Skip to content

Simd v6.2.155

Choose a tag to compare

@ermig1979 ermig1979 released this 10 Nov 08:24
· 225 commits to master since this release

Algorithms

New features
  • SSE4.1, AVX2, AVX-512BW optimizations of function SynetQuantizedScaleLayerForward.
  • SSE4.1, AVX2, AVX-512BW optimizations of function SynetQuantizedPreluLayerForward.
  • Arbitrary activation function in Base implementation of class SynetQuantizedConvolutionGemm.
  • Arbitrary activation function in Base implementation, SSE4.1, AVX2, AVX-512BW, AVX-512VNNI, AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcGemm.
  • Arbitrary activation function in Base implementation, SSE4.1, AVX2, AVX-512BW, AVX-512VNNI, AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcSpecV0.
  • Arbitrary activation function in Base implementation, SSE4.1, AVX2, AVX-512BW, AVX-512VNNI optimizations of class SynetQuantizedConvolutionNhwcDepthwiseV2.
  • Arbitrary activation function in Base implementation, AVX-512VNNI optimizations of class SynetQuantizedConvolutionNhwcDepthwiseV3.
Improve
  • AMX-BF16 optimizations of class SynetConvolution16bNhwcGemm (case of small srcC).
Bug fixing
  • Performance bug in AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcGemm.
  • Error in SSE4.1, AVX2, AVX-512BW, AVX-512VNNI optimizations of class SynetQuantizedInnerProductGemmNN.
  • Error in SSE4.1 optimizations of class SynetQuantizedConvolutionNhwcSpecV0.
  • Error in Base implementation of class SynetQuantizedConvolutionNhwcSpecV0.
  • Error in Base implementation of class SynetQuantizedConvolutionNhwcGemm.