[Openmp-commits] [PATCH] D132660: [openmp][amdgpu] Implement target_alloc_host as fine grain HSA memory

[Joseph Huber via Phabricator via Openmp-commits]

jhuber6 added a comment.

In D132660#3749006 <https://reviews.llvm.org/D132660#3749006>, @JonChesterfield wrote:

> The generic plugin maps them all to malloc, and I'm not convinced migratable means anything much in the context of the HSA model, so propose we treat 'shared' and 'host' as the same thing. It's unfortunate that 'shared' is also a totally different thing, possibly called 'dynamic_shared', which is not as easy to wire up.

It's just supposed to be an optimized memory location for DMA transfers AFAIK. To the user it behaves exactly like regular memory allocated via `malloc`, and since the generic plugin doesn't communicate via PCIe or other peripheral interface it's meaningless there.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D132660/new/

https://reviews.llvm.org/D132660