[ARM] MVE VMULL patterns

[dmgreen]

Summary

This adds MVE vmull patterns, which are conceptually the same as mul(vmovl, vmovl), and so the tablegen patterns follow the same structure.

For i8 and i16 this is simple enough, but in the i32 version the multiply (in 64bits) is illegal, meaning we need to catch the pattern earlier in a dag fold. Because bitcasts are involved in the zext versions and the patterns are a little different in little and big endian. I have only added little endian support in this patch.

[samparker]

llvm/lib/Target/ARM/ARMISelLowering.cpp

12166

I'm a bit confused here... this looks like the AND mask is taking the 'top' parts of the elements, and if so, why don't we have to handle a 'bottom' element mask? Is this to do with the revs I see in the isel patterns?

[dmgreen]

Added some tests for both top and bottom vmull's

llvm/lib/Target/ARM/ARMISelLowering.cpp

12166

I think this is bottom half? The 0 element is bottom half of the i64, 1 element would be top half of it.

And yes, the other pattern vmullt is the same pattern with some vrev's that we match in tablegen.

You reminded me that I did want to add some tests for both top and bottom at the same time.

[samparker]

LGTM

llvm/lib/Target/ARM/ARMISelLowering.cpp

12166

Hmm, not sure how I misread that!

llvm/test/CodeGen/Thumb2/mve-vmull.ll

52–53

Do you know why we sometimes can't write to q0 too?

[dmgreen]

Thanks.

llvm/test/CodeGen/Thumb2/mve-vmull.ll

52–53

The i32 version is earlyclobber because it needs to work cross-lane. I think it's UNPREDICTABLE if Qd == Qn or Qm

llvm/lib/Target/ARM/ARMISelLowering.cpp

static SDValue PerformVMULCombine(SDNode *N,
   SDLoc DL(N);
   SDValue N00 = N0->getOperand(0);
   SDValue N01 = N0->getOperand(1);
   return DAG.getNode(Opcode, DL, VT,
                      DAG.getNode(ISD::MUL, DL, VT, N00, N1),
                      DAG.getNode(ISD::MUL, DL, VT, N01, N1));
 }
 
+static SDValue PerformMVEVMULLCombine(SDNode *N, SelectionDAG &DAG,
+                                      const ARMSubtarget *Subtarget) {
+  EVT VT = N->getValueType(0);
+  if (VT != MVT::v2i64)
+    return SDValue();
+
+  SDValue N0 = N->getOperand(0);
+  SDValue N1 = N->getOperand(1);
+
+  auto IsSignExt = [&](SDValue Op) {
+    if (Op->getOpcode() != ISD::SIGN_EXTEND_INREG)
+      return SDValue();
+    EVT VT = cast<VTSDNode>(Op->getOperand(1))->getVT();
+    if (VT.getScalarSizeInBits() == 32)
+      return Op->getOperand(0);
+    return SDValue();
+  };
+  auto IsZeroExt = [&](SDValue Op) {
+    // Zero extends are a little more awkward. At the point we are matching
+    // this, we are looking for an AND with a (-1, 0, -1, 0) buildvector mask.
+    // That might be before of after a bitcast depending on how the and is
+    // placed. Because this has to look through bitcasts, it is currently only
+    // supported on LE.
+    if (!Subtarget->isLittle())
+      return SDValue();
+
+    SDValue And = Op;
+    if (And->getOpcode() == ISD::BITCAST)
+      And = And->getOperand(0);
+    if (And->getOpcode() != ISD::AND)
+      return SDValue();
+    SDValue Mask = And->getOperand(1);
+    if (Mask->getOpcode() == ISD::BITCAST)
+      Mask = Mask->getOperand(0);
+
+    if (Mask->getOpcode() != ISD::BUILD_VECTOR ||
+        Mask.getValueType() != MVT::v4i32)
+      return SDValue();
+    if (isAllOnesConstant(Mask->getOperand(0)) &&
+        isNullConstant(Mask->getOperand(1)) &&
+        isAllOnesConstant(Mask->getOperand(2)) &&
+        isNullConstant(Mask->getOperand(3)))
+      return And->getOperand(0);
+    return SDValue();
+  };
+
+  SDLoc dl(N);
+  if (SDValue Op0 = IsSignExt(N0)) {
+    if (SDValue Op1 = IsSignExt(N1)) {
+      SDValue New0a = DAG.getNode(ARMISD::VECTOR_REG_CAST, dl, MVT::v4i32, Op0);
+      SDValue New1a = DAG.getNode(ARMISD::VECTOR_REG_CAST, dl, MVT::v4i32, Op1);
+      return DAG.getNode(ARMISD::VMULLs, dl, VT, New0a, New1a);
+    }
+  }
+  if (SDValue Op0 = IsZeroExt(N0)) {
+    if (SDValue Op1 = IsZeroExt(N1)) {
+      SDValue New0a = DAG.getNode(ARMISD::VECTOR_REG_CAST, dl, MVT::v4i32, Op0);
+      SDValue New1a = DAG.getNode(ARMISD::VECTOR_REG_CAST, dl, MVT::v4i32, Op1);
+      return DAG.getNode(ARMISD::VMULLu, dl, VT, New0a, New1a);
+    }
+  }
+
+  return SDValue();
+}
+
 static SDValue PerformMULCombine(SDNode *N,
                                  TargetLowering::DAGCombinerInfo &DCI,
                                  const ARMSubtarget *Subtarget) {
   SelectionDAG &DAG = DCI.DAG;
 
+  EVT VT = N->getValueType(0);
+  if (Subtarget->hasMVEIntegerOps() && VT == MVT::v2i64)
+    return PerformMVEVMULLCombine(N, DAG, Subtarget);
+
   if (Subtarget->isThumb1Only())
     return SDValue();
 
   if (DCI.isBeforeLegalize() || DCI.isCalledByLegalizer())
     return SDValue();
 
-  EVT VT = N->getValueType(0);
   if (VT.is64BitVector() || VT.is128BitVector())
     return PerformVMULCombine(N, DCI, Subtarget);
   if (VT != MVT::i32)
     return SDValue();
 
   ConstantSDNode *C = dyn_cast<ConstantSDNode>(N->getOperand(1));
   if (!C)
     return SDValue();

llvm/lib/Target/ARM/ARMInstrInfo.td

 def SDTARMVSH : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisSameAs<0, 1>,
                                      SDTCisSameAs<0, 2>,]>;
 def ARMvshlImm   : SDNode<"ARMISD::VSHLIMM", SDTARMVSHIMM>;
 def ARMvshrsImm  : SDNode<"ARMISD::VSHRsIMM", SDTARMVSHIMM>;
 def ARMvshruImm  : SDNode<"ARMISD::VSHRuIMM", SDTARMVSHIMM>;
 def ARMvshls     : SDNode<"ARMISD::VSHLs", SDTARMVSH>;
 def ARMvshlu     : SDNode<"ARMISD::VSHLu", SDTARMVSH>;
 
