DXSaveSurfaceToFileA_ptr == nullptr) { - D3DXSaveSurfaceToFileA_ptr = libutils::try_proc("D3DXSaveSurfaceToFileA"); - - // check if function was not found, likely because d3dx9 is not loaded - if (!ATTEMPTED_D3DX9_LOAD_LIBRARY && D3DXSaveSurfaceToFileA_ptr == nullptr) { - ATTEMPTED_D3DX9_LOAD_LIBRARY = true; - - // prefer the newest installed helper while supporting older D3DX9 runtimes - for (size_t i = 43; i >= 24; i--) { - auto lib_name = fmt::format("d3dx9_{}.dll", i); - auto d3dx9 = libutils::try_library(lib_name); - - // Check if library was not found - if (d3dx9 == nullptr) { - continue; - } - - D3DXSaveSurfaceToFileA_ptr = libutils::try_proc( - d3dx9, "D3DXSaveSurfaceToFileA"); - - // Check if function was not found - if (D3DXSaveSurfaceToFileA_ptr == nullptr) { - FreeLibrary(d3dx9); - d3dx9 = nullptr; - - continue; - } - - log_info("graphics::d3d9", "found surface save function in '{}'", lib_name); - break; - } - } - } - - if (D3DXSaveSurfaceToFileA_ptr == nullptr) { - log_warning("graphics::d3d9", "Direct3D save helper function not available"); - return false; - } - - // save to file log_info("graphics::d3d9", "saving screenshot to {}", file_path); - const HRESULT save_result = D3DXSaveSurfaceToFileA_ptr( - file_path.c_str(), D3DXIFF_PNG, surface, nullptr, nullptr); - if (FAILED(save_result)) { - log_warning("graphics::d3d9", "Failed to save screenshot"); + if (!fpng::fpng_encode_image_to_file( + file_path.c_str(), + pixels.data(), + static_cast(width), + static_cast(height), + 3)) { + log_warning("graphics::d3d9", "failed to write screenshot png"); return false; } @@ -249,139 +405,9 @@ static std::string screenshot_path_for_screen(const std::string &primary_path, i .string(); } -static std::optional acquire_backbuffer_copy( - IDirect3DDevice9 *device, IDirect3DSwapChain9 *swap_chain, int screen) { - - IDirect3DSurface9 *buffer = nullptr; - HRESULT hr = swap_chain->GetBackBuffer(0, D3DBACKBUFFER_TYPE_MONO, &buffer); - if (FAILED(hr) || buffer == nullptr) { - log_warning("graphics::d3d9", - "failed to get back buffer for screen {}, hr={}", - screen, - FMT_HRESULT(hr)); - return std::nullopt; - } - - D3DSURFACE_DESC desc {}; - hr = buffer->GetDesc(&desc); - if (FAILED(hr)) { - log_warning("graphics::d3d9", - "failed to acquire back buffer descriptor, hr={}", - FMT_HRESULT(hr)); - buffer->Release(); - return std::nullopt; - } - - // TODO: cache render targets - IDirect3DSurface9 *temp_surface = nullptr; - hr = device->CreateRenderTarget( - desc.Width, desc.Height, desc.Format, desc.MultiSampleType, - desc.MultiSampleQuality, TRUE, &temp_surface, nullptr); - if (FAILED(hr) || temp_surface == nullptr) { - log_warning("graphics::d3d9", - "failed to acquire temporary surface, hr={}", - FMT_HRESULT(hr)); - buffer->Release(); - return std::nullopt; - } - - hr = device->StretchRect(buffer, nullptr, temp_surface, nullptr, D3DTEXF_NONE); - if (FAILED(hr)) { - log_warning("graphics::d3d9", - "failed to copy back buffer contents, hr={}", - FMT_HRESULT(hr)); - temp_surface->Release(); - buffer->Release(); - return std::nullopt; - } - - // release original back buffer reference - buffer->Release(); - - return BackbufferCopy { - .screen = screen, - .desc = desc, - .surface = SurfacePtr(temp_surface), - }; -} - -static void dispatch_surface_save( - const ImageRequest &request, - std::vector copies, - size_t screen_count) { - auto surface_process = [request, screen_count, copies = std::move(copies)]() { - switch (request.kind) { - case ImageRequestKind::Capture: { - const auto © = copies.front(); - save_capture( - request.screen, - copy.desc.Format, - copy.desc.Width, - copy.desc.Height, - copy.surface.get()); - break; - } - - case ImageRequestKind::Screenshot: { - std::lock_guard lock(SCREENSHOT_SAVE_M); - - std::vector screens; - screens.reserve(copies.size()); - for (const auto © : copies) { - screens.push_back(copy.screen); - } - - const auto base_path = graphics_screenshot_genpath(screens); - if (base_path.empty()) { - break; - } - - // screens missing from copies already failed to be acquired - size_t failed = screen_count - copies.size(); - std::string