Page MenuHomePhabricator

[X86][AVX] Tag VPMOVSX/VPMOVZX ymm instructions as WriteShuffle256.
ClosedPublic

Authored by RKSimon on Apr 28 2018, 3:28 PM.

Details

Summary

These are more like cross-lane shuffles than regular shuffles - we already do this for AVX512 equivalents.

@GGanesh The Znver1 cases needed a little tweaking (typos in the instregex and missing load fold cases) - please can you take a look? Agner didn't have good numbers, but it seems to match what instlatx64 reports.

Diff Detail

Repository
rL LLVM

Event Timeline

RKSimon created this revision.Apr 28 2018, 3:28 PM

The Intel changes look good to me.

Sorry! Missed this thread completely!
LGTM!

This revision is now accepted and ready to land.May 7 2018, 10:50 AM
This revision was automatically updated to reflect the committed changes.