+def SDTARMVMULL   : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
+                                         SDTCisSameAs<1, 2>]>;
+def ARMvmulls    : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
+def ARMvmullu    : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
+
 def SDTARMVCMP    : SDTypeProfile<1, 3, [SDTCisInt<0>, SDTCisSameAs<1, 2>,
                                          SDTCisInt<3>]>;
 def SDTARMVCMPZ   : SDTypeProfile<1, 2, [SDTCisInt<2>]>;
 
 def ARMvcmp      : SDNode<"ARMISD::VCMP", SDTARMVCMP>;
 def ARMvcmpz     : SDNode<"ARMISD::VCMPZ", SDTARMVCMPZ>;
 
 def ARMWLS      : SDNode<"ARMISD::WLS", SDT_ARMLoLoop, [SDNPHasChain]>;

llvm/lib/Target/ARM/ARMInstrMVE.td

 let Predicates = [HasMVEInt] in {
   def : Pat<(sext_inreg (v4i32 MQPR:$src), v4i16),
             (MVE_VMOVLs16bh MQPR:$src)>;
   def : Pat<(sext_inreg (v8i16 MQPR:$src), v8i8),
             (MVE_VMOVLs8bh MQPR:$src)>;
   def : Pat<(sext_inreg (v4i32 MQPR:$src), v4i8),
             (MVE_VMOVLs16bh (MVE_VMOVLs8bh MQPR:$src))>;
 
-  def : Pat<(sext_inreg (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src)))),
-                        v8i8), (MVE_VMOVLs8th MQPR:$src)>;
-  def : Pat<(sext_inreg (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src)))),
-                        v4i16), (MVE_VMOVLs16th MQPR:$src)>;
-  def : Pat<(ARMvbicImm (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src)))),
-                        (i32 0xAFF)), (MVE_VMOVLu8th MQPR:$src)>;
-  def : Pat<(and (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src)))),
-                 (v4i32 (ARMvmovImm (i32 0xCFF)))),
-                 (MVE_VMOVLu16th MQPR:$src)>;
+  def : Pat<(sext_inreg (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src)))), v8i8),
+            (MVE_VMOVLs8th MQPR:$src)>;
+  def : Pat<(sext_inreg (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src)))), v4i16),
+            (MVE_VMOVLs16th MQPR:$src)>;
 
-  // zext_inreg 16 -> 32
-  def : Pat<(and (v4i32 MQPR:$src), (v4i32 (ARMvmovImm (i32 0xCFF)))),
-            (MVE_VMOVLu16bh MQPR:$src)>;
   // zext_inreg 8 -> 16
   def : Pat<(ARMvbicImm (v8i16 MQPR:$src), (i32 0xAFF)),
             (MVE_VMOVLu8bh MQPR:$src)>;
+  // zext_inreg 16 -> 32
+  def : Pat<(and (v4i32 MQPR:$src), (v4i32 (ARMvmovImm (i32 0xCFF)))),
+            (MVE_VMOVLu16bh MQPR:$src)>;
+  // Same zext_inreg with vrevs, picking the top half
+  def : Pat<(ARMvbicImm (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src)))), (i32 0xAFF)),
+            (MVE_VMOVLu8th MQPR:$src)>;
+  def : Pat<(and (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src)))),
+                 (v4i32 (ARMvmovImm (i32 0xCFF)))),
+            (MVE_VMOVLu16th MQPR:$src)>;
 }
 
 
 class MVE_VSHLL_imm<string iname, string suffix, bit U, bit th,
                     Operand immtype, list<dag> pattern=[]>
   : MVE_shift_imm<(outs MQPR:$Qd), (ins MQPR:$Qm, immtype:$imm),
                   iname, suffix, "$Qd, $Qm, $imm", vpred_r, "", pattern> {
   let Inst{28} = U;
defm MVE_VMULLBp8  : MVE_VMULL_m<MVE_v16p8, int_arm_mve_vmull_poly,
                                  int_arm_mve_mull_poly_predicated, 0b0>;
 defm MVE_VMULLTp8  : MVE_VMULL_m<MVE_v16p8, int_arm_mve_vmull_poly,
                                  int_arm_mve_mull_poly_predicated, 0b1>;
 defm MVE_VMULLBp16 : MVE_VMULL_m<MVE_v8p16, int_arm_mve_vmull_poly,
                                  int_arm_mve_mull_poly_predicated, 0b0>;
 defm MVE_VMULLTp16 : MVE_VMULL_m<MVE_v8p16, int_arm_mve_vmull_poly,
                                  int_arm_mve_mull_poly_predicated, 0b1>;
 
+let Predicates = [HasMVEInt] in {
+  def : Pat<(v2i64 (ARMvmulls (v4i32 MQPR:$src1), (v4i32 MQPR:$src2))),
+            (MVE_VMULLBs32 MQPR:$src1, MQPR:$src2)>;
+  def : Pat<(v2i64 (ARMvmulls (v4i32 (ARMvrev64 (v4i32 MQPR:$src1))),
+                              (v4i32 (ARMvrev64 (v4i32 MQPR:$src2))))),
+            (MVE_VMULLTs32 MQPR:$src1, MQPR:$src2)>;
+
+  def : Pat<(mul (sext_inreg (v4i32 MQPR:$src1), v4i16),
+                 (sext_inreg (v4i32 MQPR:$src2), v4i16)),
+            (MVE_VMULLBs16 MQPR:$src1, MQPR:$src2)>;
+  def : Pat<(mul (sext_inreg (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src1)))), v4i16),
+                 (sext_inreg (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src2)))), v4i16)),
+            (MVE_VMULLTs16 MQPR:$src1, MQPR:$src2)>;
+
+  def : Pat<(mul (sext_inreg (v8i16 MQPR:$src1), v8i8),
+                 (sext_inreg (v8i16 MQPR:$src2), v8i8)),
+            (MVE_VMULLBs8 MQPR:$src1, MQPR:$src2)>;
+  def : Pat<(mul (sext_inreg (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src1)))), v8i8),
+                 (sext_inreg (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src2)))), v8i8)),
+            (MVE_VMULLTs8 MQPR:$src1, MQPR:$src2)>;
+
+  def : Pat<(v2i64 (ARMvmullu (v4i32 MQPR:$src1), (v4i32 MQPR:$src2))),
+            (MVE_VMULLBu32 MQPR:$src1, MQPR:$src2)>;
+  def : Pat<(v2i64 (ARMvmullu (v4i32 (ARMvrev64 (v4i32 MQPR:$src1))),
+                              (v4i32 (ARMvrev64 (v4i32 MQPR:$src2))))),
+            (MVE_VMULLTu32 MQPR:$src1, MQPR:$src2)>;
+
+  def : Pat<(mul (and (v4i32 MQPR:$src1), (v4i32 (ARMvmovImm (i32 0xCFF)))),
+                 (and (v4i32 MQPR:$src2), (v4i32 (ARMvmovImm (i32 0xCFF))))),
+            (MVE_VMULLBu16 MQPR:$src1, MQPR:$src2)>;
+  def : Pat<(mul (and (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src1)))),
+                      (v4i32 (ARMvmovImm (i32 0xCFF)))),
+                 (and (v4i32 (ARMVectorRegCast (ARMvrev32 (v8i16 MQPR:$src2)))),
+                      (v4i32 (ARMvmovImm (i32 0xCFF))))),
+            (MVE_VMULLTu16 MQPR:$src1, MQPR:$src2)>;
+
+  def : Pat<(mul (ARMvbicImm (v8i16 MQPR:$src1), (i32 0xAFF)),
+                 (ARMvbicImm (v8i16 MQPR:$src2), (i32 0xAFF))),
+            (MVE_VMULLBu8 MQPR:$src1, MQPR:$src2)>;
+  def : Pat<(mul (ARMvbicImm (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src1)))), (i32 0xAFF)),
+                 (ARMvbicImm (v8i16 (ARMVectorRegCast (ARMvrev16 (v16i8 MQPR:$src2)))), (i32 0xAFF))),
+            (MVE_VMULLTu8 MQPR:$src1, MQPR:$src2)>;
+}
+
 class MVE_VxMULH<string iname, string suffix, bit U, bits<2> size, bit round,
                  list<dag> pattern=[]>
   : MVE_qDest_qSrc<iname, suffix, (outs MQPR:$Qd),
                    (ins MQPR:$Qn, MQPR:$Qm), "$Qd, $Qn, $Qm",
                    vpred_r, "", pattern> {
   bits<4> Qn;
 
