This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU/SI: Don't promote to vector if the load/store is volatile
ClosedPublic

Authored by cfang on May 11 2017, 2:44 PM.

Diff Detail

Event Timeline

cfang created this revision.May 11 2017, 2:44 PM
arsenm added inline comments.May 11 2017, 2:47 PM
lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
405

As a separate patch this should reject addrspacecasts

410–412

Can combine into return condition

cfang updated this revision to Diff 98817.May 12 2017, 12:35 PM

Update based on Matt;s comments.

Will disable the case for addrspacecast in promoting to vector in a separate patch.

arsenm accepted this revision.May 12 2017, 12:49 PM

LGTM

This revision is now accepted and ready to land.May 12 2017, 12:49 PM
This revision was automatically updated to reflect the committed changes.