primary_path; - std::string notify_path; - for (const auto © : copies) { - const auto path = screenshot_path_for_screen(base_path, copy.screen); - if (!save_screenshot( - path, - copy.desc.Format, - copy.desc.Width, - copy.desc.Height, - copy.surface.get())) { - failed++; - continue; - } - if (notify_path.empty()) { - notify_path = path; - } - if (copy.screen == 0) { - primary_path = path; - } - } - - // only the primary screen goes to the clipboard, but any saved file is a success - if (!primary_path.empty()) { - clipboard::copy_image(primary_path); - } - if (!notify_path.empty()) { - overlay::notifications::add( - overlay::notifications::Severity::Success, - fmt::format("Screenshot saved: {}", fileutils::basename(notify_path))); - } else { - overlay::notifications::add( - overlay::notifications::Severity::Error, - "Screenshot failed to save"); - } - - if (failed > 0) { - log_warning("graphics::d3d9", "{} screenshot screen(s) missing", failed); - } - break; - } - } - }; - - // list of games that crash when running the screenshot processor on another thread +// games that crash or hang when the screenshot processor runs on another thread. +// D3DCREATE_MULTITHREADED is not a predictor of this; MDX omits it and threads fine +static bool image_processing_must_be_inline() { static const robin_hood::unordered_set THREAD_BAN { "JMA", #ifndef SPICE64 @@ -393,12 +419,164 @@ static void dispatch_surface_save( "LMA", }; - // run the save operation on another thread for supported games - if (THREAD_BAN.contains(avs::game::MODEL)) { - surface_process(); + return THREAD_BAN.contains(avs::game::MODEL); +} + +static void dispatch_capture_save(PendingCapture capture) { + auto capture_process = [capture = std::move(capture)]() mutable { + // an escape from here would cross a thread boundary and terminate + try { + save_capture(std::move(capture)); + } catch (const std::exception &error) { + log_warning("graphics::d3d9", "capture save failed: {}", error.what()); + } catch (...) { + log_warning("graphics::d3d9", "capture save failed"); + } + }; + + if (image_processing_must_be_inline()) { + capture_process(); } else { static auto pool = ThreadPool(2); - pool.add(std::move(surface_process)); + pool.add(std::move(capture_process)); + } +} + +// by this point the pixels are plain memory, so none of this needs the device +static void dispatch_screenshot_save(std::vector writes, size_t screen_count) { + auto screenshot_process = [writes = std::move(writes), screen_count]() mutable { + std::lock_guard lock(SCREENSHOT_SAVE_M); + + std::vector screens; + screens.reserve(writes.size()); + for (const auto &write : writes) { + screens.push_back(write.screen); + } + + const auto base_path = graphics_screenshot_genpath(screens); + if (base_path.empty()) { + return; + } + + for (auto &write : writes) { + write.path = screenshot_path_for_screen(base_path, write.screen); + } + + // screens missing from writes either failed to be acquired or failed to read + size_t failed = screen_count - writes.size(); + + // a throw here would otherwise reach a thread boundary and terminate + auto encode_one = [](PendingWrite &write) { + try { + const auto rgb = compute_image_size(write.width, write.height, RGB_PIXEL_SIZE); + std::vector pixels; + if (!rgb.has_value() || !resize_pixels(pixels, rgb->total_size)) { + write.saved = false; + return; + } + + surface_to_rgb( + write.format, + write.width, + write.height, + write.data.data(), + write.pitch, + pixels.data()); + + // the encode below is the long part; the raw copy is dead by now + write.data.clear(); + write.data.shrink_to_fit(); + + write.saved = write_screenshot_png( + write.path, write.width, write.height, pixels); + } catch (const std::exception &error) { + log_warning("graphics::d3d9", + "screenshot encode failed for {}: {}", write.path, error.what()); + write.saved = false; + } catch (...) { + log_warning("graphics::d3d9", + "screenshot encode failed for {}", write.path); + write.saved = false; + } + }; + + { + // sized up front and assigned by index: storing a future must not be able + // to throw once its task is queued, or the screen would encode twice + std::vector> pending(writes.empty() ? 