This is an archive of the discontinued LLVM Phabricator instance.

[X86] When applying the shuffle-to-zero-extend transformation on floating point, bitcast to integer first.
ClosedPublic

Authored by aymanmus on Sep 13 2017, 6:12 AM.

Diff Detail

Repository
rL LLVM

Event Timeline

aymanmus created this revision.Sep 13 2017, 6:12 AM
RKSimon added inline comments.Sep 13 2017, 6:36 AM
test/CodeGen/X86/pr34577.ll
1 ↗(On Diff #115020)

No need for a new file - put this into vector-shuffle-combining-avx2.ll and call the test pr34577

aymanmus updated this revision to Diff 115180.Sep 14 2017, 1:13 AM
aymanmus marked an inline comment as done.
RKSimon accepted this revision.Sep 14 2017, 2:39 AM

LGTM (one minor) - I'm surprised how bad the AVX512 codegen is but that's a job for another day.....

test/CodeGen/X86/vector-shuffle-combining-avx2.ll
901 ↗(On Diff #115180)

Drop the #0 ?

This revision is now accepted and ready to land.Sep 14 2017, 2:39 AM
This revision was automatically updated to reflect the committed changes.