ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-14 10:09:41 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit 'ef72a4b9bc2e5ddc63d9138cae4e5eba23d35b16' into develop

Browse Source
This commit is contained in:
github-actions[bot]
2025-05-09 08:07:08 +00:00
parent b39f59767b
commit 2ff2610908
2 changed files with 14 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
include/ck_tile/ops/gemm/warp/warp_gemm.hpp
View File

@@ -97,12 +97,20 @@ using WarpGemmMfmaF16F16F32M64N4K16 = WarpGemmImpl<WarpGemmAtrributeMfmaIterateK
4>>;
// fp16 2:4 structured sparsity
#if defined(__gfx94__) || defined(__gfx95__)
using WarpGemmSmfmacF16F16F32M32N32K16 = WarpGemmSmfmacImpl<WarpGemmAttributeSmfmac<
WarpGemmAttributeSmfmacImplF16F16F32M32N32K16<WGAttrCtlEnum::Default_>>>;
using WarpGemmSmfmacF16F16F32M16N16K32 = WarpGemmSmfmacImpl<WarpGemmAttributeSmfmac<
WarpGemmAttributeSmfmacImplF16F16F32M16N16K32<WGAttrCtlEnum::Default_>>>;
#else // gfx 90a does not support smfmac
using WarpGemmSmfmacF16F16F32M32N32K16 = WarpGemmImpl<WarpGemmAtrributeMfmaIterateK<
WarpGemmAttributeMfmaImplF16F16F32M32N32K8<WGAttrCtlEnum::Default_>,
2>>;
using WarpGemmSmfmacF16F16F32M16N16K32 = WarpGemmImpl<WarpGemmAtrributeMfmaIterateK<
WarpGemmAttributeMfmaImplBf16Bf16F32M16N16K16<WGAttrCtlEnum::Default_>,
2>>;
#endif
// bf16
using WarpGemmMfmaBf16Bf16F32M32N32K8 = WarpGemmImpl<

6
tile_engine/ops/gemm/gemm_instance_builder.py
View File

@@ -535,7 +535,11 @@ struct GemmDispatcher {
((tile[6] == 32 and tile[7] == 32 and tile[8] == 16) or
(tile[6] == 16 and tile[7] == 16 and tile[8] == 32))
content += f"""
run_kernel<{group}::GemmKernel<{tile[0]}, {tile[1]}, {tile[2]}, {tile[3]}, {tile[4]}, {tile[5]}, {tile[6]}, {tile[7]}, {tile[8]}, {BOOL_MAP(sparse)}>>(c_m_n_dev_buf, c_m_n_host_result, c_m_n_dev_result, verify, args, stream);"""
#if defined(__gfx908__)
run_kernel<{group}::GemmKernel<{tile[0]}, {tile[1]}, {tile[2]}, {tile[3]}, {tile[4]}, {tile[5]}, {tile[6]}, {tile[7]}, {tile[8]}, {BOOL_MAP(False)}>>(c_m_n_dev_buf, c_m_n_host_result, c_m_n_dev_result, verify, args, stream);
#else
run_kernel<{group}::GemmKernel<{tile[0]}, {tile[1]}, {tile[2]}, {tile[3]}, {tile[4]}, {tile[5]}, {tile[6]}, {tile[7]}, {tile[8]}, {BOOL_MAP(sparse)}>>(c_m_n_dev_buf, c_m_n_host_result, c_m_n_dev_result, verify, args, stream);
#endif"""
content += f"""
}} else {{"""
for tile in tile_params: