diff --git a/example/ck_tile/18_hstu_attention/hstu_attention_with_softmax_fwd_trload_pipeline.hpp b/example/ck_tile/18_hstu_attention/hstu_attention_with_softmax_fwd_trload_pipeline.hpp index ec409d7d0c..3230ad96df 100644 --- a/example/ck_tile/18_hstu_attention/hstu_attention_with_softmax_fwd_trload_pipeline.hpp +++ b/example/ck_tile/18_hstu_attention/hstu_attention_with_softmax_fwd_trload_pipeline.hpp @@ -552,7 +552,7 @@ struct HstuAttentionWithSoftmaxFwdPipelineQRKSVSTrLoad __builtin_amdgcn_sched_barrier(0x00000001); store_tile(v_lds_windows[number<(i_k1 + 3) % NumKVLdsBuffers>{}], - v_tiles[number{}], + v_tiles[number<(i_k1 + 1) % NumPrefetchV>{}], partition_index); __builtin_amdgcn_sched_barrier(0x00000001);