This is an archive of the discontinued LLVM Phabricator instance.

[OpenMP][AMDGPU] Optimize the linked in math libraries
ClosedPublic

Authored by jdoerfert on Jan 9 2022, 6:20 PM.

Details

Summary

Once we linked in math files, potentially even if we link in only other
"system libraries", we want to optimize the code again. This is not only
reasonable but also helps to hide various problems with the missing
attribute annotations in the math libraries.

Diff Detail

Event Timeline

jdoerfert created this revision.Jan 9 2022, 6:20 PM
jdoerfert requested review of this revision.Jan 9 2022, 6:20 PM
Herald added a project: Restricted Project. · View Herald TranscriptJan 9 2022, 6:20 PM

This fixes/hides the last remaining OvO failures with upstream on an AMD MI100 compiled with -O3.

JonChesterfield accepted this revision.Jan 10 2022, 3:34 AM

Using the optimiser to hide errors in how we're pulling in libm is clearly not right, but it leaves us less obviously broken than we are at present.

This revision is now accepted and ready to land.Jan 10 2022, 3:34 AM
This revision was landed with ongoing or failed builds.Jan 19 2022, 9:36 PM
This revision was automatically updated to reflect the committed changes.