0 : writes.size() - 1); + for (size_t i = 1; i < writes.size(); i++) { + try { + pending[i - 1] = encode_pool().add([&writes, &encode_one, i] { + encode_one(writes[i]); + }); + } catch (const std::exception &) { + // nothing to queue onto; encoding it here still makes progress + encode_one(writes[i]); + } + } + + if (!writes.empty()) { + encode_one(writes.front()); + } + + for (auto &task : pending) { + if (task.valid()) { + task.wait(); + } + } + } + + std::string primary_path; + std::string notify_path; + for (const auto &write : writes) { + if (!write.saved) { + failed++; + continue; + } + if (notify_path.empty()) { + notify_path = write.path; + } + if (write.screen == 0) { + primary_path = write.path; + } + } + + // only the primary screen goes to the clipboard, but any saved file is a success + if (!primary_path.empty()) { + clipboard::copy_image(primary_path); + } + if (!notify_path.empty()) { + overlay::notifications::add( + overlay::notifications::Severity::Success, + fmt::format("Screenshot saved: {}", fileutils::basename(notify_path))); + } else { + overlay::notifications::add( + overlay::notifications::Severity::Error, + "Screenshot failed to save"); + } + + if (failed > 0) { + log_warning("graphics::d3d9", "{} screenshot screen(s) missing", failed); + } + }; + + // genpath and the path building below allocate, so an escape from here would + // cross a thread boundary and terminate + auto guarded = [process = std::move(screenshot_process)]() mutable { + try { + process(); + } catch (const std::exception &error) { + log_warning("graphics::d3d9", "screenshot save failed: {}", error.what()); + } catch (...) { + log_warning("graphics::d3d9", "screenshot save failed"); + } + }; + + if (image_processing_must_be_inline()) { + guarded(); + } else { + static auto pool = ThreadPool(2); + pool.add(std::move(guarded)); } } @@ -427,7 +605,8 @@ static void process_image_request( screen, FMT_HRESULT(hr)); } else { - copy = acquire_backbuffer_copy(device, swap_chain, screen); + // only the API capture path runs often enough to benefit from pooling + copy = d3d9_readback::acquire_backbuffer_copy(device, swap_chain, screen, !screenshot); swap_chain->Release(); } @@ -443,7 +622,39 @@ static void process_image_request( return; } - dispatch_surface_save(request, std::move(copies), screens.size()); + if (!screenshot) { + PendingCapture capture; + if (!read_capture_surface(copies.front(), capture)) { + graphics_capture_skip(request.screen); + return; + } + + copies.clear(); + dispatch_capture_save(std::move(capture)); + return; + } + + // reading a surface touches the device, and doing that off the present thread + // has been seen to deadlock games whose device has no internal locking + std::vector writes; + writes.reserve(copies.size()); + for (const auto © : copies) { + PendingWrite write; + write.screen = copy.screen; + write.format = copy.desc.Format; + write.width = copy.desc.Width; + write.height = copy.desc.Height; + + if (read_surface_raw(copy, write.pitch, write.data) != SurfaceRead::Ok) { + continue; + } + + writes.push_back(std::move(write)); + } + + copies.clear(); + + dispatch_screenshot_save(std::move(writes), screens.size()); } void graphics_d3d9_process_screenshot( diff --git a/src/spice2x/hooks/graphics/graphics.cpp b/src/spice2x/hooks/graphics/graphics.cpp index 31f1db8..04214e6 100644 --- a/src/spice2x/hooks/graphics/graphics.cpp +++ b/src/spice2x/hooks/graphics/graphics.cpp @@ -20,6 +20,7 @@ #include "games/io.h" #include "games/sdvx/sdvx.h" #include "games/popn/popn.h" +#include "hooks/graphics/jpeg_encoder.h" #include "hooks/graphics/backends/d3d9/d3d9_backend.h" #include "hooks/graphics/backends/d3d11/d3d11_backend.h" #include "launcher/shutdown.h" @@ -64,7 +65,6 @@ static bool GRAPHICS_SCREENSHOT_TRIGGER = false; static std::set GRAPHICS_SCREENS { 0 }; static std::mutex GRAPHICS_SCREENS_M {}; static std::vector GRAPHICS_CAPTURE_SCREENS; -static const size_t GRAPHICS_CAPTURE_SCREEN_NO = 4; static std::mutex GRAPHICS_CAPTURE_SCREENS_M {}; static CaptureData