mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-04 21:51:28 +00:00
fix the lds alignment caused performance regression
This commit is contained in:
@@ -656,7 +656,7 @@ def get_fmha_fwd_tile_dict_from_dtype(dtype : str) -> Optional[dict]:
|
||||
# '64': FmhaFwdTileSize(64, 64, 64, 64, 64, 64, 4, 1, 1, 4, 1, 1, 16, 16, 32, 16, 16, 32, -1),
|
||||
},
|
||||
'128': {
|
||||
'16': FmhaFwdTileSize(16, 32, 128, 128, 32, 128, 1, 1, 1, 1, 1, 1, 16, 16, 32, 16, 16, 32, -1),
|
||||
'16': FmhaFwdTileSize(16, 32, 64, 128, 32, 128, 1, 1, 1, 1, 1, 1, 16, 16, 32, 16, 16, 32, -1),
|
||||
'32': FmhaFwdTileSize(32, 32, 128, 128, 32, 128, 1, 1, 1, 1, 1, 1, 32, 32, 16, 32, 32, 16, -1),
|
||||
# # '64': FmhaFwdTileSize(64, 64, 64, 128, 64, 128, 4, 1, 1, 4, 1, 1, 16, 16, 32, 16, 16, 32, -1),
|
||||
},
|
||||
|
||||
Reference in New Issue
Block a user