This is an archive of the discontinued LLVM Phabricator instance.

[AArch64] enable (v)select to math TLI hook (WIP)
Needs ReviewPublic

Authored by spatel on Oct 12 2019, 7:26 AM.

Details

Summary

I added more select-to-shift folds to DAGCombiner in:
rL374397
rL374555
...and noticed that AArch64 does not override a TLI hook that was added with:
rL296977

Not sure if that is intentional or oversight, so I flipped the setting completely and updated some auto-generated regression tests. It's not a universal win, but likely worth doing? If so, there are a few more regression tests that need updating/inspection.

I don't intend to do more with this patch, so feel free to commandeer/close.

Diff Detail

Event Timeline

spatel created this revision.Oct 12 2019, 7:26 AM
Herald added a project: Restricted Project. · View Herald TranscriptOct 12 2019, 7:26 AM
xbolva00 set the repository for this revision to rG LLVM Github Monorepo.
dmgreen added subscribers: sanwou01, dmgreen.EditedMay 6 2020, 11:20 AM

I think I tried turning this on once upon a time, but the (very limited set of) results left me saying "meh". That was probably just codesize I was looking at at the time though.

@sanwou01 do you have times to give this a proper test? See if it helps or hurts on something bigger?

Edit: Oh, I see you are already looking!

I'm running SPEC CPU intrate for this patch as well as this patch in combination with D78880.

I'm running SPEC CPU intrate for this patch as well as this patch in combination with D78880.

What was the result of this experiment?