GRAPHICS_CAPTURE_BUFFER[GRAPHICS_CAPTURE_SCREEN_NO] {}; static std::mutex GRAPHICS_CAPTURE_BUFFER_M[GRAPHICS_CAPTURE_SCREEN_NO] {}; @@ -1481,10 +1481,12 @@ void graphics_capture_skip(int screen) { GRAPHICS_CAPTURE_CV[screen].notify_one(); } -bool graphics_capture_receive_jpeg(int screen, TooJpeg::WRITE_ONE_BYTE receiver, - bool rgb, int quality, bool downsample, int divide, uint64_t *timestamp, +bool graphics_capture_receive_jpeg(int screen, std::vector &out, + int quality, int divide, uint64_t *timestamp, int *width, int *height) { + out.clear(); + if (screen < 0 || screen >= static_cast(GRAPHICS_CAPTURE_SCREEN_NO)) { return false; } @@ -1557,10 +1559,13 @@ bool graphics_capture_receive_jpeg(int screen, TooJpeg::WRITE_ONE_BYTE receiver, } // compress - auto success = TooJpeg::writeJpeg( - receiver, capture_data.get(), - capture_width, capture_height, - rgb, quality, downsample); + const bool success = jpeg_encoder::encode( + out, capture_data.get(), + capture_width, capture_height, quality); + + if (!success) { + out.clear(); + } // status if (timestamp) { diff --git a/src/spice2x/hooks/graphics/graphics.h b/src/spice2x/hooks/graphics/graphics.h index 717e7b4..f88bb79 100644 --- a/src/spice2x/hooks/graphics/graphics.h +++ b/src/spice2x/hooks/graphics/graphics.h @@ -12,8 +12,6 @@ #include #endif -#include "external/toojpeg/toojpeg.h" - // order must match spice2x_AutoOrientation UI enum order enum graphics_orientation { ORIENTATION_CW = 0, @@ -140,12 +138,16 @@ void graphics_screens_get(std::vector &screens); void graphics_poll_screenshot_hotkey(); void graphics_screenshot_trigger(); bool graphics_screenshot_consume(); + +inline constexpr size_t GRAPHICS_CAPTURE_SCREEN_NO = 4; + void graphics_capture_trigger(int screen); bool graphics_capture_consume(int *screen); void graphics_capture_enqueue(int screen, uint8_t *data, size_t width, size_t height); void graphics_capture_skip(int screen); -bool graphics_capture_receive_jpeg(int screen, TooJpeg::WRITE_ONE_BYTE receiver, - bool rgb = true, int quality = 80, bool downsample = true, int divide = 0, +// on success `out` holds the encoded JPEG; its storage is reused across calls +bool graphics_capture_receive_jpeg(int screen, std::vector &out, + int quality = 80, int divide = 0, uint64_t *timestamp = nullptr, int *width = nullptr, int *height = nullptr); // the returned path is for screen 0; any extra screens only reserve their suffixed names diff --git a/src/spice2x/hooks/graphics/jpeg_encoder.cpp b/src/spice2x/hooks/graphics/jpeg_encoder.cpp new file mode 100644 index 0000000..5edb0d2 --- /dev/null +++ b/src/spice2x/hooks/graphics/jpeg_encoder.cpp @@ -0,0 +1,140 @@ +#include "jpeg_encoder.h" + +#include +#include + +#include + +namespace jpeg_encoder { + +namespace { + +constexpr size_t CHUNK_SIZE = 16 * 1024; + +// libjpeg writes through a destination manager; this one appends straight into +// the caller's vector so the encoded frame is never copied +struct VectorDestination { + jpeg_destination_mgr mgr {}; + std::vector *out = nullptr; + uint8_t chunk[CHUNK_SIZE] {}; +}; + +void dest_init(j_compress_ptr cinfo) { + auto dest = reinterpret_cast(cinfo->dest); + dest->mgr.next_output_byte = dest->chunk; + dest->mgr.free_in_buffer = CHUNK_SIZE; +} + +// libjpeg cannot unwind a C++ exception out of its own frames, so growing the +// output has to fail by value and be turned into an error_exit by the caller +bool dest_append(VectorDestination *dest, size_t size) { + try { + dest->out->insert(dest->out->end(), dest->chunk, dest->chunk + size); + return true; + } catch (...) { + return false; + } +} + +boolean dest_empty(j_compress_ptr cinfo) { + auto dest = reinterpret_cast(cinfo->dest); + + // returning FALSE would mean suspension to libjpeg, not failure + if (!dest_append(dest, CHUNK_SIZE)) { + (*cinfo->err->error_exit)(reinterpret_cast(cinfo)); + } + + dest->mgr.next_output_byte = dest->chunk; + dest->mgr.free_in_buffer = CHUNK_SIZE; + return TRUE; +} + +void dest_term(j_compress_ptr cinfo) { + auto dest = reinterpret_cast(cinfo->dest); + if (!dest_append(dest, CHUNK_SIZE - dest->mgr.free_in_buffer)) { + (*cinfo->err->error_exit)(reinterpret_cast(cinfo)); + } +} + +// the