[Openmp-commits] [PATCH] D104691: [AMDGPU][Libomptarget] Move allow_access_to_all_gpu_agents to rtl.cpp

[Pushpinder Singh via Phabricator via Openmp-commits]

pdhaliwal added inline comments.


================
Comment at: openmp/libomptarget/plugins/amdgpu/impl/data.cpp:66
-
-  if (err == HSA_STATUS_SUCCESS) {
-    err = register_allocation(*ptr, size, DeviceType);
----------------
JonChesterfield wrote:
> it's not obvious this is dead - if Runtime::Malloc is called with DEVTYPE_CPU, but not from HostMalloc (where the follow on handling is added), then we miss the allow_access call. Can this Runtime::Malloc be made local to this TU, so that we can't accidentally call it instead of one of the DeviceMalloc or HostMalloc wrappers?
Malloc is already declared inside Runtime class as private static method.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D104691/new/

https://reviews.llvm.org/D104691