Cfu s8s8 (#10413)

Adding S8S8 kernels for symmetric quantized indirect conv and depthwise conv. Perf number with single thread: Nokia G10 (baseline / new) in ms Pixel 4 (baseline/new) in ms mobilenet_edgetpu 220 / 213 18.5 / 17.6 cartoongan 8537 / 8521 967 / 928 Co-authored-by: Chen Fu <[email protected]>
microsoft · Jan 28, 2022 · c4f1dfc · c4f1dfc
1 parent 1a2925a
commit c4f1dfc
Show file tree

Hide file tree

Showing 18 changed files with 3,701 additions and 222 deletions.
diff --git a/cmake/onnxruntime_mlas.cmake b/cmake/onnxruntime_mlas.cmake
@@ -47,9 +47,12 @@ function(setup_mlas_source_for_windows)
       )
 
       set(mlas_platform_preprocess_srcs
+        ${MLAS_SRC_DIR}/arm64/ConvSymS8KernelDot.asm
         ${MLAS_SRC_DIR}/arm64/ConvSymU8KernelDot.asm
+        ${MLAS_SRC_DIR}/arm64/ConvSymS8KernelNeon.asm
         ${MLAS_SRC_DIR}/arm64/ConvSymU8KernelNeon.asm
-        ${MLAS_SRC_DIR}/arm64/DepthwiseConvsymKernelNeon.asm
+        ${MLAS_SRC_DIR}/arm64/DepthwiseQConvSymS8KernelNeon.asm
+        ${MLAS_SRC_DIR}/arm64/DepthwiseQConvSymU8KernelNeon.asm
         ${MLAS_SRC_DIR}/arm64/DepthwiseQConvKernelSize9Neon.asm
         ${MLAS_SRC_DIR}/arm64/QgemmU8X8KernelNeon.asm
         ${MLAS_SRC_DIR}/arm64/QgemmS8S8KernelNeon.asm
@@ -272,9 +275,12 @@ else()
     if(ARM64 AND MLAS_SOURCE_IS_NOT_SET )
         enable_language(ASM)
         set(mlas_platform_srcs
+          ${MLAS_SRC_DIR}/aarch64/ConvSymS8KernelDot.S
           ${MLAS_SRC_DIR}/aarch64/ConvSymU8KernelDot.S
+          ${MLAS_SRC_DIR}/aarch64/ConvSymS8KernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/ConvSymU8KernelNeon.S
-          ${MLAS_SRC_DIR}/aarch64/DepthwiseConvSymKernelNeon.S
+          ${MLAS_SRC_DIR}/aarch64/DepthwiseQConvSymS8KernelNeon.S
+          ${MLAS_SRC_DIR}/aarch64/DepthwiseQConvSymU8KernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/DepthwiseQConvKernelSize9Neon.S
           ${MLAS_SRC_DIR}/aarch64/QgemmU8X8KernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/QgemmS8S8KernelNeon.S