This is an archive of the discontinued LLVM Phabricator instance.

Allow targets to opt-in to codegen in SCC order
ClosedPublic

Authored by arsenm on Mar 30 2017, 2:49 PM.

Details

Reviewers
MatzeB
Summary

Decouple this setting from EnableIRPA.

To support function calls on AMDGPU, it is necessary to
report the global register usage throughout the kernel's
call graph, so callees need to be handled first.

Diff Detail

Event Timeline

arsenm created this revision.Mar 30 2017, 2:49 PM
arsenm updated this revision to Diff 93592.Mar 30 2017, 11:00 PM

Fix missing piece

MatzeB edited edge metadata.Apr 4 2017, 11:34 AM

I'd lean towards putting the option into TargetPassConfig instead of TargetMachine, what do you think?

I'd lean towards putting the option into TargetPassConfig instead of TargetMachine, what do you think?

I wasn't sure on this. Is there any reason passes would need to check this? If not, then TargetPassConfig is probably the right place.

I'd lean towards putting the option into TargetPassConfig instead of TargetMachine, what do you think?

I wasn't sure on this. Is there any reason passes would need to check this? If not, then TargetPassConfig is probably the right place.

I can't think of a reason why passes would want to know this (and they certainly shouldn't rely on a specific order).

arsenm updated this revision to Diff 94135.Apr 4 2017, 4:05 PM

Move to TargetPassConfig

MatzeB accepted this revision.Apr 4 2017, 4:20 PM

LGTM

This revision is now accepted and ready to land.Apr 4 2017, 4:20 PM
arsenm closed this revision.Apr 4 2017, 4:57 PM

r299487