ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-07-17 17:19:12 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix hd32 error and boost performance

Browse Source
This commit is contained in:
danyao12
2024-07-25 16:16:30 +08:00
parent b2510c0541
commit dcc3593fe4
3 changed files with 123 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/ck_tile/01_fmha/codegen/ops/fmha_bwd.py
View File

@@ -448,7 +448,7 @@ class FmhaBwdDQDKDVKernel:
def get_fmha_bwd_dq_dk_dv_tile_ppl_dict_from_dtype(dtype : str) -> Optional[dict]:
if dtype == 'fp16' or dtype == 'bf16':
return {
# '32' : [FmhaBwdDQDKDVTileSize( 64, 64, 32, 64, 32, 64, 64, 32, 32, 1, 2, 1, 2, 1, 1, 2, 1, 1, 32, 32, 16, 32, 32, 16, 1),
# '32' : [FmhaBwdDQDKDVTileSize( 32, 128, 32, 32, 32, 32, 64, 32, 32, 1, 4, 1, 4, 1, 1, 2, 2, 1, 16, 16, 32, 16, 16, 16, 1),
# "kr_ktr_vr"],
'64' : [FmhaBwdDQDKDVTileSize( 64, 128, 64, 64, 64, 64, 64, 64, 64, 1, 4, 1, 4, 1, 1, 2, 2, 1, 32, 32, 16, 32, 32, 16, 1),
"kr_ktr_vr"],