This is an archive of the discontinued LLVM Phabricator instance.

[Polly] Add invalidation of the loop SCEVs after merge block generation
ClosedPublic

Authored by mbdharan on May 15 2017, 6:44 AM.

Diff Detail

Repository
rL LLVM

Event Timeline

mbdharan created this revision.May 15 2017, 6:44 AM
Meinersbur edited edge metadata.May 15 2017, 8:19 AM

Hi Baranidharan,

thank you for the patch. Could you please add a test case as well (e.g. the .ll file from llvm.org/PR32536, minimized as far as possible and a CHECK:-line for something that is relevant. e.g. that the %num.0 = phi i32 [ %add, %for.body15 ], [ 0, %for.cond13.pre_entry_bb ] is in the right position)

lib/CodeGen/BlockGenerators.cpp
784 ↗(On Diff #98999)

Please add a newline before.

mbdharan updated this revision to Diff 99758.May 22 2017, 7:05 AM

Add changes to reflect the comments.

Meinersbur accepted this revision.May 22 2017, 7:35 AM

Hi Baranidharan,

the patch LGTM. An explanation into the doxygen of invalidateScalarEvolution about the added code and why it is there (like you explained in the summary) would be nice.

Do you have commit rights? If not, I'd commit for you.

lib/CodeGen/BlockGenerators.cpp
862 ↗(On Diff #99758)

Nit: Dot at the end of sentence.

This revision is now accepted and ready to land.May 22 2017, 7:35 AM
mbdharan updated this revision to Diff 99767.May 22 2017, 8:15 AM

Hi @Meinersbur, Updated the diff to according comments.

I don't have commit rights, so you'd have to commit for me.

This revision was automatically updated to reflect the committed changes.