[PATCH] D116496: [unroll] Prune all but first copy of invariant exit

Nikita Popov via Phabricator via llvm-commits llvm-commits at lists.llvm.org
Sun Jan 2 12:54:59 PST 2022


nikic added a comment.

In D116496#3216398 <https://reviews.llvm.org/D116496#3216398>, @reames wrote:

> In D116496#3216380 <https://reviews.llvm.org/D116496#3216380>, @nikic wrote:
>
>> This generally makes sense to me, but I think you need to adjust some of the tests to make the condition non-invariant, because I doubt that this is what they were intending to test. Replace an argument with a call return value.
>
> Did you have any particular tests in mind?  I skimmed the tests and don't see any where the change obviously breaks intent of the test author.

Always hard to say, but pr31718.ll is apparently a test for LCSSA and simplification interaction, which might not be representative if some of the exits (with LCSSA phi nodes) are dropped. I don't know what specifically runtime-loop-multiexit-dom-verify.ll is testing, but the relevant domtree pattern might not be present with some of the edges gone.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D116496/new/

https://reviews.llvm.org/D116496



More information about the llvm-commits mailing list