triton.experimental.gluon.language.amd.cdna5.tdm.async_store

triton.experimental.gluon.language.amd.cdna5.tdm.async_store(dest: tensor_descriptor, offsets: List[ttgl.constexpr | ttgl.tensor] = None, src: shared_memory_descriptor = None, mbarrier: shared_memory_descriptor = None, cache_modifier='', _semantic=None) → None

Store a block of tensor specified in tensor descriptor from shared memory to global memory asynchronously.

See async_load() for how the descriptor is positioned (a prior update_tensor_descriptor() or the offsets convenience). Stores take no pred.

Parameters:

• dest (tensor_descriptor) – the destination tensor descriptor.

• offsets (List[int], optional) – if given, the offsets from the base pointer used to position the descriptor before the store.

• src (shared_memory_descriptor) – the shared memory source to load the data.

• mbarrier (shared_memory_descriptor, optional) – The barrier object to signal “arrive” on.