default handler calls exit(), which is not an option inside a game process +struct ErrorManager { + jpeg_error_mgr mgr {}; + jmp_buf escape {}; +}; + +void on_error(j_common_ptr cinfo) { + longjmp(reinterpret_cast(cinfo->err)->escape, 1); +} + +void on_message(j_common_ptr) { +} + +} // namespace + +bool encode( + std::vector &out, + const uint8_t *pixels, + int width, + int height, + int quality) { + + if (!pixels || width <= 0 || height <= 0) { + return false; + } + + if (quality < 1) { + quality = 1; + } else if (quality > 100) { + quality = 100; + } + + // zero init is load bearing: the trap below is armed before the struct is + // created, and jpeg_destroy_compress only tolerates that on a zeroed struct + jpeg_compress_struct cinfo {}; + ErrorManager err; + VectorDestination dest; + + cinfo.err = jpeg_std_error(&err.mgr); + err.mgr.error_exit = on_error; + err.mgr.output_message = on_message; + + if (setjmp(err.escape)) { + jpeg_destroy_compress(&cinfo); + return false; + } + + jpeg_create_compress(&cinfo); + + dest.out = &out; + dest.mgr.init_destination = dest_init; + dest.mgr.empty_output_buffer = dest_empty; + dest.mgr.term_destination = dest_term; + cinfo.dest = &dest.mgr; + + cinfo.image_width = static_cast(width); + cinfo.image_height = static_cast(height); + cinfo.in_color_space = JCS_RGB; + cinfo.input_components = 3; + + jpeg_set_defaults(&cinfo); + jpeg_set_quality(&cinfo, quality, TRUE); + + // 4:2:0, matching what the capture path asked the previous encoder for + cinfo.comp_info[0].h_samp_factor = 2; + cinfo.comp_info[0].v_samp_factor = 2; + + jpeg_start_compress(&cinfo, TRUE); + + const size_t pitch = static_cast(width) * 3; + while (cinfo.next_scanline < cinfo.image_height) { + auto row = const_cast(pixels + cinfo.next_scanline * pitch); + JSAMPROW rows[1] = { row }; + jpeg_write_scanlines(&cinfo, rows, 1); + } + + jpeg_finish_compress(&cinfo); + jpeg_destroy_compress(&cinfo); + + return true; +} +} diff --git a/src/spice2x/hooks/graphics/jpeg_encoder.h b/src/spice2x/hooks/graphics/jpeg_encoder.h new file mode 100644 index 0000000..b1629bc --- /dev/null +++ b/src/spice2x/hooks/graphics/jpeg_encoder.h @@ -0,0 +1,15 @@ +#pragma once + +#include +#include + +namespace jpeg_encoder { + + // appends a baseline 4:2:0 JPEG of packed 24bpp RGB pixels to `out` + bool encode( + std::vector &out, + const uint8_t *pixels, + int width, + int height, + int quality); +} diff --git a/src/spice2x/licenses.txt b/src/spice2x/licenses.txt index d7a9b3a..ea32d87 100644 --- a/src/spice2x/licenses.txt +++ b/src/spice2x/licenses.txt @@ -527,25 +527,78 @@ HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABI TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -toojpeg (zlib) +fpng (Unlicense) ------------------------------------------- -zlib License +Copyright (c) 2021 Richard Geldreich, Jr. +Incorporates public domain code by Alex Evans, the original miniz by +Richard Geldreich, Jr., and Huffman code size work by Alistair Moffat and +Jyrki Katajainen. -Copyright (c) 2011-2016 Stephan Brumme +This is free and unencumbered software released into the public domain. -This software is provided 'as-is', without any express or implied warranty. In -no event will the authors be held liable for any damages arising from the use -of this software. -Permission is granted to anyone to use this software for any purpose, including -commercial applications, and to alter it and redistribute it freely, subject to -the following restrictions: -1. The origin of this software must not be misrepresented; you must not claim - that you wrote the original software. If you use this software in a product, - an acknowledgment in the product documentation would be appreciated but is - not required. -2. Altered source versions must be plainly marked as such, and must not be - misrepresented as being the original software. -3. This notice may not be removed or altered from any source distribution. +Anyone is free to copy, modify, publish, use, compile, sell, or distribute +this software, either in source code form or as a compiled binary, for