[PATCH] D75757: [InstComine] Forego of one-use check in `(X - (X & Y)) --> (X & ~Y)` if Y is a constant

[Dave Green via Phabricator via llvm-commits]

dmgreen added a comment.

Like I said in D75145 <https://reviews.llvm.org/D75145>, this looks like it's causing a lot of other regressions, at least on the ARM benchmarks I'm running. I'm wondering if it's to do with the LTO pipeline we have. There's a chance that this is actually doing better, allowing loops after unrolling to be analysable, but now the compiler has a lot more opportunities to shoot itself in the foot.


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D75757/new/

https://reviews.llvm.org/D75757