   let Inst{28} = U;

llvm/lib/Target/ARM/ARMInstrNEON.td

 
 def SDTARMVSHUF2  : SDTypeProfile<2, 2, [SDTCisVec<0>, SDTCisSameAs<0, 1>,
                                          SDTCisSameAs<0, 2>,
                                          SDTCisSameAs<0, 3>]>;
 def NEONzip       : SDNode<"ARMISD::VZIP", SDTARMVSHUF2>;
 def NEONuzp       : SDNode<"ARMISD::VUZP", SDTARMVSHUF2>;
 def NEONtrn       : SDNode<"ARMISD::VTRN", SDTARMVSHUF2>;
 
-def SDTARMVMULL   : SDTypeProfile<1, 2, [SDTCisInt<0>, SDTCisInt<1>,
-                                         SDTCisSameAs<1, 2>]>;
-def NEONvmulls    : SDNode<"ARMISD::VMULLs", SDTARMVMULL>;
-def NEONvmullu    : SDNode<"ARMISD::VMULLu", SDTARMVMULL>;
-
 def SDTARMVTBL1   : SDTypeProfile<1, 2, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
                                          SDTCisVT<2, v8i8>]>;
 def SDTARMVTBL2   : SDTypeProfile<1, 3, [SDTCisVT<0, v8i8>, SDTCisVT<1, v8i8>,
                                          SDTCisVT<2, v8i8>, SDTCisVT<3, v8i8>]>;
 def NEONvtbl1     : SDNode<"ARMISD::VTBL1", SDTARMVTBL1>;
 def NEONvtbl2     : SDNode<"ARMISD::VTBL2", SDTARMVTBL2>;
 
 
def : Pat<(v4i32 (int_arm_neon_vqrdmulh (v4i32 QPR:$src1),
                                           (DSubReg_i32_reg imm:$lane))),
                                   (SubReg_i32_lane imm:$lane)))>;
 }
 
 //   VMULL    : Vector Multiply Long (integer and polynomial) (Q = D * D)
 let PostEncoderMethod = "NEONThumb2DataIPostEncoder",
     DecoderNamespace = "NEONData" in {
   defm VMULLs   : N3VL_QHS<0,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
-                           "vmull", "s", NEONvmulls, 1>;
+                           "vmull", "s", ARMvmulls, 1>;
   defm VMULLu   : N3VL_QHS<1,1,0b1100,0, IIC_VMULi16D, IIC_VMULi32D,
-                           "vmull", "u", NEONvmullu, 1>;
+                           "vmull", "u", ARMvmullu, 1>;
   def  VMULLp8   :  N3VLInt<0, 1, 0b00, 0b1110, 0, IIC_VMULi16D, "vmull", "p8",
                             v8i16, v8i8, int_arm_neon_vmullp, 1>;
   def  VMULLp64  : N3VLIntnp<0b00101, 0b10, 0b1110, 0, 0, NoItinerary,
                           "vmull", "p64", v2i64, v1i64, int_arm_neon_vmullp, 1>,
                     Requires<[HasV8, HasCrypto]>;
 }
-defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", NEONvmulls>;
-defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", NEONvmullu>;
+defm VMULLsls : N3VLSL_HS<0, 0b1010, IIC_VMULi16D, "vmull", "s", ARMvmulls>;
+defm VMULLslu : N3VLSL_HS<1, 0b1010, IIC_VMULi16D, "vmull", "u", ARMvmullu>;
 
 //   VQDMULL  : Vector Saturating Doubling Multiply Long (Q = D * D)
 defm VQDMULL  : N3VLInt_HS<0,1,0b1101,0, IIC_VMULi16D, IIC_VMULi32D,
                            "vqdmull", "s", int_arm_neon_vqdmull, 1>;
 defm VQDMULLsl: N3VLIntSL_HS<0, 0b1011, IIC_VMULi16D,
                              "vqdmull", "s", int_arm_neon_vqdmull>;
 
 // Vector Multiply-Accumulate and Multiply-Subtract Operations.
def : Pat<(v4f32 (fadd_mlx (v4f32 QPR:$src1),
                            (v4f32 QPR:$src2),
                            (v2f32 (EXTRACT_SUBREG QPR:$src3,
                                    (DSubReg_i32_reg imm:$lane))),
                            (SubReg_i32_lane imm:$lane)))>,
           Requires<[HasNEON, UseFPVMLx]>;
 
 //   VMLAL    : Vector Multiply Accumulate Long (Q += D * D)
 defm VMLALs   : N3VLMulOp_QHS<0,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
-                              "vmlal", "s", NEONvmulls, add>;
+                              "vmlal", "s", ARMvmulls, add>;
 defm VMLALu   : N3VLMulOp_QHS<1,1,0b1000,0, IIC_VMACi16D, IIC_VMACi32D,
-                              "vmlal", "u", NEONvmullu, add>;
+                              "vmlal", "u", ARMvmullu, add>;
 
-defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", NEONvmulls, add>;
-defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", NEONvmullu, add>;
+defm VMLALsls : N3VLMulOpSL_HS<0, 0b0010, "vmlal", "s", ARMvmulls, add>;
+defm VMLALslu : N3VLMulOpSL_HS<1, 0b0010, "vmlal", "u", ARMvmullu, add>;
 
 let Predicates = [HasNEON, HasV8_1a] in {
   // v8.1a Neon Rounding Double Multiply-Op vector operations,
   // VQRDMLAH : Vector Saturating Rounding Doubling Multiply Accumulate Long
   //            (Q += D * D)
   defm VQRDMLAH : N3VInt3_HS<1, 0, 0b1011, 1, IIC_VMACi16D, IIC_VMACi32D,
                              IIC_VMACi16Q, IIC_VMACi32Q, "vqrdmlah", "s",
                              null_frag>;
def : Pat<(v4f32 (fsub_mlx (v4f32 QPR:$src1),
           (v4f32 (VMLSslfq (v4f32 QPR:$src1), (v4f32 QPR:$src2),
                            (v2f32 (EXTRACT_SUBREG QPR:$src3,
                                    (DSubReg_i32_reg imm:$lane))),
                            (SubReg_i32_lane imm:$lane)))>,
           Requires<[HasNEON, UseFPVMLx]>;
 
