From 25bb0d2fee306a6192d68ab3d3477affce125427 Mon Sep 17 00:00:00 2001 From: aska-0096 Date: Wed, 23 Apr 2025 02:08:44 +0000 Subject: [PATCH] add flags to avoid vectorizer problem --- example/01_gemm/CMakeLists.txt | 2 +- example/01_gemm/gemm_xdl_fp8_v3.cpp | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-) diff --git a/example/01_gemm/CMakeLists.txt b/example/01_gemm/CMakeLists.txt index 02482d5b39..39212d2904 100755 --- a/example/01_gemm/CMakeLists.txt +++ b/example/01_gemm/CMakeLists.txt @@ -39,7 +39,7 @@ add_example_dependencies(example_gemm_xdl example_gemm_xdl_fp16_fp8_streamk_v3) add_example_executable(example_gemm_xdl_bf16_v3 gemm_xdl_bf16_v3.cpp) add_example_dependencies(example_gemm_xdl example_gemm_xdl_bf16_v3) set(GEMM_OPTIONS) -# list(APPEND GEMM_OPTIONS -mllvm -greedy-reverse-local-assignment=1) +list(APPEND GEMM_OPTIONS "SHELL: -mllvm -greedy-reverse-local-assignment=1 -mllvm --slp-threshold=-16") list(APPEND GEMM_OPTIONS -v --save-temps -Wno-gnu-line-marker) target_compile_options(example_gemm_xdl_bf16_v3 PRIVATE ${GEMM_OPTIONS}) target_compile_options(example_gemm_xdl_fp8_v3 PRIVATE ${GEMM_OPTIONS}) diff --git a/example/01_gemm/gemm_xdl_fp8_v3.cpp b/example/01_gemm/gemm_xdl_fp8_v3.cpp index 0270ffe591..55a6c60273 100644 --- a/example/01_gemm/gemm_xdl_fp8_v3.cpp +++ b/example/01_gemm/gemm_xdl_fp8_v3.cpp @@ -28,10 +28,10 @@ using DeviceGemmV2Instance = ADataType, BDataType, CDataType, AccDataType, CShuffleDataType, PassThrough, PassThrough, PassThrough, GemmDefault, 256, - 128, 128, + 256, 256, 128, 16, 16, 16, 16, - 4, 4, + 8, 8, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>, 2, 16, 16, 0, S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>,