ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-29 03:07:02 +00:00
Code Issues Packages Projects Releases Wiki Activity

modified the tile size to 256, 128x128x128.

Browse Source
This commit is contained in:
mtgu0705
2025-02-13 15:53:21 +08:00
parent 85a7aa89d5
commit 2d25691347
1 changed files with 7 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
example/01_gemm/gemm_xdl_fp8_pk_i4_bpreshuffle_v3.cpp
View File

@@ -36,16 +36,16 @@ using DeviceGemmV2Instance =
ALayout, BLayout, CLayout,
ADataType, BDataType, CDataType, AccDataType, CShuffleDataType,
AElementOp, BElementOp, CElementOp, GemmDefault,
128,
16, 128,
256,
128, 128,
KPerBlock, 16, 32,
16, 16,
1, 4,
S<8, 16, 1>, S<1, 0, 2>, S<1, 0, 2>,
32, 32,
4, 1,
S<8, 32, 1>, S<1, 0, 2>, S<1, 0, 2>,
2, 16, 16, 0,
S<4, 32, 1>, S<1, 0, 2>, S<1, 0, 2>,
S<4, 64, 1>, S<1, 0, 2>, S<1, 0, 2>,
2, 32, 32, 0,
1, 1, S<1, 16, 1, 8>, 4,
1, 1, S<1, 32, 1, 8>, 4,
ck::BlockGemmPipelineScheduler::Intrawave, ck::BlockGemmPipelineVersion::v1, F8, F8, PermuteA, PermuteB>;
// clang-format on