 //   VMLSL    : Vector Multiply Subtract Long (Q -= D * D)
 defm VMLSLs   : N3VLMulOp_QHS<0,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
-                              "vmlsl", "s", NEONvmulls, sub>;
+                              "vmlsl", "s", ARMvmulls, sub>;
 defm VMLSLu   : N3VLMulOp_QHS<1,1,0b1010,0, IIC_VMACi16D, IIC_VMACi32D,
-                              "vmlsl", "u", NEONvmullu, sub>;
+                              "vmlsl", "u", ARMvmullu, sub>;
 
-defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", NEONvmulls, sub>;
-defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", NEONvmullu, sub>;
+defm VMLSLsls : N3VLMulOpSL_HS<0, 0b0110, "vmlsl", "s", ARMvmulls, sub>;
+defm VMLSLslu : N3VLMulOpSL_HS<1, 0b0110, "vmlsl", "u", ARMvmullu, sub>;
 
 //   VQDMLSL  : Vector Saturating Doubling Multiply Subtract Long (Q -= D * D)
 defm VQDMLSL  : N3VLInt3_HS<0, 1, 0b1011, 0, IIC_VMACi16D, IIC_VMACi32D,
                             "vqdmlsl", "s", null_frag>;
 defm VQDMLSLsl: N3VLInt3SL_HS<0, 0b0111, "vqdmlsl", "s", null_frag>;
 
 let Predicates = [HasNEON] in {
 def : Pat<(v4i32 (ssubsat (v4i32 QPR:$src1),

llvm/test/CodeGen/Thumb2/mve-vecreduce-mla.ll

entry:
   %m = mul <4 x i64> %xx, %yy
   %z = call i64 @llvm.experimental.vector.reduce.add.v4i64(<4 x i64> %m)
   ret i64 %z
 }
 
 define arm_aapcs_vfpcc i64 @add_v2i32_v2i64_zext(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: add_v2i32_v2i64_zext:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov r2, s6
-; CHECK-NEXT:    vmov r3, s2
-; CHECK-NEXT:    umull r0, r1, r1, r0
-; CHECK-NEXT:    umlal r0, r1, r3, r2
+; CHECK-NEXT:    vmullb.u32 q2, q0, q1
+; CHECK-NEXT:    vmov r0, s10
+; CHECK-NEXT:    vmov r3, s8
+; CHECK-NEXT:    vmov r1, s11
+; CHECK-NEXT:    vmov r2, s9
+; CHECK-NEXT:    adds r0, r0, r3
+; CHECK-NEXT:    adcs r1, r2
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = zext <2 x i32> %x to <2 x i64>
   %yy = zext <2 x i32> %y to <2 x i64>
   %m = mul <2 x i64> %xx, %yy
   %z = call i64 @llvm.experimental.vector.reduce.add.v2i64(<2 x i64> %m)
   ret i64 %z
 }
 
 define arm_aapcs_vfpcc i64 @add_v2i32_v2i64_sext(<2 x i32> %x, <2 x i32> %y) {
 ; CHECK-LABEL: add_v2i32_v2i64_sext:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    vmov r2, s6
-; CHECK-NEXT:    vmov r3, s2
-; CHECK-NEXT:    smull r0, r1, r1, r0
-; CHECK-NEXT:    smlal r0, r1, r3, r2
+; CHECK-NEXT:    vmullb.s32 q2, q0, q1
+; CHECK-NEXT:    vmov r0, s10
+; CHECK-NEXT:    vmov r3, s8
+; CHECK-NEXT:    vmov r1, s11
+; CHECK-NEXT:    vmov r2, s9
+; CHECK-NEXT:    adds r0, r0, r3
+; CHECK-NEXT:    adcs r1, r2
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = sext <2 x i32> %x to <2 x i64>
   %yy = sext <2 x i32> %y to <2 x i64>
   %m = mul <2 x i64> %xx, %yy
   %z = call i64 @llvm.experimental.vector.reduce.add.v2i64(<2 x i64> %m)
   ret i64 %z
 }

 ; CHECK-NEXT:    vmov.u8 r0, q0[12]
 ; CHECK-NEXT:    vmov.16 q3[4], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[13]
 ; CHECK-NEXT:    vmov.16 q3[5], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[14]
 ; CHECK-NEXT:    vmov.16 q3[6], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[15]
 ; CHECK-NEXT:    vmov.16 q3[7], r0
-; CHECK-NEXT:    vmovlb.u8 q2, q2
-; CHECK-NEXT:    vmovlb.u8 q3, q3
 ; CHECK-NEXT:    vmov.u8 r0, q1[0]
-; CHECK-NEXT:    vmul.i16 q2, q3, q2
+; CHECK-NEXT:    vmullb.u8 q2, q3, q2
 ; CHECK-NEXT:    vmov.16 q3[0], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[1]
 ; CHECK-NEXT:    vmov.16 q3[1], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[2]
 ; CHECK-NEXT:    vmov.16 q3[2], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[3]
 ; CHECK-NEXT:    vmov.16 q3[3], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[4]
 ; CHECK-NEXT:    vmov.16 q3[4], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[5]
 ; CHECK-NEXT:    vmov.16 q3[5], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[6]
 ; CHECK-NEXT:    vmov.16 q3[6], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[7]
 ; CHECK-NEXT:    vmov.16 q3[7], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[0]
-; CHECK-NEXT:    vmovlb.u8 q1, q3
-; CHECK-NEXT:    vmov.16 q3[0], r0
+; CHECK-NEXT:    vmov.16 q1[0], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[1]
-; CHECK-NEXT:    vmov.16 q3[1], r0
+; CHECK-NEXT:    vmov.16 q1[1], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[2]
-; CHECK-NEXT:    vmov.16 q3[2], r0
+; CHECK-NEXT:    vmov.16 q1[2], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[3]
-; CHECK-NEXT:    vmov.16 q3[3], r0
+; CHECK-NEXT:    vmov.16 q1[3], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[4]
-; CHECK-NEXT:    vmov.16 q3[4], r0
+; CHECK-NEXT:    vmov.16 q1[4], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[5]
-; CHECK-NEXT:    vmov.16 q3[5], r0
+; CHECK-NEXT:    vmov.16 q1[5], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[6]
-; CHECK-NEXT:    vmov.16 q3[6], r0
+; CHECK-NEXT:    vmov.16 q1[6], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[7]
-; CHECK-NEXT:    vmov.16 q3[7], r0
-; CHECK-NEXT:    vmovlb.u8 q0, q3
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmov.16 q1[7], r0
+; CHECK-NEXT:    vmullb.u8 q0, q1, q3
 ; CHECK-NEXT:    vadd.i16 q0, q0, q2
 ; CHECK-NEXT:    vaddv.u16 r0, q0
 ; CHECK-NEXT:    uxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = zext <16 x i8> %x to <16 x i16>
   %yy = zext <16 x i8> %y to <16 x i16>
   %m = mul <16 x i16> %xx, %yy

