[llvm] [offload] Add olMemPrefetch to liboffload (PR #206752)

[Alex Duran via llvm-commits]

=?utf-8?q?Łukasz?= Plewa <lukasz.plewa at intel.com>
Message-ID:
In-Reply-To: <llvm.org/llvm/llvm-project/pull/206752 at github.com>


================

----------------
adurang wrote:

We currently support the following extension:
`int ompx_target_prefetch_shared_mem(size_t num_ptrs, void **ptrs, size_t *sizes, int device_num)`
which accepts multiple pointers. I suspect in some applications the already have these precomputed for other reasons (e.g. sparse structures) so it's more user-friendly to support those.

About the low-level APIs CUDA supports batched prefetch (cuMemPrefetchBatchAsync) so I wouldn't be surprised if other APIs would also support this in the future. 

https://github.com/llvm/llvm-project/pull/206752