triton.experimental.gluon.language.nvidia.rubin.async_store

triton.experimental.gluon.language.nvidia.rubin.async_store(dst, value, mbarrier, _semantic=None)

Store a tensor to shared memory asynchronously and signal an mbarrier on completion. Requires a CTA cluster with at least two CTAs.

Parameters:

• dst (shared_memory_descriptor) – Destination shared memory descriptor.

• value (tensor) – Tensor whose contents to store.

• mbarrier (shared_memory_descriptor) – Barrier signaled when the store completes.