 ; CHECK-NEXT:    vmov.u8 r0, q0[12]
 ; CHECK-NEXT:    vmov.16 q3[4], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[13]
 ; CHECK-NEXT:    vmov.16 q3[5], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[14]
 ; CHECK-NEXT:    vmov.16 q3[6], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[15]
 ; CHECK-NEXT:    vmov.16 q3[7], r0
-; CHECK-NEXT:    vmovlb.s8 q2, q2
-; CHECK-NEXT:    vmovlb.s8 q3, q3
 ; CHECK-NEXT:    vmov.u8 r0, q1[0]
-; CHECK-NEXT:    vmul.i16 q2, q3, q2
+; CHECK-NEXT:    vmullb.s8 q2, q3, q2
 ; CHECK-NEXT:    vmov.16 q3[0], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[1]
 ; CHECK-NEXT:    vmov.16 q3[1], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[2]
 ; CHECK-NEXT:    vmov.16 q3[2], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[3]
 ; CHECK-NEXT:    vmov.16 q3[3], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[4]
 ; CHECK-NEXT:    vmov.16 q3[4], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[5]
 ; CHECK-NEXT:    vmov.16 q3[5], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[6]
 ; CHECK-NEXT:    vmov.16 q3[6], r0
 ; CHECK-NEXT:    vmov.u8 r0, q1[7]
 ; CHECK-NEXT:    vmov.16 q3[7], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[0]
-; CHECK-NEXT:    vmovlb.s8 q1, q3
-; CHECK-NEXT:    vmov.16 q3[0], r0
+; CHECK-NEXT:    vmov.16 q1[0], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[1]
-; CHECK-NEXT:    vmov.16 q3[1], r0
+; CHECK-NEXT:    vmov.16 q1[1], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[2]
-; CHECK-NEXT:    vmov.16 q3[2], r0
+; CHECK-NEXT:    vmov.16 q1[2], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[3]
-; CHECK-NEXT:    vmov.16 q3[3], r0
+; CHECK-NEXT:    vmov.16 q1[3], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[4]
-; CHECK-NEXT:    vmov.16 q3[4], r0
+; CHECK-NEXT:    vmov.16 q1[4], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[5]
-; CHECK-NEXT:    vmov.16 q3[5], r0
+; CHECK-NEXT:    vmov.16 q1[5], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[6]
-; CHECK-NEXT:    vmov.16 q3[6], r0
+; CHECK-NEXT:    vmov.16 q1[6], r0
 ; CHECK-NEXT:    vmov.u8 r0, q0[7]
-; CHECK-NEXT:    vmov.16 q3[7], r0
-; CHECK-NEXT:    vmovlb.s8 q0, q3
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmov.16 q1[7], r0
+; CHECK-NEXT:    vmullb.s8 q0, q1, q3
 ; CHECK-NEXT:    vadd.i16 q0, q0, q2
 ; CHECK-NEXT:    vaddv.u16 r0, q0
 ; CHECK-NEXT:    sxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = sext <16 x i8> %x to <16 x i16>
   %yy = sext <16 x i8> %y to <16 x i16>
   %m = mul <16 x i16> %xx, %yy
   %z = call i16 @llvm.experimental.vector.reduce.add.v16i16(<16 x i16> %m)
   ret i16 %z
 }
 
 define arm_aapcs_vfpcc zeroext i16 @add_v8i8_v8i16_zext(<8 x i8> %x, <8 x i8> %y) {
 ; CHECK-LABEL: add_v8i8_v8i16_zext:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.u8 q1, q1
-; CHECK-NEXT:    vmovlb.u8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullb.u8 q0, q0, q1
 ; CHECK-NEXT:    vaddv.u16 r0, q0
 ; CHECK-NEXT:    uxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = zext <8 x i8> %x to <8 x i16>
   %yy = zext <8 x i8> %y to <8 x i16>
   %m = mul <8 x i16> %xx, %yy
   %z = call i16 @llvm.experimental.vector.reduce.add.v8i16(<8 x i16> %m)
   ret i16 %z
 }
 
 define arm_aapcs_vfpcc signext i16 @add_v8i8_v8i16_sext(<8 x i8> %x, <8 x i8> %y) {
 ; CHECK-LABEL: add_v8i8_v8i16_sext:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.s8 q1, q1
-; CHECK-NEXT:    vmovlb.s8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullb.s8 q0, q0, q1
 ; CHECK-NEXT:    vaddv.u16 r0, q0
 ; CHECK-NEXT:    sxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = sext <8 x i8> %x to <8 x i16>
   %yy = sext <8 x i8> %y to <8 x i16>
   %m = mul <8 x i16> %xx, %yy
   %z = call i16 @llvm.experimental.vector.reduce.add.v8i16(<8 x i16> %m)
entry:
   ret i64 %r
 }
 
 define arm_aapcs_vfpcc i64 @add_v2i32_v2i64_acc_zext(<2 x i32> %x, <2 x i32> %y, i64 %a) {
 ; CHECK-LABEL: add_v2i32_v2i64_acc_zext:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    .save {r7, lr}
 ; CHECK-NEXT:    push {r7, lr}
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vmov r3, s0
-; CHECK-NEXT:    vmov r12, s6
-; CHECK-NEXT:    umull r2, lr, r3, r2
-; CHECK-NEXT:    vmov r3, s2
-; CHECK-NEXT:    umlal r2, lr, r3, r12
+; CHECK-NEXT:    vmullb.u32 q2, q0, q1
+; CHECK-NEXT:    vmov r2, s10
+; CHECK-NEXT:    vmov r3, s8
+; CHECK-NEXT:    vmov r12, s11
+; CHECK-NEXT:    vmov lr, s9
+; CHECK-NEXT:    adds r2, r2, r3
+; CHECK-NEXT:    adc.w r3, lr, r12
 ; CHECK-NEXT:    adds r0, r0, r2
-; CHECK-NEXT:    adc.w r1, r1, lr
+; CHECK-NEXT:    adcs r1, r3
 ; CHECK-NEXT:    pop {r7, pc}
 entry:
   %xx = zext <2 x i32> %x to <2 x i64>
   %yy = zext <2 x i32> %y to <2 x i64>
   %m = mul <2 x i64> %xx, %yy
   %z = call i64 @llvm.experimental.vector.reduce.add.v2i64(<2 x i64> %m)
   %r = add i64 %z, %a
   ret i64 %r
 }
 
