mirror of
https://chromium.googlesource.com/libyuv/libyuv
synced 2026-11-09 14:49:57 +08:00
78277c25e6
x86 fixes:
- Prefer CopyRow_AVX512BW and SetRow_AVX2 over ERMS rep movsb/stosb
in CopyPlane, SetPlane and RotatePlane. ERMS was checked last and
replaced the vector rows on all modern CPUs; on Zen 4 CopyRow_AVX512BW
is 1.2x to 5x faster than rep movsb for 64 to 3840 byte rows.
- Only report AMX-INT8 and ACEv1 when XCR0 enables tile state,
CPUID reports AMX-TILE and, on Linux, the process has XTILEDATA
permission (ARCH_GET_XCOMP_PERM). Without permission tile
instructions raise SIGILL. libyuv does not request the permission.
- Detect AVX512-BMM from AMD leaf 0x80000021 EAX[23] instead of basic
leaf 0x21, and detect ERMS from leaf 0x80000021 without requiring AVX.
- Allow AVX10 CPUs without the legacy AVX512F bit when CPUID leaf 0x24
reports 512-bit vectors.
- Remove unused kCpuHasAVX512VBITALG and reuse its bit for kCpuHasAPX
(CPUID.7.1:EDX[21]).
- Restrict CLANG_HAS_AVX512 to x86_64 and mark xgetbv asm volatile
with a plain "edx" clobber.
- Use 64-bit division in FixedDiv_X86 and FixedDiv1_X86 on x86_64 so
quotients >= 2^31 no longer raise SIGFPE.
- Sample the center pixel when scaling 32768 or more pixels down to 1
instead of stretching the destination and sampling pixel 0.
- Check InterpolateRow_16_AVX2 alignment against 16 pixels, matching
its loop, instead of 32.
- ARGBBlur: reject a NULL cumulative sum buffer, check the result of
ARGBComputeCumulativeSum, and use uint32_t arithmetic in the C
cumulative sum rows to avoid signed overflow on 4K frames.
Fix red and blue channels being swapped by SplitARGBRow, MergeARGBRow,
SplitXRGBRow and MergeXRGBRow for WebAssembly SIMD128 and LoongArch
LSX/LASX. libyuv ARGB is B, G, R, A in memory. The Wasm kernels stored
byte 0 to dst_r and interleaved (r, g) first. The LoongArch kernels
named their pointer parameters in B, G, R order while callers pass
R, G, B. Split and Merge had matching errors, so they cancelled out.
The SplitARGBPlane, MergeARGBPlane, SplitXRGBPlane, MergeXRGBPlane,
SplitRGBPlane and MergeRGBPlane tests only compared a full split and
merge round trip against C, which hides a channel swap that is
mirrored in both directions. The Split tests now compare each plane
against C, the Merge tests merge the C planes with both paths, and the
new SplitMergePlane_ChannelOrder test checks fixed channel values
without relying on the C code.
Bump version to 1980.
Test: libyuv_unittest --gtest_filter=*Split*:*Merge*:*ChannelOrder*
Test: libyuv_unittest (x86_64, x86_32, LIBYUV_ENABLE_ROWWIN, GCC,
MSan, UBSan)
Bug: libyuv:573815252
Change-Id: I61df6a542bc95a00f7e69441ee5e012f1c00eae7
Reviewed-on: https://chromium-review.googlesource.com/c/libyuv/libyuv/+/8749626
Reviewed-by: richard winterton <rrwinterton@gmail.com>
Commit-Queue: Frank Barchard <fbarchard@google.com>
|
||
|---|---|---|
| build_overrides | ||
| docs | ||
| include | ||
| infra/config | ||
| riscv_script | ||
| source | ||
| tools_libyuv | ||
| unit_test | ||
| util | ||
| .clang-format | ||
| .gitignore | ||
| .gn | ||
| Android.bp | ||
| Android.mk | ||
| AUTHORS | ||
| BUILD.bazel | ||
| BUILD.gn | ||
| CM_linux_packages.cmake | ||
| CMakeLists.txt | ||
| codereview.settings | ||
| DEPS | ||
| DIR_METADATA | ||
| download_vs_toolchain.py | ||
| GEMINI.md | ||
| libyuv.bzl | ||
| libyuv.gni | ||
| libyuv.gyp | ||
| libyuv.gypi | ||
| LICENSE | ||
| linux.mk | ||
| OWNERS | ||
| PATENTS | ||
| PRESUBMIT.py | ||
| public.mk | ||
| pylintrc | ||
| README.chromium | ||
| README.md | ||
| vpython.toml | ||
| vpython.toml.uv.lock | ||
| winarm.mk | ||
| WORKSPACE.bazel | ||
libyuv is an open source project that includes YUV scaling and conversion functionality.
- Scale YUV to prepare content for compression, with point, bilinear or box filter.
- Convert to YUV from webcam formats for compression.
- Convert to RGB formats for rendering/effects.
- Rotate by 90/180/270 degrees to adjust for mobile devices in portrait mode.
- Optimized for SSSE3/AVX2 on x86/x64.
- Optimized for Neon/SVE2/SME on Arm.
- Optimized for MSA on Mips.
- Optimized for RVV on RISC-V.
Development
See Getting started for instructions on how to get started developing.
You can also browse the docs directory for more documentation.