[PATCH] D138452: [ConstraintElim] Queue facts and checks directly.
Florian Hahn via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Mon Dec 5 09:04:29 PST 2022
fhahn added a comment.
In D138452#3971049 <https://reviews.llvm.org/D138452#3971049>, @nikic wrote:
> LGTM. A question I had is why we use this kind of upfront collection + sort approach at all. Wouldn't it be possible to do essentially the same in a single walk over the dominator tree?
Thanks! IIRC the main reason was to t make it easier to inject additional conditions for which no direct IR instruction exists, but that's not used in the version on `main` at the moment. I haven't thought a lot about what a single DT walk would look like, but a straight forward walk would probably be a recursive implementation and the worklist approach seems slightly simpler compared to that. I could look into the DT walk approach a bit more, if you think there's a significant benefit.
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D138452/new/
https://reviews.llvm.org/D138452
More information about the llvm-commits
mailing list