This is an archive of the discontinued LLVM Phabricator instance.

[ValueTracking] Cleanup condition for dup incoming edge for phi nodes; NFC
AbandonedPublic

Authored by goldstein.w.n on Aug 13 2023, 10:33 AM.

Details

Reviewers
nikic

Diff Detail

Event Timeline

goldstein.w.n created this revision.Aug 13 2023, 10:33 AM
Herald added a project: Restricted Project. · View Herald TranscriptAug 13 2023, 10:33 AM
goldstein.w.n requested review of this revision.Aug 13 2023, 10:33 AM
Herald added a project: Restricted Project. · View Herald TranscriptAug 13 2023, 10:33 AM
This revision is now accepted and ready to land.Aug 14 2023, 2:57 AM
goldstein.w.n abandoned this revision.Aug 15 2023, 10:41 PM

This isn't NFC. It breaks some tests in clang. Going to drop it.

Can you extract the failing clang test into an LLVM test? I don't understand why this isn't NFC.

Can you extract the failing clang test into an LLVM test? I don't understand why this isn't NFC.

Clang :: CodeGenCXX/temp-order.cpp
Clang :: Headers/__clang_hip_math.hip

I looked into temp-order.cpp and its properly different.
All the print calls disappear.

It looks like the issue occurs during inlining. I expect what happens is that not the terminator hasn't been remapped yet and references blocks from the old function. This seems like something that is just waiting to blow up...