 define arm_aapcs_vfpcc i64 @add_v2i32_v2i64_acc_sext(<2 x i32> %x, <2 x i32> %y, i64 %a) {
 ; CHECK-LABEL: add_v2i32_v2i64_acc_sext:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    .save {r7, lr}
 ; CHECK-NEXT:    push {r7, lr}
-; CHECK-NEXT:    vmov r2, s4
-; CHECK-NEXT:    vmov r3, s0
-; CHECK-NEXT:    vmov r12, s6
-; CHECK-NEXT:    smull r2, lr, r3, r2
-; CHECK-NEXT:    vmov r3, s2
-; CHECK-NEXT:    smlal r2, lr, r3, r12
+; CHECK-NEXT:    vmullb.s32 q2, q0, q1
+; CHECK-NEXT:    vmov r2, s10
+; CHECK-NEXT:    vmov r3, s8
+; CHECK-NEXT:    vmov r12, s11
+; CHECK-NEXT:    vmov lr, s9
+; CHECK-NEXT:    adds r2, r2, r3
+; CHECK-NEXT:    adc.w r3, lr, r12
 ; CHECK-NEXT:    adds r0, r0, r2
-; CHECK-NEXT:    adc.w r1, r1, lr
+; CHECK-NEXT:    adcs r1, r3
 ; CHECK-NEXT:    pop {r7, pc}
 entry:
   %xx = sext <2 x i32> %x to <2 x i64>
   %yy = sext <2 x i32> %y to <2 x i64>
   %m = mul <2 x i64> %xx, %yy
   %z = call i64 @llvm.experimental.vector.reduce.add.v2i64(<2 x i64> %m)
   %r = add i64 %z, %a
   ret i64 %r

 ; CHECK-NEXT:    vmov.u8 r1, q0[12]
 ; CHECK-NEXT:    vmov.16 q3[4], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[13]
 ; CHECK-NEXT:    vmov.16 q3[5], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[14]
 ; CHECK-NEXT:    vmov.16 q3[6], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[15]
 ; CHECK-NEXT:    vmov.16 q3[7], r1
-; CHECK-NEXT:    vmovlb.u8 q2, q2
-; CHECK-NEXT:    vmovlb.u8 q3, q3
 ; CHECK-NEXT:    vmov.u8 r1, q1[0]
-; CHECK-NEXT:    vmul.i16 q2, q3, q2
+; CHECK-NEXT:    vmullb.u8 q2, q3, q2
 ; CHECK-NEXT:    vmov.16 q3[0], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[1]
 ; CHECK-NEXT:    vmov.16 q3[1], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[2]
 ; CHECK-NEXT:    vmov.16 q3[2], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[3]
 ; CHECK-NEXT:    vmov.16 q3[3], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[4]
 ; CHECK-NEXT:    vmov.16 q3[4], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[5]
 ; CHECK-NEXT:    vmov.16 q3[5], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[6]
 ; CHECK-NEXT:    vmov.16 q3[6], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[7]
 ; CHECK-NEXT:    vmov.16 q3[7], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[0]
-; CHECK-NEXT:    vmovlb.u8 q1, q3
-; CHECK-NEXT:    vmov.16 q3[0], r1
+; CHECK-NEXT:    vmov.16 q1[0], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[1]
-; CHECK-NEXT:    vmov.16 q3[1], r1
+; CHECK-NEXT:    vmov.16 q1[1], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[2]
-; CHECK-NEXT:    vmov.16 q3[2], r1
+; CHECK-NEXT:    vmov.16 q1[2], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[3]
-; CHECK-NEXT:    vmov.16 q3[3], r1
+; CHECK-NEXT:    vmov.16 q1[3], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[4]
-; CHECK-NEXT:    vmov.16 q3[4], r1
+; CHECK-NEXT:    vmov.16 q1[4], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[5]
-; CHECK-NEXT:    vmov.16 q3[5], r1
+; CHECK-NEXT:    vmov.16 q1[5], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[6]
-; CHECK-NEXT:    vmov.16 q3[6], r1
+; CHECK-NEXT:    vmov.16 q1[6], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[7]
-; CHECK-NEXT:    vmov.16 q3[7], r1
-; CHECK-NEXT:    vmovlb.u8 q0, q3
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmov.16 q1[7], r1
+; CHECK-NEXT:    vmullb.u8 q0, q1, q3
 ; CHECK-NEXT:    vadd.i16 q0, q0, q2
 ; CHECK-NEXT:    vaddva.u16 r0, q0
 ; CHECK-NEXT:    uxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = zext <16 x i8> %x to <16 x i16>
   %yy = zext <16 x i8> %y to <16 x i16>
   %m = mul <16 x i16> %xx, %yy

 ; CHECK-NEXT:    vmov.u8 r1, q0[12]
 ; CHECK-NEXT:    vmov.16 q3[4], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[13]
 ; CHECK-NEXT:    vmov.16 q3[5], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[14]
 ; CHECK-NEXT:    vmov.16 q3[6], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[15]
 ; CHECK-NEXT:    vmov.16 q3[7], r1
-; CHECK-NEXT:    vmovlb.s8 q2, q2
-; CHECK-NEXT:    vmovlb.s8 q3, q3
 ; CHECK-NEXT:    vmov.u8 r1, q1[0]
-; CHECK-NEXT:    vmul.i16 q2, q3, q2
+; CHECK-NEXT:    vmullb.s8 q2, q3, q2
 ; CHECK-NEXT:    vmov.16 q3[0], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[1]
 ; CHECK-NEXT:    vmov.16 q3[1], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[2]
 ; CHECK-NEXT:    vmov.16 q3[2], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[3]
 ; CHECK-NEXT:    vmov.16 q3[3], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[4]
 ; CHECK-NEXT:    vmov.16 q3[4], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[5]
 ; CHECK-NEXT:    vmov.16 q3[5], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[6]
 ; CHECK-NEXT:    vmov.16 q3[6], r1
 ; CHECK-NEXT:    vmov.u8 r1, q1[7]
 ; CHECK-NEXT:    vmov.16 q3[7], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[0]
-; CHECK-NEXT:    vmovlb.s8 q1, q3
-; CHECK-NEXT:    vmov.16 q3[0], r1
+; CHECK-NEXT:    vmov.16 q1[0], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[1]
-; CHECK-NEXT:    vmov.16 q3[1], r1
+; CHECK-NEXT:    vmov.16 q1[1], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[2]
-; CHECK-NEXT:    vmov.16 q3[2], r1
+; CHECK-NEXT:    vmov.16 q1[2], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[3]
-; CHECK-NEXT:    vmov.16 q3[3], r1
+; CHECK-NEXT:    vmov.16 q1[3], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[4]
-; CHECK-NEXT:    vmov.16 q3[4], r1
+; CHECK-NEXT:    vmov.16 q1[4], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[5]
-; CHECK-NEXT:    vmov.16 q3[5], r1
+; CHECK-NEXT:    vmov.16 q1[5], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[6]
-; CHECK-NEXT:    vmov.16 q3[6], r1
+; CHECK-NEXT:    vmov.16 q1[6], r1
 ; CHECK-NEXT:    vmov.u8 r1, q0[7]
-; CHECK-NEXT:    vmov.16 q3[7], r1
-; CHECK-NEXT:    vmovlb.s8 q0, q3
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmov.16 q1[7], r1
+; CHECK-NEXT:    vmullb.s8 q0, q1, q3
 ; CHECK-NEXT:    vadd.i16 q0, q0, q2
 ; CHECK-NEXT:    vaddva.u16 r0, q0
 ; CHECK-NEXT:    sxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = sext <16 x i8> %x to <16 x i16>
   %yy = sext <16 x i8> %y to <16 x i16>
   %m = mul <16 x i16> %xx, %yy
   %z = call i16 @llvm.experimental.vector.reduce.add.v16i16(<16 x i16> %m)
   %r = add i16 %z, %a
   ret i16 %r
 }
 
