diff --git a/dsv4/kernels/attention/fmha.py b/dsv4/kernels/attention/fmha.py index 1f65f246..fcd7ce93 100644 --- a/dsv4/kernels/attention/fmha.py +++ b/dsv4/kernels/attention/fmha.py @@ -265,7 +265,8 @@ class FmhaKernel: cute.arch.fence_proxy("async.shared", space="cta") softmax_done_bar.arrive() - print(f"[SMEM-P DEEP DEBUG] === END ===") cute.arch.fence_view_async_tmem_store() + print(f"[SMEM-P DEEP DEBUG] === END ===") + cute.arch.fence_view_async_tmem_store() kvh.release() acc_pipe.producer_commit(acc_st); acc_st.advance() final_o_bar.arrive()