For the complete documentation index, see llms.txt. Markdown versions of all pages are available by appending .md to any URL (e.g. /get-started.md).
Mojo function
cp_async_bulk_prefetch_tensor
def cp_async_bulk_prefetch_tensor[rank: Int](tma_descriptor: Pointer[NoneType], coords: IndexList[rank])
Initiates an asynchronous prefetch of a tensor tile into the L2 cache.
Reads the box that the TMA descriptor places at coords from global
memory into L2. There is no shared memory destination and no barrier.
This is a hint to the memory subsystem: a later TMA load of the same box
may hit L2.
Requires sm_90 or higher.
Parameters:
- rank (
Int): The dimensionality of the tensor (1 to 5).
Args:
- tma_descriptor (
Pointer[NoneType]): Pointer to the TMA descriptor of the tensor. - coords (
IndexList[rank]): The coordinates of the box, innermost dimension first.