Index: lib/Target/R600/AMDGPUISelLowering.cpp
===================================================================
--- lib/Target/R600/AMDGPUISelLowering.cpp
+++ lib/Target/R600/AMDGPUISelLowering.cpp
@@ -719,32 +719,46 @@
   }
 
   SDLoc DL(Op);
-  const SDValue &Value = Store->getValue();
+  SDValue Value = Store->getValue();
   EVT VT = Value.getValueType();
-  const SDValue &Ptr = Store->getBasePtr();
+  EVT ElemVT = VT.getVectorElementType();
+  SDValue Ptr = Store->getBasePtr();
   EVT MemEltVT = MemVT.getVectorElementType();
   unsigned MemEltBits = MemEltVT.getSizeInBits();
   unsigned MemNumElements = MemVT.getVectorNumElements();
-  EVT PackedVT = EVT::getIntegerVT(*DAG.getContext(), MemVT.getSizeInBits());
-  SDValue Mask = DAG.getConstant((1 << MemEltBits) - 1, PackedVT);
+  unsigned PackedSize = MemVT.getStoreSizeInBits();
+  SDValue Mask = DAG.getConstant((1 << MemEltBits) - 1, MVT::i32);
+
+  assert(Value.getValueType().getScalarSizeInBits() >= 32);
 
   SDValue PackedValue;
   for (unsigned i = 0; i < MemNumElements; ++i) {
-    EVT ElemVT = VT.getVectorElementType();
     SDValue Elt = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, DL, ElemVT, Value,
                               DAG.getConstant(i, MVT::i32));
-    Elt = DAG.getZExtOrTrunc(Elt, DL, PackedVT);
-    Elt = DAG.getNode(ISD::AND, DL, PackedVT, Elt, Mask);
-    SDValue Shift = DAG.getConstant(MemEltBits * i, PackedVT);
-    Elt = DAG.getNode(ISD::SHL, DL, PackedVT, Elt, Shift);
+    Elt = DAG.getZExtOrTrunc(Elt, DL, MVT::i32);
+    Elt = DAG.getNode(ISD::AND, DL, MVT::i32, Elt, Mask); // getZeroExtendInReg
+
+    SDValue Shift = DAG.getConstant(MemEltBits * i, MVT::i32);
+    Elt = DAG.getNode(ISD::SHL, DL, MVT::i32, Elt, Shift);
+
     if (i == 0) {
       PackedValue = Elt;
     } else {
-      PackedValue = DAG.getNode(ISD::OR, DL, PackedVT, PackedValue, Elt);
+      PackedValue = DAG.getNode(ISD::OR, DL, MVT::i32, PackedValue, Elt);
     }
   }
+
+  if (PackedSize < 32) {
+    EVT PackedVT = EVT::getIntegerVT(*DAG.getContext(), PackedSize);
+    return DAG.getTruncStore(Store->getChain(), DL, PackedValue, Ptr,
+                             Store->getMemOperand()->getPointerInfo(),
+                             PackedVT,
+                             Store->isNonTemporal(), Store->isVolatile(),
+                             Store->getAlignment());
+  }
+
   return DAG.getStore(Store->getChain(), DL, PackedValue, Ptr,
-                      MachinePointerInfo(Store->getMemOperand()->getValue()),
+                      Store->getMemOperand()->getPointerInfo(),
                       Store->isVolatile(),  Store->isNonTemporal(),
                       Store->getAlignment());
 }
Index: test/CodeGen/R600/store.ll
===================================================================
--- test/CodeGen/R600/store.ll
+++ test/CodeGen/R600/store.ll
@@ -75,7 +75,11 @@
 ; EG-CHECK: MEM_RAT MSKOR
 ; EG-CHECK-NOT: MEM_RAT MSKOR
 ; SI-CHECK-LABEL: @store_v2i8
-; SI-CHECK: BUFFER_STORE_SHORT
+; SI-CHECK: S_LSHL_B32 [[LSHL:s[0-9]+]], s{{[0-9]+}}, 8
+; SI-CHECK: S_AND_B32 [[AND:s[0-9]+]], s{{[0-9]+}}, 255
+; SI-CHECK: S_OR_B32 [[RESULT:s[0-9]+]], [[AND]], [[LSHL]]
+; SI-CHECK: V_MOV_B32_e32 [[VRESULT:v[0-9]+]], [[RESULT]]
+; SI-CHECK: BUFFER_STORE_SHORT [[VRESULT]]
 define void @store_v2i8(<2 x i8> addrspace(1)* %out, <2 x i32> %in) {
 entry:
   %0 = trunc <2 x i32> %in to <2 x i8>