This is an archive of the discontinued LLVM Phabricator instance.

[MLIR][SCF] Find all innermost loops for parallel loop tiling
ClosedPublic

Authored by frgossen on Nov 26 2020, 9:09 AM.

Details

Summary

Overcome the assumption that parallel loops are only nested in other parallel
loops.

Diff Detail

Event Timeline

frgossen created this revision.Nov 26 2020, 9:09 AM
frgossen requested review of this revision.Nov 26 2020, 9:09 AM
mehdi_amini accepted this revision.Nov 26 2020, 4:53 PM

You have two typos in the commit message: prarallel in the title and assumtion in the description.

LGTM otherwise.

mlir/lib/Dialect/SCF/Transforms/ParallelLoopTiling.cpp
131

Nit: SmallVectorImpl for arguments.

135

Can we spell out the types here? They are fairly trivial I think (if it isn't Region& and Block& then auto would be misleading anyway)x

This revision is now accepted and ready to land.Nov 26 2020, 4:53 PM
frgossen updated this revision to Diff 307983.Nov 27 2020, 1:06 AM
frgossen marked 2 inline comments as done.

Address comments

frgossen retitled this revision from [MLIR][SCF] Find all innermost loops for prarallel loop tiling to [MLIR][SCF] Find all innermost loops for parallel loop tiling.Nov 27 2020, 1:08 AM
frgossen edited the summary of this revision. (Show Details)