Enable AVX2 optimizations for libwebp on x86/x64 platforms

[mattleibow]

Summary

Enable AVX2 SIMD optimizations in libwebp for improved WebP encoding/decoding performance on modern x86/x64 CPUs.

Background

libwebp 1.6.0 added AVX2 optimizations in two new source files:

• src/dsp/lossless_avx2.c
• src/dsp/lossless_enc_avx2.c

Upstream commit: webmproject/libwebp@f2b3f527

During the libwebp 1.3.2 → 1.6.0 update (#3461), we had to disable AVX2 to fix Windows build failures. This issue tracks re-enabling it properly.

Current Workaround

We disabled AVX2 using -mno-avx2 compiler flag in third_party/libwebp/BUILD.gn:

config("libwebp_no_avx2") {
  if (current_cpu == "x86" || current_cpu == "x64") {
    cflags = [ "-mno-avx2" ]
  }
}

This prevents the compiler from defining __AVX2__, which would otherwise trigger AVX2 code paths.

Why Windows Failed (and others did not)

The AVX2 code path is enabled via preprocessor macros in src/dsp/cpu.h:

// Lines 59-62: MSVC-compatible compilers trigger this
#if defined(_MSC_VER) && _MSC_VER >= 1700 && \
    (defined(_M_X64) || defined(_M_IX86))
#define WEBP_MSC_AVX2
#endif

// Lines 88-95: This enables the AVX2 code path
#if (defined(__AVX2__) || defined(WEBP_MSC_AVX2)) && \
    (!defined(HAVE_CONFIG_H) || defined(WEBP_HAVE_AVX2))
#define WEBP_USE_AVX2
#endif

Why Windows fails: SkiaSharp uses clang-cl on Windows, which defines _MSC_VER for MSVC compatibility. This triggers WEBP_MSC_AVX2 → WEBP_HAVE_AVX2, causing lossless.c and lossless_enc.c to call VP8LDspInitAVX2() and VP8LEncDspInitAVX2(). But these functions are defined in the AVX2 source files which are not compiled → linker error.

Why macOS/Linux x64 works: Native Clang/GCC does not define _MSC_VER, and does not define __AVX2__ by default, so AVX2 is not enabled.

Why ARM works: AVX2 is an x86 instruction set, ARM platforms have no AVX2 code paths.

How to Enable AVX2

Add a new AVX2 target following the existing SSE4.1 pattern:

third_party("libwebp_avx2") {
  public_include_dirs = [
    "../externals/libwebp/src",
    "../externals/libwebp",
  ]
  configs = [ ":libwebp_defines" ]
  sources = [
    "../externals/libwebp/src/dsp/lossless_avx2.c",
    "../externals/libwebp/src/dsp/lossless_enc_avx2.c",
  ]
  
  # Enable AVX2 compilation on x86/x64
  if (current_cpu == "x86" || current_cpu == "x64") {
    if (is_win && !is_clang) {
      # MSVC
      cflags_c = [ "/arch:AVX2" ]
    } else {
      # Clang (macOS, Linux, Windows clang-cl)
      cflags_c = [ "-mavx2" ]
    }
  }
}

third_party("libwebp") {
  # ...
  deps = [ ":libwebp_sse41", ":libwebp_avx2" ]  # Add AVX2 dependency
  # ...
}

Then remove the config("libwebp_no_avx2") workaround.

Compiler Flags Summary

Platform	Compiler	Flag to enable AVX2
Windows (MSVC)	cl.exe	/arch:AVX2
Windows (clang-cl)	clang-cl	-mavx2
macOS	clang	-mavx2
Linux	clang/gcc	-mavx2

Safety: Runtime CPU Detection

libwebp has runtime CPU detection - the AVX2 code only runs on CPUs that support it:

// src/dsp/lossless.c lines 658-661
#if defined(WEBP_HAVE_AVX2)
  if (VP8GetCPUInfo(kAVX2)) {  // Runtime check!
    VP8LDspInitAVX2();
  }
#endif

So even on older CPUs without AVX2 support, the code safely falls back to SSE/C implementations.

What AVX2 Does

AVX2 (Advanced Vector Extensions 2) is a SIMD instruction set that processes 256 bits at once (vs 128 bits for SSE). This provides faster:

• Lossless WebP encoding
• Lossless WebP decoding
• Color space conversions

Available on: Intel Haswell (2013+), AMD Excavator (2015+), and newer CPUs.

Files to Modify

• externals/skia/third_party/libwebp/BUILD.gn - Add AVX2 target, remove -mno-avx2 workaround

References

• libwebp AVX2 commit - Original upstream commit adding AVX2 support
• libwebp 1.6.0 release
• PR #3461 - libwebp 1.6.0 update

[github-actions]

Triage Summary

Labels will be applied to indicate the performance focus (tenet/performance) and the affected platforms (os/Windows-Classic and os/Linux) regarding AVX2 optimizations in the libwebp library.

This issue is not a regression, as it is primarily concerned with performance improvements rather than any historical failures.

Additional remarks:

• The issue emphasizes enabling AVX2 optimizations specifically for the libwebp library on x86/x64 platforms, without indicating any compatibility or reliability issues in the current context.
• No relevant area labels apply to the optimizations discussed, as they do not pertain to documentation or bindings related to SkiaSharp or HarfBuzz.

Detailed Summary and Actions

Summary of the triage:

• The issue is focused on performance, specifically enabling AVX2 optimizations in the libwebp library.
• The affected platforms include Windows and Linux environments, as the optimizations are related to these systems.
• No regression is indicated since this relates to enhancements rather than fixing previous faults.

Summary of the actions that will be performed:

Action	Item	Description
Apply Label	tenet/performance	The issue focuses on enabling AVX2 optimizations in libwebp to enhance performance in encoding and decoding.
Apply Label	os/Windows-Classic	The issue discusses enabling AVX2 optimizations that address build failures specifically on Windows platforms.
Apply Label	os/Linux	The issue considers enabling AVX2 optimizations for Linux x64 platforms, indicating their involvement.

This entire triage process was automated by AI and mistakes may have been made. Please let us know so we can continue to improve.