[Fix][NPU] Fix performance degradation caused by serial execution of two-stage ACL op calls on graph (#40814)

This commit is contained in:
Kurkur
2026-09-28 09:32:16 +08:00
committed by GitHub
parent 0823651b0b
commit 6cbdfec453
@@ -178,8 +178,8 @@ class NPUCudaGraphBackend(BaseCudaGraphBackend):
update_future = self._update_executor.submit(
graph.update, cpu_update_input=cpu_update_input
)
update_future.result()
graph.replay()
update_future.result()
return self._outputs[shape_key]
def cleanup(self) -> None: