This is an archive of the discontinued LLVM Phabricator instance.

[X86] Split masked integer vector stores into vXi32/vXi64 variants (PR45975). NFC
ClosedPublic

Authored by andreadb on May 19 2020, 3:35 AM.

Details

Summary

This effectively splits the scheduling WriteVecMaskedStore(Y) classes
into four different classes (one per each variant).

VecMaskedStores are now correctly marked as 'unsupported' by the bdver2
and btver2 models (since those are only used for AVX2 VPMASKMOV variants).

No functional change intended.

Diff Detail

Event Timeline

andreadb created this revision.May 19 2020, 3:35 AM
Herald added a project: Restricted Project. · View Herald TranscriptMay 19 2020, 3:35 AM
Herald added a subscriber: hiraditya. · View Herald Transcript
andreadb updated this revision to Diff 264841.May 19 2020, 3:43 AM

Patch updated. This time with context.

RKSimon added inline comments.May 19 2020, 6:51 AM
llvm/lib/Target/X86/X86SchedBroadwell.td
243

TBH it's better if we can keep the integer vector entries all together, not put in the float vector sections - same for all other models

andreadb updated this revision to Diff 264913.May 19 2020, 7:56 AM

Addressed review comments.

andreadb marked an inline comment as done.May 19 2020, 7:56 AM
RKSimon accepted this revision.May 19 2020, 8:44 AM

LGTM - cheers

This revision is now accepted and ready to land.May 19 2020, 8:44 AM
This revision was automatically updated to reflect the committed changes.