[PATCH] [LAA] Try to prove non-wrapping of pointers if SCEV cannot

[Andrew Trick]

LGTM. Sorry for the delay.

It's not clear to me how comprehensive this analysis is, but at least it covers some important cases.


================
Comment at: lib/Analysis/LoopAccessAnalysis.cpp:511
@@ +510,3 @@
+                           ScalarEvolution *SE, const Loop *L) {
+  // FIXME: This should probably only return true for NUW.
+  if (AR->getNoWrapFlags(SCEV::NoWrapMask))
----------------
I agree with you here. If NSW/NW are valid conditions, it's not obvious to me and the original author needs to explain it.

http://reviews.llvm.org/D10472

EMAIL PREFERENCES
  http://reviews.llvm.org/settings/panel/emailpreferences/