The order of blocks in BlocksWithDeadTerminators may impact uselists.
They can differ from run to run for the same input. The attached test case is
only meant to show the problem to anyone interested. It is not a test I plan to
commit, at least not in its current form.
Details
Details
Diff Detail
Diff Detail
- Repository
- rL LLVM
- Build Status
Buildable 33979 Build 33978: arc lint + arc unit
Event Timeline
Comment Actions
Maybe worth explicitly mentioning in the commit message that AggressiveDeadCodeElimination::updateDeadRegions loops over BlocksWithDeadTerminators.
It's fine to commit non-determinism patches without a test, I think, because it's hard to write a useful regression test for these issues. I mean, you could theoretically CHECK the use-list itself, but that isn't really productive.