ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-04 21:51:28 +00:00
Code Issues Packages Projects Releases Wiki Activity

bug fix, clang format;

Browse Source
This commit is contained in:
aska-0096
2025-08-08 09:04:02 +00:00
parent 3b9fb6af38
commit 78edd7303b
16 changed files with 180 additions and 80 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -644,6 +644,8 @@ def get_fwd_blobs(kernel_filter : Optional[str], receipt, optdim_list, mask_impl
# non qr_async_trload only support km0=128 tile size when hdim is not 128
# non qr_async only support kn0=128 tile size when hdim is 128
continue
if pipeline.tag == 'qr_async_trload' and (((hdim, hdim_v) == (128, 128) and tile.F_bn0 == 128) or ((hdim, hdim_v) not in [(64, 64), (128, 128)])):
continue
# logits_soft_cap is only allowed if no bias
if not ((pipeline.F_logits == 't' and pipeline.F_bias == 'no') or pipeline.F_logits == 'f'):
continue