[Openmp-commits] [PATCH] D109875: [AMDGPU][Libomptarget] Refactor device/host memory pools setup
Pushpinder Singh via Phabricator via Openmp-commits
openmp-commits at lists.llvm.org
Fri Sep 17 01:52:59 PDT 2021
pdhaliwal marked an inline comment as done.
pdhaliwal added inline comments.
================
Comment at: openmp/libomptarget/plugins/amdgpu/src/rtl.cpp:636
return HSA_STATUS_SUCCESS;
}
----------------
JonChesterfield wrote:
> This is going to return success if it finds a pool for either kernarg or fine grain. I think we should only be claiming success if we've found (at least?) one for each. We could handle the control flow for that by initialising both pools to {0} and checking neither is {0} before returning, or by keeping track of whether both have been initialized
Thanks for pointing this out. It actually found an issue when only one memory pool is available and it is both kernarg and fine grained.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D109875/new/
https://reviews.llvm.org/D109875
More information about the Openmp-commits
mailing list