diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 66a4b427a2d..bdc0f62f967 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -12263,6 +12263,8 @@ SDValue DAGCombiner::visitINSERT_VECTOR_ELT(SDNode *N) {
 
 SDValue DAGCombiner::ReplaceExtractVectorEltOfLoadWithNarrowedLoad(
     SDNode *EVE, EVT InVecVT, SDValue EltNo, LoadSDNode *OriginalLoad) {
+  assert(!OriginalLoad->isVolatile());
+
   EVT ResultVT = EVE->getValueType(0);
   EVT VecEltVT = InVecVT.getVectorElementType();
   unsigned Align = OriginalLoad->getAlignment();
@@ -12457,9 +12459,12 @@ SDValue DAGCombiner::visitEXTRACT_VECTOR_ELT(SDNode *N) {
       ISD::isNormalLoad(InVec.getNode()) &&
       !N->getOperand(1)->hasPredecessor(InVec.getNode())) {
     SDValue Index = N->getOperand(1);
-    if (LoadSDNode *OrigLoad = dyn_cast<LoadSDNode>(InVec))
-      return ReplaceExtractVectorEltOfLoadWithNarrowedLoad(N, VT, Index,
-                                                           OrigLoad);
+    if (LoadSDNode *OrigLoad = dyn_cast<LoadSDNode>(InVec)) {
+      if (!OrigLoad->isVolatile()) {
+        return ReplaceExtractVectorEltOfLoadWithNarrowedLoad(N, VT, Index,
+                                                             OrigLoad);
+      }
+    }
   }
 
   // Perform only after legalization to ensure build_vector / vector_shuffle
diff --git a/test/CodeGen/AMDGPU/extractelt-to-trunc.ll b/test/CodeGen/AMDGPU/extractelt-to-trunc.ll
index 3c5b59c321a..e160c20a03a 100644
--- a/test/CodeGen/AMDGPU/extractelt-to-trunc.ll
+++ b/test/CodeGen/AMDGPU/extractelt-to-trunc.ll
@@ -41,3 +41,37 @@ define void @bitcast_int_to_fpvector_extract_0(float addrspace(1)* %out, i64 add
    store float %extract, float addrspace(1)* %out
    ret void
 }
+
+; GCN-LABEL: {{^}}no_extract_volatile_load_extract0:
+; GCN: buffer_load_dwordx4
+; GCN: buffer_store_dword v
+define void @no_extract_volatile_load_extract0(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
+entry:
+  %vec = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
+  %elt0 = extractelement <4 x i32> %vec, i32 0
+  store i32 %elt0, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}no_extract_volatile_load_extract2:
+; GCN: buffer_load_dwordx4
+; GCN: buffer_store_dword v
+
+define void @no_extract_volatile_load_extract2(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
+entry:
+  %vec = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
+  %elt2 = extractelement <4 x i32> %vec, i32 2
+  store i32 %elt2, i32 addrspace(1)* %out
+  ret void
+}
+
+; GCN-LABEL: {{^}}no_extract_volatile_load_dynextract:
+; GCN: buffer_load_dwordx4
+; GCN: buffer_store_dword v
+define void @no_extract_volatile_load_dynextract(i32 addrspace(1)* %out, <4 x i32> addrspace(1)* %in, i32 %idx) {
+entry:
+  %vec = load volatile <4 x i32>, <4 x i32> addrspace(1)* %in
+  %eltN = extractelement <4 x i32> %vec, i32 %idx
+  store i32 %eltN, i32 addrspace(1)* %out
+  ret void
+}