This repo is an attempt to port some of the SIMD code written in assembly to C intrinsics, to be used with WebAssembly.
git clone --depth 1 https://github.com/AlexVestin/x264-wasm x264
mkdir build
make wasm-libx264-simd
Needs nasm (there's a target for this in the Makefile) for native builds with assembly, and emsdk for wasm builds.
Substituting only the SAD, and SATD 8x8 function (the 4x4 is already very optimized) with SIMD code there was around a 1.6x speedup running a very hastily made benchmark.
https://software.intel.com/sites/landingpage/IntrinsicsGuide/#techs=SSE2,SSE3,SSSE3,SSE4_1,AVX
emscripten/wasm generally only supports SSE1 & SSE2 instructionss (https://emscripten.org/docs/porting/simd.html#compiling-simd-code-targeting-x86-sse-instruction-set), and some emulated instructions up to AVX
build with -pg and --enable-gprof for x264 (maybe -no-pie/-fno-pie)
- https://wiki.videolan.org/X264_asm_intro/ ( https://wiki.videolan.org/X264asm/ )
- http://www.apsipa.org/proceedings_2012/papers/201.pdf
- https://eiken.dev/blog/2020/04/optimizing-the-walsh-hadamard-transform-using-simd-intrinsics/
- https://mailman.videolan.org/pipermail/x264-devel/attachments/20130423/ffd6bfb6/attachment-0001.pdf
- https://www.researchgate.net/publication/4222524_Introduction_to_H264_advanced_video_coding
- https://www.researchgate.net/profile/Daniel_Lun/publication/220598752_Improved_SIMD_Architecture_for_High_Performance_Video_Processors/links/5498d1cf0cf2519f5a1dea03/Improved-SIMD-Architecture-for-High-Performance-Video-Processors.pdf
- https://pdfs.semanticscholar.org/9942/1f78e7f0c0dbfaaedf6c73d3bf9c3bea18fc.pdf