Page MenuHomePhabricator

[x86] add AVX2 cost model entries for maxnum of 256-bit vectors
ClosedPublic

Authored by spatel on Nov 2 2020, 7:46 AM.

Details

Summary

As noticed in D90554 , the AVX2 costs for 256-bit vectors did not include FMAXNUM entries, so we fell back to AVX1 which assumes those ops will be split into 128-bit halves.

Diff Detail

Event Timeline

spatel created this revision.Nov 2 2020, 7:46 AM
Herald added a project: Restricted Project. · View Herald TranscriptNov 2 2020, 7:46 AM
spatel requested review of this revision.Nov 2 2020, 7:46 AM
RKSimon accepted this revision.Nov 2 2020, 8:07 AM

LGTM - cheers

This revision is now accepted and ready to land.Nov 2 2020, 8:07 AM
pengfei added inline comments.Nov 2 2020, 5:08 PM
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
2531

Nit: the comments should be MAXSS + CMPUNORDSS.
The same below.

spatel added inline comments.Nov 3 2020, 5:15 AM
llvm/lib/Target/X86/X86TargetTransformInfo.cpp
2531

Thanks - updated with 9af561e.