Simd v6.2.154

Algorithms

New features

• SSE4.1, AVX2, AVX-512BW, AVX-512VNNI, AMX-INT8 optimizations of class SynetQuantizedMergedConvolutionCdc.
• Base implementation, SSE4.1, AVX2, AVX-512BW, AVX-512VNNI, AMX-INT8 optimizations of class SynetQuantizedMergedConvolutionCd.
• Base implementation, SSE4.1, AVX2, AVX-512BW, AVX-512VNNI, AMX-INT8 optimizations of class SynetQuantizedMergedConvolutionDc.
• Base implementation of function SynetQuantizedScaleLayerForward.
• Base implementation of function SynetQuantizedPreluLayerForward.

Improve

• AVX-512VNNI optimizations of class SynetQuantizedConvolutionNhwcDepthwiseV3.
• Performance of AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcGemm (case of batch > 1).
• Performance of AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcSpecV0 (case of batch > 1).
• Performance of AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcGemm (case of small srcC).
• Performance of AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcSpecV0 (case of small srcC).

Bug fixing

• Error in AVX-512BW optimizations of function SynetQuantizedConcatLayerForward.
• Error in function Base::CpuModel (Windows Server 2025).
• Error in Base implementation, SSE4.1, AVX2, AVX-512BW optimizations of class SynetQuantizedAddUniform.
• Error in Base implementation, SSE4.1, AVX2, AVX-512BW optimizations of function QuantizedMergedConvolutionAddInputToOutput.
• Error in AMX-INT8 optimizations of class SynetQuantizedConvolutionNhwcGemm (case of batch > 1).

Test framework

New features

• Tests for verifying functionality of function SynetQuantizedScaleLayerForward.
• Tests for verifying functionality of function SynetQuantizedPreluLayerForward.

Infrastructure

Bug fixing

• Fix bug in step 'Host Properties' in Github actions script for MSBuild.
• Fix bug in step 'Host Properties' in Github actions script for CMake.

Removing

• Support of Microsoft Visual Studio 2019.