Tiny fix in using v_tiles[] index

This commit is contained in:
Qianfeng Zhang
2025-12-25 15:37:22 +00:00
parent 2d53d67b6d
commit df902c6a06

View File

@@ -552,7 +552,7 @@ struct HstuAttentionWithSoftmaxFwdPipelineQRKSVSTrLoad
__builtin_amdgcn_sched_barrier(0x00000001);
store_tile(v_lds_windows[number<(i_k1 + 3) % NumKVLdsBuffers>{}],
v_tiles[number<i_k1 + 1>{}],
v_tiles[number<(i_k1 + 1) % NumPrefetchV>{}],
partition_index);
__builtin_amdgcn_sched_barrier(0x00000001);