[aarch64] Implement QGEMM S8S8 kernel with SMMLA instructions

microsoft · Aug 16, 2023 · 28ea61f · 28ea61f
1 parent ad9fe3e
commit 28ea61f
Show file tree

Hide file tree

Showing 5 changed files with 1,919 additions and 0 deletions.
diff --git a/cmake/onnxruntime_mlas.cmake b/cmake/onnxruntime_mlas.cmake
@@ -62,6 +62,7 @@ function(setup_mlas_source_for_windows)
         ${MLAS_SRC_DIR}/qgemm_kernel_udot.cpp
         ${MLAS_SRC_DIR}/qgemm_kernel_sdot.cpp
         ${MLAS_SRC_DIR}/qgemm_kernel_ummla.cpp
+        ${MLAS_SRC_DIR}/qgemm_kernel_smmla.cpp
       )
 
       set(mlas_platform_preprocess_srcs
@@ -311,6 +312,7 @@ else()
           ${MLAS_SRC_DIR}/aarch64/QgemmU8X8KernelUdot.S
           ${MLAS_SRC_DIR}/aarch64/QgemmU8X8KernelUmmla.S
           ${MLAS_SRC_DIR}/aarch64/QgemmS8S8KernelSdot.S
+          ${MLAS_SRC_DIR}/aarch64/QgemmS8S8KernelSmmla.S
           ${MLAS_SRC_DIR}/aarch64/SgemmKernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/SgemvKernelNeon.S
           ${MLAS_SRC_DIR}/aarch64/SymQgemmS8KernelNeon.S
@@ -320,6 +322,7 @@ else()
           ${MLAS_SRC_DIR}/qgemm_kernel_udot.cpp
           ${MLAS_SRC_DIR}/qgemm_kernel_sdot.cpp
           ${MLAS_SRC_DIR}/qgemm_kernel_ummla.cpp
+          ${MLAS_SRC_DIR}/qgemm_kernel_smmla.cpp
         )
         if (NOT APPLE)
           set(mlas_platform_srcs