mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-06-05 20:55:59 +00:00
Add s_barrier to sync threads
This commit is contained in:
@@ -145,6 +145,7 @@ struct BlockFmhaFwdAppendKVPipeline
|
||||
rotary_sin_window,
|
||||
rotary_dim,
|
||||
thread_end);
|
||||
__builtin_amdgcn_s_barrier();
|
||||
}
|
||||
|
||||
if constexpr(kIsPagedKV)
|
||||
@@ -222,6 +223,7 @@ struct BlockFmhaFwdAppendKVPipeline
|
||||
|
||||
BlockRotaryEmbedding<RotaryEnum>::apply(
|
||||
q_tile, q_window, rotary_cos_window, rotary_sin_window, rotary_dim, thread_end);
|
||||
__builtin_amdgcn_s_barrier();
|
||||
|
||||
store_tile(q_dram_block_window, q_tile);
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user