diff --git a/lib/Target/ARM/ARMISelDAGToDAG.cpp b/lib/Target/ARM/ARMISelDAGToDAG.cpp
index bdc29402b45..a9b86f02848 100644
--- a/lib/Target/ARM/ARMISelDAGToDAG.cpp
+++ b/lib/Target/ARM/ARMISelDAGToDAG.cpp
@@ -1445,18 +1445,61 @@ SDNode *ARMDAGToDAGISel::Select(SDValue Op) {
       SDValue MemAddr, MemUpdate, MemOpc;
       if (!SelectAddrMode6(Op, N->getOperand(2), MemAddr, MemUpdate, MemOpc))
         return NULL;
+      if (VT.is64BitVector()) {
+        switch (VT.getSimpleVT().SimpleTy) {
+        default: llvm_unreachable("unhandled vld4 type");
+        case MVT::v8i8:  Opc = ARM::VLD4d8; break;
+        case MVT::v4i16: Opc = ARM::VLD4d16; break;
+        case MVT::v2f32:
+        case MVT::v2i32: Opc = ARM::VLD4d32; break;
+        }
+        SDValue Chain = N->getOperand(0);
+        const SDValue Ops[] = { MemAddr, MemUpdate, MemOpc, Chain };
+        std::vector<EVT> ResTys(4, VT);
+        ResTys.push_back(MVT::Other);
+        return CurDAG->getMachineNode(Opc, dl, ResTys, Ops, 4);
+      }
+      // Quad registers are loaded with two separate instructions, where one
+      // loads the even registers and the other loads the odd registers.
+      EVT RegVT = VT;
+      unsigned Opc2 = 0;
       switch (VT.getSimpleVT().SimpleTy) {
       default: llvm_unreachable("unhandled vld4 type");
-      case MVT::v8i8:  Opc = ARM::VLD4d8; break;
-      case MVT::v4i16: Opc = ARM::VLD4d16; break;
-      case MVT::v2f32:
-      case MVT::v2i32: Opc = ARM::VLD4d32; break;
+      case MVT::v16i8:
+        Opc = ARM::VLD4q8a;  Opc2 = ARM::VLD4q8b;  RegVT = MVT::v8i8; break;
+      case MVT::v8i16:
+        Opc = ARM::VLD4q16a; Opc2 = ARM::VLD4q16b; RegVT = MVT::v4i16; break;
+      case MVT::v4f32:
+        Opc = ARM::VLD4q32a; Opc2 = ARM::VLD4q32b; RegVT = MVT::v2f32; break;
+      case MVT::v4i32:
+        Opc = ARM::VLD4q32a; Opc2 = ARM::VLD4q32b; RegVT = MVT::v2i32; break;
       }
       SDValue Chain = N->getOperand(0);
-      const SDValue Ops[] = { MemAddr, MemUpdate, MemOpc, Chain };
-      std::vector<EVT> ResTys(4, VT);
+      // Enable writeback to the address register.
+      MemOpc = CurDAG->getTargetConstant(ARM_AM::getAM6Opc(true), MVT::i32);
+
+      std::vector<EVT> ResTys(4, RegVT);
+      ResTys.push_back(MemAddr.getValueType());
       ResTys.push_back(MVT::Other);
-      return CurDAG->getMachineNode(Opc, dl, ResTys, Ops, 4);
+
+      const SDValue OpsA[] = { MemAddr, MemUpdate, MemOpc, Chain };
+      SDNode *VLdA = CurDAG->getMachineNode(Opc, dl, ResTys, OpsA, 4);
+      Chain = SDValue(VLdA, 5);
+
+      const SDValue OpsB[] = { SDValue(VLdA, 4), MemUpdate, MemOpc, Chain };
+      SDNode *VLdB = CurDAG->getMachineNode(Opc2, dl, ResTys, OpsB, 4);
+      Chain = SDValue(VLdB, 5);
+
+      SDNode *Q0 = PairDRegs(VT, SDValue(VLdA, 0), SDValue(VLdB, 0));
+      SDNode *Q1 = PairDRegs(VT, SDValue(VLdA, 1), SDValue(VLdB, 1));
+      SDNode *Q2 = PairDRegs(VT, SDValue(VLdA, 2), SDValue(VLdB, 2));
+      SDNode *Q3 = PairDRegs(VT, SDValue(VLdA, 3), SDValue(VLdB, 3));
+      ReplaceUses(SDValue(N, 0), SDValue(Q0, 0));
+      ReplaceUses(SDValue(N, 1), SDValue(Q1, 0));
+      ReplaceUses(SDValue(N, 2), SDValue(Q2, 0));
+      ReplaceUses(SDValue(N, 3), SDValue(Q3, 0));
+      ReplaceUses(SDValue(N, 4), Chain);
+      return NULL;
     }
 
