[PATCH] D20128: MemCpyOpt: combine local load/store sequences into memcpy.

[Tim Northover via llvm-commits]

t.p.northover added inline comments.

================
Comment at: lib/Transforms/Scalar/MemCpyOptimizer.cpp:577
@@ +576,3 @@
+
+      if (NextLoad || MTI->isVolatile() || !isa<ConstantInt>(MTI->getLength()))
+        break;
----------------
majnemer wrote:
> Atomic?
I don't think MemTransferInsts (i.e. @llvm.mem* calls) can be atomic, can they? For normal loads & stores both volatile and atomic are covered by the isSimple check.


Repository:
  rL LLVM

http://reviews.llvm.org/D20128