This is an archive of the discontinued LLVM Phabricator instance.

[AMDGPU] Split unaligned LDS access instead of scalarizing
ClosedPublic

Authored by rampitec on Apr 7 2022, 11:48 AM.

Details

Summary

There is no need to fully scalarize an unaligned operation in
some case, just split it to alignment.

Diff Detail

Event Timeline

rampitec created this revision.Apr 7 2022, 11:48 AM
Herald added a project: Restricted Project. · View Herald TranscriptApr 7 2022, 11:48 AM
rampitec requested review of this revision.Apr 7 2022, 11:48 AM
Herald added a project: Restricted Project. · View Herald TranscriptApr 7 2022, 11:48 AM
Herald added a subscriber: wdng. · View Herald Transcript
rampitec added inline comments.Apr 7 2022, 12:08 PM
llvm/test/CodeGen/AMDGPU/load-local.96.ll
355

This is the most essential effect. Here and in some other places.

arsenm accepted this revision.Apr 7 2022, 2:08 PM

expandUnalignedLoad/Store should probably be smarter

This revision is now accepted and ready to land.Apr 7 2022, 2:08 PM
This revision was landed with ongoing or failed builds.Apr 7 2022, 2:27 PM
This revision was automatically updated to reflect the committed changes.