From 0afe56970324894311883e8ea56b39217f7abe70 Mon Sep 17 00:00:00 2001 From: Mingtao Gu <145657261+mtgu0705@users.noreply.github.com> Date: Fri, 9 May 2025 23:25:31 +0800 Subject: [PATCH] fix moe gemm2 for gfx950 (#2164) Co-authored-by: mtgu0705 [ROCm/composable_kernel commit: a23390163d604d9f00ea43e920822ac7cfb0884f] --- example/65_gemm_multiply_multiply/CMakeLists.txt | 2 +- example/65_gemm_multiply_multiply/moe_gemm2_xdl_pk_i4.cpp | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/example/65_gemm_multiply_multiply/CMakeLists.txt b/example/65_gemm_multiply_multiply/CMakeLists.txt index 5d2a097576..8d51d43c65 100644 --- a/example/65_gemm_multiply_multiply/CMakeLists.txt +++ b/example/65_gemm_multiply_multiply/CMakeLists.txt @@ -7,7 +7,7 @@ add_example_executable(example_gemm_multiply_multiply_xdl_int8 gemm_multiply_mul add_example_executable(example_moe_gemm1_xdl_fp8 moe_gemm1_xdl_fp8.cpp) add_example_executable(example_moe_gemm2_xdl_fp8 moe_gemm2_xdl_fp8.cpp) -list(APPEND gpu_list gfx942) +list(APPEND gpu_list gfx942 gfx950) set(target 0) foreach(gpu IN LISTS GPU_TARGETS) if(gpu IN_LIST gpu_list AND target EQUAL 0) diff --git a/example/65_gemm_multiply_multiply/moe_gemm2_xdl_pk_i4.cpp b/example/65_gemm_multiply_multiply/moe_gemm2_xdl_pk_i4.cpp index b9621cc9b3..3745e3d0af 100644 --- a/example/65_gemm_multiply_multiply/moe_gemm2_xdl_pk_i4.cpp +++ b/example/65_gemm_multiply_multiply/moe_gemm2_xdl_pk_i4.cpp @@ -281,7 +281,7 @@ int main(int argc, char* argv[]) break; case 4: a0_t_k_k.GenerateTensorValue(GeneratorTensor_1{}); - b0_e_n_k.GenerateTensorValue(GeneratorTensor_2{-2, 2}); + b0_e_n_k.GenerateTensorValue(GeneratorTensor_2{-2, 2}); d0_t_n.GenerateTensorValue(GeneratorTensor_1{}); d1_e_n.GenerateTensorValue(GeneratorTensor_1{}); d2_e_n.GenerateTensorValue(GeneratorTensor_1{});