[PATCH] D13460: [IndVars] Don't break dominance in `eliminateIdentitySCEV`

[Mehdi AMINI via llvm-commits]

joker.eph added a subscriber: joker.eph.

================
Comment at: lib/Transforms/Utils/SimplifyIndVar.cpp:328
@@ +327,3 @@
+  // getSCEV(%M) == getSCEV(%X) == {0,+,1}, but %X does not dominate %M, and
+  // %M.replaceAllUsesWith(%X) would be incorrect.
+
----------------
Not sure if it is related, but about live-out value for loops, if the exit value is the same you could RAUW uses that are outside of the loop. Does it mean that we will pessimize this case? Is it handled somewhere else?


http://reviews.llvm.org/D13460