This is an archive of the discontinued LLVM Phabricator instance.

[mlir] Support broadcast dimensions in ProgressiveVectorToSCF
ClosedPublic

Authored by springerm on Apr 21 2021, 10:07 PM.

Details

Summary

This commit adds support for broadcast dimensions in permutation maps of vector transfer ops.

Also fixes a bug in VectorToSCF that generated incorrect in-bounds checks for broadcast dimensions.

Depends On D101013

Diff Detail

Event Timeline

springerm created this revision.Apr 21 2021, 10:07 PM
springerm requested review of this revision.Apr 21 2021, 10:07 PM
Herald added a project: Restricted Project. · View Herald TranscriptApr 21 2021, 10:07 PM

(no change)

springerm updated this revision to Diff 339872.Apr 22 2021, 9:33 PM

(no change)

(no changes)

Looks good, approval conditioned on -1 -> Optional, thanks!

mlir/lib/Conversion/VectorToSCF/ProgressiveVectorToSCF.cpp
75

Can we use Optional<unsigned> ?
I have a deep aversion for magic constants as we have quite some tech debt related to those remaining deep in the code.

77–82

trivial braces

547–548

same re Optional

This revision is now accepted and ready to land.Apr 22 2021, 11:50 PM
springerm updated this revision to Diff 339920.Apr 23 2021, 1:03 AM

Use llvm::Optional instead of -1.

springerm updated this revision to Diff 339947.Apr 23 2021, 2:00 AM
springerm marked 2 inline comments as done.

(new export)

This revision was landed with ongoing or failed builds.Apr 23 2021, 2:02 AM
This revision was automatically updated to reflect the committed changes.