[Openmp-commits] [PATCH] D142512: [OpenMP][libomptarget] Fix mapping/prelock.cpp test

Kevin Sala via Phabricator via Openmp-commits openmp-commits at lists.llvm.org
Wed Jan 25 06:49:55 PST 2023

kevinsala added a comment.

I don't have a strong opinion regarding returning or not the agent pointer; I don't know which use cases may have (could be the use of `is_device_ptr`?). However, I have doubts about the correctness of passing a non-host pointer to a `map` clause and be treated as part of the host data environment.

> The lock function could operate in a "side effect" mode, where the user asks the runtime to lock memory and the ROCr keeps track of it. Does this make sense?

That makes sense to me. It's what we are assuming in the nextgen plugins. The user should pass the host pointer, the plugin detects that it's pinned memory, and we internally use the associated agent pointer for the asynchronous transfer.

  rG LLVM Github Monorepo



More information about the Openmp-commits mailing list