ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-15 18:42:06 +00:00
Code Issues Packages Projects Releases Wiki Activity

Use kM0=128 kN0=64 to completely remove the vgprs spilling

Browse Source
This commit is contained in:
Qianfeng Zhang
2025-04-15 15:08:46 +00:00
parent 3cd1b13e46
commit d1749b3aae
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
example/ck_tile/18_hstu_attention/hstu_attention_fwd_setting.hpp
View File

@@ -35,7 +35,7 @@ struct HstuAttentionFwdBlockTile<64>
template <>
struct HstuAttentionFwdBlockTile<128>
{
using type = ck_tile::sequence<128, 128, 32, 128, 32, 128>;
using type = ck_tile::sequence<128, 64, 32, 128, 32, 128>;
using gemm0_warps = ck_tile::sequence<4, 1, 1>;
using gemm1_warps = ck_tile::sequence<4, 1, 1>;
};