     case Intrinsic::arm_neon_vld2lane: {
diff --git a/lib/Target/ARM/ARMInstrNEON.td b/lib/Target/ARM/ARMInstrNEON.td
index c7ff523b437..75ff78a63dc 100644
--- a/lib/Target/ARM/ARMInstrNEON.td
+++ b/lib/Target/ARM/ARMInstrNEON.td
@@ -227,11 +227,26 @@ class VLD4D<string OpcodeStr>
           (ins addrmode6:$addr), IIC_VLD4,
           !strconcat(OpcodeStr, "\t\\{$dst1,$dst2,$dst3,$dst4\\}, $addr"),
           "", []>;
+class VLD4WB<string OpcodeStr>
+  : NLdSt<(outs DPR:$dst1, DPR:$dst2, DPR:$dst3, DPR:$dst4, GPR:$wb),
+          (ins addrmode6:$addr), IIC_VLD4,
+          !strconcat(OpcodeStr, "\t\\{$dst1,$dst2,$dst3,$dst4\\}, $addr"),
+          "$addr.addr = $wb", []>;
 
 def  VLD4d8   : VLD4D<"vld4.8">;
 def  VLD4d16  : VLD4D<"vld4.16">;
 def  VLD4d32  : VLD4D<"vld4.32">;
 
+// vld4 to double-spaced even registers.
+def  VLD4q8a  : VLD4WB<"vld4.8">;
+def  VLD4q16a : VLD4WB<"vld4.16">;
+def  VLD4q32a : VLD4WB<"vld4.32">;
+
+// vld4 to double-spaced odd registers.
+def  VLD4q8b  : VLD4WB<"vld4.8">;
+def  VLD4q16b : VLD4WB<"vld4.16">;
+def  VLD4q32b : VLD4WB<"vld4.32">;
+
 //   VLD2LN   : Vector Load (single 2-element structure to one lane)
 class VLD2LND<string OpcodeStr>
   : NLdSt<(outs DPR:$dst1, DPR:$dst2),
diff --git a/lib/Target/ARM/NEONPreAllocPass.cpp b/lib/Target/ARM/NEONPreAllocPass.cpp
index fab62f6cfc2..1232f896c42 100644
--- a/lib/Target/ARM/NEONPreAllocPass.cpp
+++ b/lib/Target/ARM/NEONPreAllocPass.cpp
@@ -101,6 +101,24 @@ static bool isNEONMultiRegOp(int Opcode, unsigned &FirstOpnd, unsigned &NumRegs,
     NumRegs = 4;
     return true;
 
+  case ARM::VLD4q8a:
+  case ARM::VLD4q16a:
+  case ARM::VLD4q32a:
+    FirstOpnd = 0;
+    NumRegs = 4;
+    Offset = 0;
+    Stride = 2;
+    return true;
+
+  case ARM::VLD4q8b:
+  case ARM::VLD4q16b:
+  case ARM::VLD4q32b:
+    FirstOpnd = 0;
+    NumRegs = 4;
+    Offset = 1;
+    Stride = 2;
+    return true;
+
   case ARM::VST2d8:
   case ARM::VST2d16:
   case ARM::VST2d32:
diff --git a/test/CodeGen/ARM/vld4.ll b/test/CodeGen/ARM/vld4.ll
index 6e6bb75c168..08f67a793a5 100644
--- a/test/CodeGen/ARM/vld4.ll
+++ b/test/CodeGen/ARM/vld4.ll
@@ -5,6 +5,11 @@
 %struct.__neon_int32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
 %struct.__neon_float32x2x4_t = type { <2 x float>, <2 x float>, <2 x float>, <2 x float> }
 
