This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU/GlobalISel: Fix broken expansion of 64-bit vector sext_inreg
AbandonedPublic

Authored by arsenm on Nov 15 2022, 11:36 AM.

Details

Reviewers
Pierre-vh
foad
Petar.Avramovic
Group Reviewers
Restricted Project

Diff Detail

Event Timeline

arsenm created this revision.Nov 15 2022, 11:36 AM
Herald added a project: Restricted Project. · View Herald TranscriptNov 15 2022, 11:36 AM
arsenm requested review of this revision.Nov 15 2022, 11:36 AM
Herald added a project: Restricted Project. · View Herald TranscriptNov 15 2022, 11:36 AM
Herald added a subscriber: wdng. · View Herald Transcript
foad requested changes to this revision.Nov 15 2022, 12:28 PM

I'm pretty sure this was correct and you've broken it.

This revision now requires changes to proceed.Nov 15 2022, 12:28 PM
arsenm abandoned this revision.Nov 15 2022, 12:54 PM

I'm getting confused by the shift amount being subtracted by the amount to get the sext_inreg encoding

foad added a comment.Nov 15 2022, 1:20 PM

I try not to think about shifts at all. The second input operand is just the bit width to extend from.

I try not to think about shifts at all. The second input operand is just the bit width to extend from.

It's hard to not when trying to lift into IR for alive