Page MenuHomePhabricator

AMDGPU: Add patterns for i32/i64 local atomic load/store
ClosedPublic

Authored by arsenm on Jun 21 2018, 8:08 AM.

Details

Reviewers
t-tye
kzhuravl
Summary

Not sure why the 32/64 split is needed in the atomic_load
store hierarchies. The regular PatFrags do this, but we don't
do it for the existing handling for global.

I'm not 100% sure if anything else is needed to handle the different orderings correctly

Diff Detail

Event Timeline

arsenm created this revision.Jun 21 2018, 8:08 AM
kzhuravl accepted this revision.Jun 21 2018, 8:55 AM

LGTM. Orderings are already handled correctly.

This revision is now accepted and ready to land.Jun 21 2018, 8:55 AM

Now this is implemented it may be worth converting the memorylegalizer tests from MIR to IR tests.

Now this is implemented it may be worth converting the memorylegalizer tests from MIR to IR tests.

I was planning to do it in a separate change once I get free.

arsenm closed this revision.Jun 22 2018, 1:44 AM

r335325