 define arm_aapcs_vfpcc zeroext i16 @add_v8i8_v8i16_acc_zext(<8 x i8> %x, <8 x i8> %y, i16 %a) {
 ; CHECK-LABEL: add_v8i8_v8i16_acc_zext:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.u8 q1, q1
-; CHECK-NEXT:    vmovlb.u8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullb.u8 q0, q0, q1
 ; CHECK-NEXT:    vaddva.u16 r0, q0
 ; CHECK-NEXT:    uxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = zext <8 x i8> %x to <8 x i16>
   %yy = zext <8 x i8> %y to <8 x i16>
   %m = mul <8 x i16> %xx, %yy
   %z = call i16 @llvm.experimental.vector.reduce.add.v8i16(<8 x i16> %m)
   %r = add i16 %z, %a
   ret i16 %r
 }
 
 define arm_aapcs_vfpcc signext i16 @add_v8i8_v8i16_acc_sext(<8 x i8> %x, <8 x i8> %y, i16 %a) {
 ; CHECK-LABEL: add_v8i8_v8i16_acc_sext:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.s8 q1, q1
-; CHECK-NEXT:    vmovlb.s8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullb.s8 q0, q0, q1
 ; CHECK-NEXT:    vaddva.u16 r0, q0
 ; CHECK-NEXT:    sxth r0, r0
 ; CHECK-NEXT:    bx lr
 entry:
   %xx = sext <8 x i8> %x to <8 x i16>
   %yy = sext <8 x i8> %y to <8 x i16>
   %m = mul <8 x i16> %xx, %yy
   %z = call i16 @llvm.experimental.vector.reduce.add.v8i16(<8 x i16> %m)

llvm/test/CodeGen/Thumb2/mve-vmull-loop.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=CHECK
 
 define arm_aapcs_vfpcc void @test32(i32* noalias nocapture readonly %x, i32* noalias nocapture readonly %y, i32* nocapture %z, i32 %n) {
 ; CHECK-LABEL: test32:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    .save {r5, lr}
 ; CHECK-NEXT:    push {r5, lr}
-; CHECK-NEXT:    .vsave {d8, d9}
-; CHECK-NEXT:    vpush {d8, d9}
 ; CHECK-NEXT:    cmp r3, #1
-; CHECK-NEXT:    blt .LBB0_2
+; CHECK-NEXT:    it lt
+; CHECK-NEXT:    poplt {r5, pc}
 ; CHECK-NEXT:  .LBB0_1: @ %vector.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
-; CHECK-NEXT:    vldrw.u32 q2, [r0], #16
-; CHECK-NEXT:    vldrw.u32 q3, [r1], #16
+; CHECK-NEXT:    vldrw.u32 q1, [r0], #16
+; CHECK-NEXT:    vldrw.u32 q2, [r1], #16
 ; CHECK-NEXT:    subs r3, #4
-; CHECK-NEXT:    vrev64.32 q1, q2
-; CHECK-NEXT:    vrev64.32 q4, q3
-; CHECK-NEXT:    vmov r12, s4
-; CHECK-NEXT:    vmov lr, s16
-; CHECK-NEXT:    smull r12, r5, lr, r12
+; CHECK-NEXT:    vmullt.s32 q3, q2, q1
+; CHECK-NEXT:    vmov r5, s13
+; CHECK-NEXT:    vmov r12, s12
 ; CHECK-NEXT:    lsrl r12, r5, #31
 ; CHECK-NEXT:    vmov.32 q0[0], r12
-; CHECK-NEXT:    vmov r12, s6
+; CHECK-NEXT:    vmov r12, s14
 ; CHECK-NEXT:    vmov.32 q0[1], r5
-; CHECK-NEXT:    vmov r5, s18
-; CHECK-NEXT:    smull r12, r5, r5, r12
+; CHECK-NEXT:    vmov r5, s15
 ; CHECK-NEXT:    lsrl r12, r5, #31
+; CHECK-NEXT:    vmullb.s32 q3, q2, q1
 ; CHECK-NEXT:    vmov.32 q0[2], r12
-; CHECK-NEXT:    vmov r12, s8
+; CHECK-NEXT:    vmov r12, s12
 ; CHECK-NEXT:    vmov.32 q0[3], r5
-; CHECK-NEXT:    vmov r5, s12
-; CHECK-NEXT:    smull r12, r5, r5, r12
+; CHECK-NEXT:    vmov r5, s13
 ; CHECK-NEXT:    lsrl r12, r5, #31
 ; CHECK-NEXT:    vmov.32 q1[0], r12
-; CHECK-NEXT:    vmov r12, s10
+; CHECK-NEXT:    vmov r12, s14
 ; CHECK-NEXT:    vmov.32 q1[1], r5
-; CHECK-NEXT:    vmov r5, s14
-; CHECK-NEXT:    smull r12, r5, r5, r12
+; CHECK-NEXT:    vmov r5, s15
 ; CHECK-NEXT:    lsrl r12, r5, #31
 ; CHECK-NEXT:    vmov.32 q1[2], r12
 ; CHECK-NEXT:    vmov.32 q1[3], r5
 ; CHECK-NEXT:    vmov.f32 s8, s6
 ; CHECK-NEXT:    vmov.f32 s9, s7
 ; CHECK-NEXT:    vmov.f32 s6, s0
 ; CHECK-NEXT:    vmov.f32 s7, s1
 ; CHECK-NEXT:    vmov.f32 s10, s2
 ; CHECK-NEXT:    vmov.f32 s5, s6
 ; CHECK-NEXT:    vmov.f32 s11, s3
 ; CHECK-NEXT:    vmov.f32 s6, s8
 ; CHECK-NEXT:    vmov.f32 s7, s10
 ; CHECK-NEXT:    vstrb.8 q1, [r2], #16
 ; CHECK-NEXT:    bne .LBB0_1
-; CHECK-NEXT:  .LBB0_2: @ %for.cond.cleanup
-; CHECK-NEXT:    vpop {d8, d9}
+; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
 ; CHECK-NEXT:    pop {r5, pc}
 entry:
   %0 = and i32 %n, 3
   %cmp = icmp eq i32 %0, 0
   %cmp113 = icmp sgt i32 %n, 0
   br i1 %cmp113, label %vector.body, label %for.cond.cleanup
 
