This is an archive of the discontinued LLVM Phabricator instance.

Correct openmp-offload-builders lld, openmp
AcceptedPublic

Authored by ronlieb on Aug 8 2021, 9:41 AM.

Details

Summary

Remove openmp from depends_on_projects, as it incorrectly builds amdgpu runtime stack corruption at runtime.
We need to use -DLLVM_ENABLE_RUNTIMES=openmp instead, and its presence conflicts if openmp is present.
Also adding "-DCLANG_DEFAULT_LINKER=lld" as it seems harmless.
Adding 'lld' to depends_on_projects so it gets built.

Diff Detail

Event Timeline

ronlieb created this revision.Aug 8 2021, 9:41 AM
ronlieb requested review of this revision.Aug 8 2021, 9:41 AM
Herald added a project: Restricted Project. · View Herald Transcript
This revision is now accepted and ready to land.Aug 9 2021, 8:28 AM
This revision was landed with ongoing or failed builds.Aug 9 2021, 8:38 AM
This revision was automatically updated to reflect the committed changes.
Meinersbur reopened this revision.EditedSep 9 2021, 12:37 PM

The enable_runtimes=['openmp'], parameter already sets -DLLVM_ENABLE_RUNTIMES=openmp (and ensures that changes to openmp triggers a build). Do not add it separately.

If you do not want support a LLVM_ENABLE_PROJECT=openmp build, remove the openmp-offload-amdgpu-project builder instead. There is no point in two builders doing the same thing.

This revision is now accepted and ready to land.Sep 9 2021, 12:37 PM

i reverted it, sent email to galina requesting update to master. and requested info on a better way to test locally ...