This is an archive of the discontinued LLVM Phabricator instance.

[CUDA] Annotate all calls in CUDA device mode as convergent.
AbandonedPublic

Authored by jlebar on Feb 16 2016, 4:48 PM.

Details

Reviewers
majnemer
rnk
Summary

We need the notion of convergent functions -- which may expose
convergent behavior to callers -- and convergent calls, which are calls
where we would like to preserve convergent behavior in the callee, if
possible.

In CUDA device mode, all calls and functions are convergent. The
optimizer can then strip this away under some circumstances.

Diff Detail

Event Timeline

jlebar updated this revision to Diff 48127.Feb 16 2016, 4:48 PM
jlebar retitled this revision from to [CUDA] Annotate all calls in CUDA device mode as convergent..
jlebar updated this object.
jlebar added reviewers: rnk, majnemer.
jlebar added subscribers: tra, cfe-commits.
jlebar updated this revision to Diff 48143.Feb 16 2016, 7:11 PM

Fix typo.

rnk accepted this revision.Feb 18 2016, 5:52 PM
rnk edited edge metadata.

lgtm

This revision is now accepted and ready to land.Feb 18 2016, 5:52 PM
jlebar abandoned this revision.Feb 24 2016, 2:03 PM

Subsumed by D17056.