This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU/SI: Don't run SLSR and NaryReassociate passes
AbandonedPublic

Authored by tstellarAMD on Aug 31 2016, 12:52 PM.

Details

Reviewers
arsenm
Summary

These can cause infinite compile times like for example:
https://llvm.org/bugs/show_bug.cgi?id=28721

Diff Detail

Event Timeline

tstellarAMD retitled this revision from to AMDGPU/SI: Don't run SLSR and NaryReassociate passes.
tstellarAMD updated this object.
tstellarAMD added a reviewer: arsenm.
tstellarAMD added a subscriber: llvm-commits.
tstellarAMD added a project: Restricted Project.Aug 31 2016, 12:52 PM
arsenm requested changes to this revision.Aug 31 2016, 12:53 PM
arsenm edited edge metadata.

The bug should be fixed instead of this. However I noticed this is run for -O0 which should be fixed

This revision now requires changes to proceed.Aug 31 2016, 12:53 PM

The bug should be fixed instead of this. However I noticed this is run for -O0 which should be fixed

I agree, but these bugs are preventing us from being able to compiler some programs, and I'm not sure when someone will be able to look into fixing this bug.

The bug should be fixed instead of this. However I noticed this is run for -O0 which should be fixed

The bug should be fixed instead of this. However I noticed this is run for -O0 which should be fixed

I agree, but these bugs are preventing us from being able to compiler some programs, and I'm not sure when someone will be able to look into fixing this bug.

With just ripping this out, it's more likely that nobody will ever get around to re-enabling these. How about just adding a flag to allow disabling this so those programs can work around it for now? Alternatively this bug should be higher priority

tstellarAMD abandoned this revision.Nov 1 2016, 10:46 AM