ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 13:11:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix grouped gemm check to avoid overflow (#1545)

Browse Source
This commit is contained in:
Bartłomiej Kocot
2024-10-04 17:32:43 +02:00
committed by GitHub
parent aeb7c91f48
commit 6b54d2faf8
4 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multi_abd_xdl_fixed_nk.hpp
View File

@@ -64,7 +64,7 @@ __global__ void
const index_t N = gemm_desc_ptr[group_id].N;
const index_t K = gemm_desc_ptr[group_id].K;
if(M * N * K == 0)
if(M == 0 || N == 0 || K == 0)
return;
const auto StrideAs = gemm_desc_ptr[group_id].StrideAs;

2
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_splitk_xdl_cshuffle_two_stage.hpp
View File

@@ -345,7 +345,7 @@ struct DeviceGroupedGemmMultipleDSplitKXdlCShuffleTwoStage
const index_t N = gemm_descs[i].N_;
const index_t K = gemm_descs[i].K_;
if(M * N * K == 0)
if(M == 0 || N == 0 || K == 0)
{
skipped_group_count_++;
continue;

2
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_multiple_d_xdl_cshuffle_tile_loop.hpp
View File

@@ -109,7 +109,7 @@ __global__ void
N = gemm_desc_ptr[group_id].N;
K = gemm_desc_ptr[group_id].K;
if(M * N * K == 0)
if(M == 0 || N == 0 || K == 0)
{
grid_size_grp = 0;
continue;

2
include/ck/tensor_operation/gpu/device/impl/device_grouped_gemm_xdl_fixed_nk.hpp
View File

@@ -68,7 +68,7 @@ __global__ void
const index_t N = gemm_desc_ptr[group_id].N;
const index_t K = gemm_desc_ptr[group_id].K;
if(M * N * K == 0)
if(M == 0 || N == 0 || K == 0)
return;
const auto StrideA = gemm_desc_ptr[group_id].StrideA;