This is an archive of the discontinued LLVM Phabricator instance.

Use proper dataflow ordering to speed convergence. This will converge the testcase on bug 26055 in 2 iterations. (data structures speedups to come to make even that faster)
ClosedPublic

Authored by dberlin on Jan 9 2016, 8:07 PM.

Diff Detail

Repository
rL LLVM

Event Timeline

dberlin updated this revision to Diff 44417.Jan 9 2016, 8:07 PM
dberlin retitled this revision from to Use proper dataflow ordering to speed convergence. This will converge the testcase on bug 26055 in 2 iterations. (data structures speedups to come to make even that faster).
dberlin updated this object.
dberlin added a subscriber: llvm-commits.
echristo accepted this revision.Jan 9 2016, 8:11 PM
echristo edited edge metadata.

LGTM, thanks. One silly nit.

-eric

lib/CodeGen/LiveDebugValues.cpp
366 ↗(On Diff #44417)

Nit: "when both worklists are empty"

This revision is now accepted and ready to land.Jan 9 2016, 8:11 PM
tvvikram edited edge metadata.Jan 10 2016, 1:06 AM

Maybe add an option to enable/disable this pass in this patch itself?

This revision was automatically updated to reflect the committed changes.

Let's commit your enable/disable patch separately.

I think we can get this to the point where we generally don't have to
disable it except for ridiculously sized programs.