cccl
cccl copied to clipboard
`mdspan` cache policy accessors
Address https://github.com/NVIDIA/cccl/issues/2472
Add custom CUDA mdspan
accessors to enable cache operators.
The PR covers the following features:
- A
cache_policy_accessor
for load and store operation - A
cache_policy_accessor
for load-only operation - A
accessor_reference
for dispatching load and store operation in different ways - Low-level memory accesses rely on
cub::ThreadLoad
andcub::ThreadStore
(related issue https://github.com/NVIDIA/cccl/issues/2486 for improving the two methods)
(names to finalize later)