+%struct.__neon_int8x16x4_t = type { <16 x i8>,  <16 x i8>,  <16 x i8>, <16 x i8> }
+%struct.__neon_int16x8x4_t = type { <8 x i16>, <8 x i16>, <8 x i16>, <8 x i16> }
+%struct.__neon_int32x4x4_t = type { <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32> }
+%struct.__neon_float32x4x4_t = type { <4 x float>, <4 x float>, <4 x float>, <4 x float> }
+
 define <8 x i8> @vld4i8(i8* %A) nounwind {
 ;CHECK: vld4i8:
 ;CHECK: vld4.8
@@ -45,7 +50,56 @@ define <2 x float> @vld4f(float* %A) nounwind {
 	ret <2 x float> %tmp4
 }
 
+define <16 x i8> @vld4Qi8(i8* %A) nounwind {
+;CHECK: vld4Qi8:
+;CHECK: vld4.8
+;CHECK: vld4.8
+	%tmp1 = call %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8* %A)
+        %tmp2 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 0
+        %tmp3 = extractvalue %struct.__neon_int8x16x4_t %tmp1, 2
+        %tmp4 = add <16 x i8> %tmp2, %tmp3
+	ret <16 x i8> %tmp4
+}
+
+define <8 x i16> @vld4Qi16(i16* %A) nounwind {
+;CHECK: vld4Qi16:
+;CHECK: vld4.16
+;CHECK: vld4.16
+	%tmp1 = call %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i16* %A)
+        %tmp2 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 0
+        %tmp3 = extractvalue %struct.__neon_int16x8x4_t %tmp1, 2
+        %tmp4 = add <8 x i16> %tmp2, %tmp3
+	ret <8 x i16> %tmp4
+}
+
+define <4 x i32> @vld4Qi32(i32* %A) nounwind {
+;CHECK: vld4Qi32:
+;CHECK: vld4.32
+;CHECK: vld4.32
+	%tmp1 = call %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i32* %A)
+        %tmp2 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 0
+        %tmp3 = extractvalue %struct.__neon_int32x4x4_t %tmp1, 2
+        %tmp4 = add <4 x i32> %tmp2, %tmp3
+	ret <4 x i32> %tmp4
+}
+
+define <4 x float> @vld4Qf(float* %A) nounwind {
+;CHECK: vld4Qf:
+;CHECK: vld4.32
+;CHECK: vld4.32
+	%tmp1 = call %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(float* %A)
+        %tmp2 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 0
+        %tmp3 = extractvalue %struct.__neon_float32x4x4_t %tmp1, 2
+        %tmp4 = add <4 x float> %tmp2, %tmp3
+	ret <4 x float> %tmp4
+}
+
 declare %struct.__neon_int8x8x4_t @llvm.arm.neon.vld4.v8i8(i8*) nounwind readonly
 declare %struct.__neon_int16x4x4_t @llvm.arm.neon.vld4.v4i16(i8*) nounwind readonly
 declare %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32(i8*) nounwind readonly
 declare %struct.__neon_float32x2x4_t @llvm.arm.neon.vld4.v2f32(i8*) nounwind readonly
+
+declare %struct.__neon_int8x16x4_t @llvm.arm.neon.vld4.v16i8(i8*) nounwind readonly
+declare %struct.__neon_int16x8x4_t @llvm.arm.neon.vld4.v8i16(i8*) nounwind readonly
+declare %struct.__neon_int32x4x4_t @llvm.arm.neon.vld4.v4i32(i8*) nounwind readonly
+declare %struct.__neon_float32x4x4_t @llvm.arm.neon.vld4.v4f32(i8*) nounwind readonly