triton.experimental.gluon.language.nvidia.hopper.fence_async_shared

triton.experimental.gluon.language.nvidia.hopper.fence_async_shared(cluster=False, _semantic=None)

Order generic-proxy and asynchronous-proxy shared memory accesses.

This fence does not wait for asynchronous operations to complete.

Parameters:

cluster (bool) – Whether to fence across cluster. Defaults to False.