any +purpose, commercial or non-commercial, and by any means. + +In jurisdictions that recognize copyright laws, the author or authors of this +software dedicate any and all copyright interest in the software to the public +domain. We make this dedication for the benefit of the public at large and to +the detriment of our heirs and successors. We intend this dedication to be an +overt act of relinquishment in perpetuity of all present and future rights to +this software under copyright law. + +THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE +AUTHORS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN +ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION +WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. + +libjpeg-turbo (IJG) +------------------------------------------- +This software is based in part on the work of the Independent JPEG Group. + +Only the libjpeg API library is used here. Per the libjpeg-turbo licensing +terms that portion is covered by the IJG License, reproduced below; the SIMD +sources bear the zlib License, whose terms are subsumed by the IJG License in +the context of the overall libjpeg API library. The Modified (3-clause) BSD +License covers the TurboJPEG API library and build system, neither of which is +distributed here. + +The authors make NO WARRANTY or representation, either express or implied, +with respect to this software, its quality, accuracy, merchantability, or +fitness for a particular purpose. This software is provided "AS IS", and you, +its user, assume the entire risk as to its quality and accuracy. + +This software is copyright (C) 1991-2020, Thomas G. Lane, Guido Vollbeding. +All Rights Reserved except as specified below. + +Permission is hereby granted to use, copy, modify, and distribute this +software (or portions thereof) for any purpose, without fee, subject to these +conditions: +(1) If any part of the source code for this software is distributed, then this +README file must be included, with this copyright and no-warranty notice +unaltered; and any additions, deletions, or changes to the original files +must be clearly indicated in accompanying documentation. +(2) If only executable code is distributed, then the accompanying +documentation must state that "this software is based in part on the work of +the Independent JPEG Group". +(3) Permission for use of this software is granted only if the user accepts +full responsibility for any undesirable consequences; the authors accept +NO LIABILITY for damages of any kind. + +These conditions apply to any software derived from or based on the IJG code, +not just to the unmodified library. If you use our work, you ought to +acknowledge us. + +Permission is NOT granted for the use of any IJG author's name or company name +in advertising or publicity relating to this software or products derived from +it. This software may be referred to only as "the Independent JPEG Group's +software". + +We specifically permit and encourage the use of this software as the basis of +commercial products, provided that all warranty or liability claims are +assumed by the product vendor. robin_hood.h (MIT) ------------------------------------------- diff --git a/src/spice2x/misc/clipboard.cpp b/src/spice2x/misc/clipboard.cpp index 0a4c2f5..5cf17fd 100644 --- a/src/spice2x/misc/clipboard.cpp +++ b/src/spice2x/misc/clipboard.cpp @@ -58,17 +58,18 @@ namespace clipboard { return; } - // If the screenshot button is print screen, the OpenClipboard call seems to fail often if we only - // call it once, probably due to a race condition. So, we can try calling a lot until we can open it. + // print screen key leaves the OS briefly holding the clipboard, so retry + // spinning without yielding starves the thread we are waiting on and looks like a hang bool clipboard_open = false; - for (int i = 0; i < 1000000; i++) { + for (int i = 0; i < 100; i++) { if (OpenClipboard(nullptr)) { clipboard_open = true; break; } + Sleep(5); } if (!clipboard_open) { - log_warning("clipboard", "Failed to open clipboard"); + log_warning("clipboard", "failed to open clipboard"); return; }