[PATCH] D18684: [SCEV] See through op.with.overflow intrinsics

[Sanjoy Das via llvm-commits]

sanjoy added inline comments.

================
Comment at: include/llvm/Analysis/ValueTracking.h:332
@@ +331,3 @@
+  /// not overflowing, \p II being an <op>.with.overflow intrinsic.
+  bool isOverflowIntrinsicNoWrap(IntrinsicInst *II, DominatorTree &DT);
+
----------------
regehr wrote:
> would something isOverflowGuard() work better?
I'd say that would be misleading since we're not checking if the arithmetic operation can overflow, but that the "arithmetic result" projection of the intrinsic can be treated as no-wrap.  So, as an example, the routine can (and will, in most cases) return true for an overflow intrinsic for which the "arithmetic result" projection is unused.


http://reviews.llvm.org/D18684