[PATCH] D73410: [InstCombine] Push negation through multiply (PR44234)

Sat Jan 25 02:08:12 PST 2020

nikic marked an inline comment as done.
nikic added inline comments.

================
Comment at: llvm/test/Transforms/InstCombine/and-or-icmps.ll:207

 define void @simplify_before_foldAndOfICmps() {
 ; CHECK-LABEL: @simplify_before_foldAndOfICmps(
----------------
We have a regression here... the relevant difference is that the %B7 neg gets reassociated to act on %B23 instead, which is then pushed through to the %B11 udiv, which is a sext at that point and gets folded to zext. That somehow prevents other folds. As this is a fuzzer generated test case I'm not particularly worried about this, but can take a closer look if people think this is important.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D73410/new/

https://reviews.llvm.org/D73410