ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-04-20 06:49:15 +00:00
Code Issues Packages Projects Releases Wiki Activity

flashattention fwd add (80, 96) instance (#3415)

Browse Source 
* add hdim (96,96) instance

* change to (80,96)

* format py

* remove 96 in optdim

* when N=6 change to llvm_amdgcn_raw_buffer_load_i32x3
This commit is contained in:
ltqin
2025-12-18 01:16:11 +08:00
committed by GitHub
parent fe3d52d9b0
commit 92653168c2
6 changed files with 127 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
include/ck_tile/ops/fmha/pipeline/tile_fmha_shape.hpp
View File

@@ -12,6 +12,8 @@ static CK_TILE_HOST_DEVICE constexpr index_t ceil_to_qualified_tile_length()
{
if constexpr(Headdim == 48)
return 48;
else if constexpr(Headdim == 80)
return 96;
else if constexpr(Headdim == 96)
return 128;
else if constexpr(Headdim == 160)