fix LDS read&write conflict when KPerBlock is 128 for pt2

8edf2a72 · danyao12 · ce2aa757 · 8edf2a72
Commit 8edf2a72 authored Jun 13, 2023 by danyao12
Show whitespace changes
Inline Side-by-side

Showing with 2 additions and 0 deletions

include/ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_pt7.hpp ...batched_multihead_attention_backward_xdl_cshuffle_pt7.hpp +2 -0

No files found.
--- a/include/ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_pt7.hpp
+++ b/include/ck/tensor_operation/gpu/grid/gridwise_batched_multihead_attention_backward_xdl_cshuffle_pt7.hpp
@@ -1785,6 +1785,8 @@ struct GridwiseBatchedMultiheadAttentionBackward_Xdl_CShuffle_V2
                                                    make_tuple(I0, I0, I0, I0, I0, I0),
                                                    y_dot_ygrad_thread_buf);

+            block_sync_lds();
+
            lse_thread_copy_global_to_vgpr.Run(lse_grid_desc_mb_m0_m1_m2_m3_m4,
                                               lse_grid_buf,
                                               lse_thread_desc_mb_m0_m1_m2_m3_m4,