diff --git a/dsv4/kernels/attention/fmha.py b/dsv4/kernels/attention/fmha.py index d356ac25..6f9b3c98 100644 --- a/dsv4/kernels/attention/fmha.py +++ b/dsv4/kernels/attention/fmha.py @@ -430,6 +430,7 @@ class FmhaKernel: rP_bf16_frg[None, j].store(s_vec.to(self.q_dtype)) cute.copy(tiled_tmem_store, rP_words, tTMEM_STOREtP) + cute.arch.fence_view_async_tmem_store() else: # SMEM-P: compute softmax, write P to SMEM (TODO) for j in range(frg_cnt):