This is an archive of the discontinued LLVM Phabricator instance.

LoadStoreVectorizer: Skip optnone functions
ClosedPublic

Authored by arsenm on Apr 25 2016, 3:14 PM.

Diff Detail

Event Timeline

arsenm updated this revision to Diff 54926.Apr 25 2016, 3:14 PM
arsenm retitled this revision from to LoadStoreVectorizer: Skip optnone functions.
arsenm updated this object.
arsenm added reviewers: escha, resistor.
arsenm added a subscriber: llvm-commits.
arsenm updated this revision to Diff 56812.May 10 2016, 2:00 PM
arsenm edited edge metadata.

Use skipFunction

jlebar accepted this revision.Jun 13 2016, 2:06 PM
jlebar added a reviewer: jlebar.
jlebar added a subscriber: jlebar.
jlebar added inline comments.
test/Transforms/LoadStoreVectorizer/AMDGPU/optnone.ll
12

Nit, If you're paranoid, consider having a test right here that ensures that a non-optnone function with the same body gets vectorized. You never know. :)

This revision is now accepted and ready to land.Jun 13 2016, 2:06 PM
arsenm closed this revision.Jun 30 2016, 4:38 PM

r274296

MatzeB added a subscriber: MatzeB.Jun 30 2016, 4:41 PM
MatzeB added inline comments.
lib/Transforms/Vectorize/LoadStoreVectorizer.cpp
158

I would recommend to make skipFunction() the first thing in runOnFunction(). Otherwise I am not sure about the effects on opt bisection when the number of calls to skipFunction() depends on the number of NoImplicitFloat attributes in the IR...