[PATCH] D11220: Correct lowering of memmove in NVPTX

[Jingyue Wu]

jingyue added inline comments.

================
Comment at: lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp:125
@@ +124,3 @@
+// }
+void convertMemMoveToLoop(Instruction *splitAt, Value *srcAddr, Value *dstAddr,
+                          Value *len, bool srcVolatile, bool dstVolatile,
----------------
Argument names start with upper case. 

================
Comment at: lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp:168
@@ +167,3 @@
+  Value *Element = LoopBuilder.CreateLoad(
+      LoopBuilder.CreateInBoundsGEP(srcAddr, {IndexPtr}), "element");
+  LoopBuilder.CreateStore(Element,
----------------
CreateInBoundsGEP(srcAddr, IndexPtr) should work now. 

================
Comment at: lib/Target/NVPTX/NVPTXTargetMachine.cpp:69
@@ -67,2 +68,3 @@
+  PassRegistry &PR = *PassRegistry::getPassRegistry();
   initializeNVVMReflectPass(*PassRegistry::getPassRegistry());
   initializeGenericToNVVMPass(*PassRegistry::getPassRegistry());
----------------
Do you intend to use `PR` instead? 

================
Comment at: test/CodeGen/NVPTX/lower-aggr-copies.ll:21
@@ +20,3 @@
+; IR:         loadstoreloop:
+; IR:         [[LOADPTR:%[0-9]+]] = getelementptr i8, i8* %src, i64
+; IR-NEXT:    [[VAL:%[0-9]+]] = load i8, i8* [[LOADPTR]]
----------------
How come we don't see inbounds here? You created them using `CreateInBoundsGEP`. 


Repository:
  rL LLVM

http://reviews.llvm.org/D11220