[PATCH] D87778: [MemorySSA] Be more conservative when traversing MemoryPhis.

[Alina Sbirlea via Phabricator via llvm-commits]

asbirlea added a comment.

I'm in favor of being conservative here and checking for only constant indexes on the geps.



================
Comment at: llvm/include/llvm/Analysis/MemorySSA.h:1262
         if (Translator.getAddr() != Location.Ptr) {
-          CurrentPair.second = Location.getWithNewPtr(Translator.getAddr());
+          CurrentPair.second = Location.getWithNewPtr(Translator.getAddr())
+                                   .getWithNewSize(LocationSize::unknown());
----------------
I'm not clear that the size should be always unknown after phi translation with different ptr, but feel free to correct me.
Here's my suggestion.
```
if (Translator.getAddr() != Location.Ptr) {
  CurrentPair.second = CurrentPair.second.getWithNewPtr(Translator.getAddr());
  if (PerformedPhiTranslation)
            *PerformedPhiTranslation = true;
} else {
  CurrentPair.second = Location.getWithNewSize(LocationSize::unknown());
}
```


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D87778/new/

https://reviews.llvm.org/D87778