triton.experimental.gluon.language.amd.cdna4.async_copy.load_shared_relaxed

triton.experimental.gluon.language.amd.cdna4.async_copy.load_shared_relaxed(smem, layout, _semantic=None)

Load a tensor from shared memory with extra hints for the underlying compiler to avoid emitting unnecessary waits before loading from the target shared memory.

Parameters:
  • smem (shared_memory_descriptor) – Shared memory descriptor to load from.

  • layout (DistributedLayout) – The destination layout of the tensor.

Returns:

A Gluon tensor containing the loaded data.

Return type:

tensor