use 4x16s4 gemm kernel for meteor lake

google · May 27, 2024 · 0143aab · 0143aab
1 parent 041ea32
commit 0143aab
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 2 deletions.
diff --git a/cmake/DownloadCpuinfo.cmake b/cmake/DownloadCpuinfo.cmake
@@ -17,8 +17,8 @@ ENDIF()
 
 INCLUDE(ExternalProject)
 ExternalProject_Add(cpuinfo
-  URL https://github.com/pytorch/cpuinfo/archive/d6860c477c99f1fce9e28eb206891af3c0e1a1d7.zip
-  URL_HASH SHA256=a615cac78fad03952cc3e1fd231ce789a8df6e81a5957b64350cb8200364b385
+  URL https://github.com/xujuntwt95329/cpuinfo/archive/10bf0a03827877a4c6896189f27e291f78a62ce5.zip
+  URL_HASH SHA256=2128b3087bf8c190128fbe8f1003552f84f5f0ed79c79868212d99eee11296eb
   SOURCE_DIR "${CMAKE_BINARY_DIR}/cpuinfo-source"
   BINARY_DIR "${CMAKE_BINARY_DIR}/cpuinfo"
   CONFIGURE_COMMAND ""

diff --git a/src/configs/gemm-config.c b/src/configs/gemm-config.c
@@ -789,6 +789,7 @@ static void init_f32_gemm_config(void) {
       switch (cpuinfo_get_core(0)->uarch) {
         case cpuinfo_uarch_zen:
         case cpuinfo_uarch_dhyana:
+        case cpuinfo_uarch_meteor_lake:
           f32_gemm_config.minmax.gemm[XNN_MR_TO_INDEX(1)] = xnn_init_hmp_gemm_ukernel((xnn_gemm_ukernel_fn) xnn_f32_gemm_minmax_ukernel_1x16s4__fma3_broadcast);
           f32_gemm_config.minmax.gemm[XNN_MR_TO_INDEX(4)] = xnn_init_hmp_gemm_ukernel((xnn_gemm_ukernel_fn) xnn_f32_gemm_minmax_ukernel_4x16s4__fma3_broadcast);
           f32_gemm_config.minmax.igemm[XNN_MR_TO_INDEX(1)] = xnn_init_hmp_igemm_ukernel((xnn_igemm_ukernel_fn) xnn_f32_igemm_minmax_ukernel_1x16s4__fma3_broadcast);