diff --git a/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp b/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
index 91c872d709c..35298565e58 100644
--- a/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
+++ b/lib/Target/NVPTX/NVPTXLowerAggrCopies.cpp
@@ -87,14 +87,15 @@ void convertMemCpyToLoop(Instruction *ConvertedInst, Value *SrcAddr,
   // load from SrcAddr+LoopIndex
   // TODO: we can leverage the align parameter of llvm.memcpy for more efficient
   // word-sized loads and stores.
-  Value *Element = LoopBuilder.CreateLoad(
-      LoopBuilder.CreateGEP(LoopBuilder.getInt8Ty(), SrcAddr, LoopIndex),
-      SrcIsVolatile);
+  Value *Element =
+      LoopBuilder.CreateLoad(LoopBuilder.CreateInBoundsGEP(
+                                 LoopBuilder.getInt8Ty(), SrcAddr, LoopIndex),
+                             SrcIsVolatile);
   // store at DstAddr+LoopIndex
-  LoopBuilder.CreateStore(
-      Element,
-      LoopBuilder.CreateGEP(LoopBuilder.getInt8Ty(), DstAddr, LoopIndex),
-      DstIsVolatile);
+  LoopBuilder.CreateStore(Element,
+                          LoopBuilder.CreateInBoundsGEP(LoopBuilder.getInt8Ty(),
+                                                        DstAddr, LoopIndex),
+                          DstIsVolatile);
 
   // The value for LoopIndex coming from backedge is (LoopIndex + 1)
   Value *NewIndex =
@@ -225,7 +226,8 @@ void convertMemSetToLoop(Instruction *ConvertedInst, Value *DstAddr,
   LoopIndex->addIncoming(ConstantInt::get(CopyLen->getType(), 0), OrigBB);
 
   LoopBuilder.CreateStore(
-      SetValue, LoopBuilder.CreateGEP(SetValue->getType(), DstAddr, LoopIndex),
+      SetValue,
+      LoopBuilder.CreateInBoundsGEP(SetValue->getType(), DstAddr, LoopIndex),
       false);
 
   Value *NewIndex =
diff --git a/test/CodeGen/NVPTX/lower-aggr-copies.ll b/test/CodeGen/NVPTX/lower-aggr-copies.ll
index 68d9c2322a6..ef570982b80 100644
--- a/test/CodeGen/NVPTX/lower-aggr-copies.ll
+++ b/test/CodeGen/NVPTX/lower-aggr-copies.ll
@@ -18,9 +18,9 @@ entry:
 
 ; IR-LABEL:   @memcpy_caller
 ; IR:         loadstoreloop:
-; IR:         [[LOADPTR:%[0-9]+]] = getelementptr i8, i8* %src, i64
+; IR:         [[LOADPTR:%[0-9]+]] = getelementptr inbounds i8, i8* %src, i64
 ; IR-NEXT:    [[VAL:%[0-9]+]] = load i8, i8* [[LOADPTR]]
-; IR-NEXT:    [[STOREPTR:%[0-9]+]] = getelementptr i8, i8* %dst, i64
+; IR-NEXT:    [[STOREPTR:%[0-9]+]] = getelementptr inbounds i8, i8* %dst, i64
 ; IR-NEXT:    store i8 [[VAL]], i8* [[STOREPTR]]
 
 ; PTX-LABEL:  .visible .func (.param .b64 func_retval0) memcpy_caller
@@ -61,8 +61,8 @@ entry:
 ; IR-LABEL:   @memcpy_casting_caller
 ; IR:         [[DSTCAST:%[0-9]+]] = bitcast i32* %dst to i8*
 ; IR:         [[SRCCAST:%[0-9]+]] = bitcast i32* %src to i8*
-; IR:         getelementptr i8, i8* [[SRCCAST]]
-; IR:         getelementptr i8, i8* [[DSTCAST]]
+; IR:         getelementptr inbounds i8, i8* [[SRCCAST]]
+; IR:         getelementptr inbounds i8, i8* [[DSTCAST]]
 }
 
 define i8* @memset_caller(i8* %dst, i32 %c, i64 %n) #0 {
@@ -74,7 +74,7 @@ entry:
 ; IR-LABEL:   @memset_caller
 ; IR:         [[VAL:%[0-9]+]] = trunc i32 %c to i8
 ; IR:         loadstoreloop:
-; IR:         [[STOREPTR:%[0-9]+]] = getelementptr i8, i8* %dst, i64
+; IR:         [[STOREPTR:%[0-9]+]] = getelementptr inbounds i8, i8* %dst, i64
 ; IR-NEXT:    store i8 [[VAL]], i8* [[STOREPTR]]
 
 ; PTX-LABEL:  .visible .func (.param .b64 func_retval0) memset_caller(