 vector.body:                                      ; preds = %vector.body, %entry

 ; CHECK-LABEL: test16:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    cmp r3, #1
 ; CHECK-NEXT:    it lt
 ; CHECK-NEXT:    bxlt lr
 ; CHECK-NEXT:  .LBB1_1: @ %vector.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
 ; CHECK-NEXT:    vldrh.u16 q0, [r0], #16
-; CHECK-NEXT:    vldrh.u16 q2, [r1], #16
+; CHECK-NEXT:    vldrh.u16 q1, [r1], #16
 ; CHECK-NEXT:    subs r3, #8
-; CHECK-NEXT:    vmovlt.s16 q1, q0
-; CHECK-NEXT:    vmovlt.s16 q3, q2
-; CHECK-NEXT:    vmovlb.s16 q0, q0
-; CHECK-NEXT:    vmovlb.s16 q2, q2
-; CHECK-NEXT:    vmul.i32 q1, q3, q1
-; CHECK-NEXT:    vmul.i32 q0, q2, q0
-; CHECK-NEXT:    vshr.u32 q1, q1, #15
+; CHECK-NEXT:    vmullt.s16 q2, q1, q0
+; CHECK-NEXT:    vmullb.s16 q0, q1, q0
+; CHECK-NEXT:    vshr.u32 q2, q2, #15
 ; CHECK-NEXT:    vshr.u32 q0, q0, #15
-; CHECK-NEXT:    vmovnt.i32 q0, q1
+; CHECK-NEXT:    vmovnt.i32 q0, q2
 ; CHECK-NEXT:    vstrb.8 q0, [r2], #16
 ; CHECK-NEXT:    bne .LBB1_1
 ; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = and i32 %n, 7
   %cmp = icmp eq i32 %0, 0
   %cmp113 = icmp sgt i32 %n, 0

 ; CHECK-LABEL: test8:
 ; CHECK:       @ %bb.0: @ %entry
 ; CHECK-NEXT:    cmp r3, #1
 ; CHECK-NEXT:    it lt
 ; CHECK-NEXT:    bxlt lr
 ; CHECK-NEXT:  .LBB2_1: @ %vector.body
 ; CHECK-NEXT:    @ =>This Inner Loop Header: Depth=1
 ; CHECK-NEXT:    vldrb.u8 q0, [r0], #16
-; CHECK-NEXT:    vldrb.u8 q2, [r1], #16
+; CHECK-NEXT:    vldrb.u8 q1, [r1], #16
 ; CHECK-NEXT:    subs r3, #16
-; CHECK-NEXT:    vmovlt.u8 q1, q0
-; CHECK-NEXT:    vmovlt.u8 q3, q2
-; CHECK-NEXT:    vmovlb.u8 q0, q0
-; CHECK-NEXT:    vmovlb.u8 q2, q2
-; CHECK-NEXT:    vmul.i16 q1, q3, q1
-; CHECK-NEXT:    vmul.i16 q0, q2, q0
-; CHECK-NEXT:    vshr.u16 q1, q1, #7
+; CHECK-NEXT:    vmullt.u8 q2, q1, q0
+; CHECK-NEXT:    vmullb.u8 q0, q1, q0
+; CHECK-NEXT:    vshr.u16 q2, q2, #7
 ; CHECK-NEXT:    vshr.u16 q0, q0, #7
-; CHECK-NEXT:    vmovnt.i16 q0, q1
+; CHECK-NEXT:    vmovnt.i16 q0, q2
 ; CHECK-NEXT:    vstrb.8 q0, [r2], #16
 ; CHECK-NEXT:    bne .LBB2_1
 ; CHECK-NEXT:  @ %bb.2: @ %for.cond.cleanup
 ; CHECK-NEXT:    bx lr
 entry:
   %0 = and i32 %n, 15
   %cmp = icmp eq i32 %0, 0
   %cmp117 = icmp sgt i32 %n, 0

llvm/test/CodeGen/Thumb2/mve-vmull.ll

 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc -mtriple=thumbv8.1m.main-arm-none-eabi -mattr=+mve -verify-machineinstrs %s -o - | FileCheck %s --check-prefix=CHECK
 
 define arm_aapcs_vfpcc <2 x i64> @sext_02(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: sext_02:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    smull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q2[0], r0
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.32 q2[1], r1
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    smull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q2[2], r0
-; CHECK-NEXT:    vmov.32 q2[3], r1
+; CHECK-NEXT:    vmullb.s32 q2, q0, q1
 ; CHECK-NEXT:    vmov q0, q2
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <4 x i32> %src1, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
   %out1 = sext <2 x i32> %shuf1 to <2 x i64>
   %shuf2 = shufflevector <4 x i32> %src2, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
   %out2 = sext <2 x i32> %shuf2 to <2 x i64>
   %out = mul <2 x i64> %out1, %out2
   ret <2 x i64> %out
 }
 
 define arm_aapcs_vfpcc <2 x i64> @sext_13(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: sext_13:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vrev64.32 q2, q1
-; CHECK-NEXT:    vrev64.32 q1, q0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    smull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q0[0], r0
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.32 q0[1], r1
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    smull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q0[2], r0
-; CHECK-NEXT:    vmov.32 q0[3], r1
+; CHECK-NEXT:    vmullt.s32 q2, q0, q1
+; CHECK-NEXT:    vmov q0, q2
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <4 x i32> %src1, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
   %out1 = sext <2 x i32> %shuf1 to <2 x i64>
   %shuf2 = shufflevector <4 x i32> %src2, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
   %out2 = sext <2 x i32> %shuf2 to <2 x i64>
   %out = mul <2 x i64> %out1, %out2
   ret <2 x i64> %out
 }
 
 define arm_aapcs_vfpcc <2 x i64> @zext_02(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: zext_02:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmov r0, s4
-; CHECK-NEXT:    vmov r1, s0
-; CHECK-NEXT:    umull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q2[0], r0
-; CHECK-NEXT:    vmov r0, s6
-; CHECK-NEXT:    vmov.32 q2[1], r1
-; CHECK-NEXT:    vmov r1, s2
-; CHECK-NEXT:    umull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q2[2], r0
-; CHECK-NEXT:    vmov.32 q2[3], r1
+; CHECK-NEXT:    vmullb.u32 q2, q0, q1
 ; CHECK-NEXT:    vmov q0, q2
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <4 x i32> %src1, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
   %out1 = zext <2 x i32> %shuf1 to <2 x i64>
   %shuf2 = shufflevector <4 x i32> %src2, <4 x i32> undef, <2 x i32> <i32 0, i32 2>
   %out2 = zext <2 x i32> %shuf2 to <2 x i64>
   %out = mul <2 x i64> %out1, %out2
   ret <2 x i64> %out
 }
 
 define arm_aapcs_vfpcc <2 x i64> @zext_13(<4 x i32> %src1, <4 x i32> %src2) {
 ; CHECK-LABEL: zext_13:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vrev64.32 q2, q1
-; CHECK-NEXT:    vrev64.32 q1, q0
-; CHECK-NEXT:    vmov r0, s8
-; CHECK-NEXT:    vmov r1, s4
-; CHECK-NEXT:    umull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q0[0], r0
-; CHECK-NEXT:    vmov r0, s10
-; CHECK-NEXT:    vmov.32 q0[1], r1
-; CHECK-NEXT:    vmov r1, s6
-; CHECK-NEXT:    umull r0, r1, r1, r0
-; CHECK-NEXT:    vmov.32 q0[2], r0
-; CHECK-NEXT:    vmov.32 q0[3], r1
+; CHECK-NEXT:    vmullt.u32 q2, q0, q1
+; CHECK-NEXT:    vmov q0, q2
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <4 x i32> %src1, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
   %out1 = zext <2 x i32> %shuf1 to <2 x i64>
   %shuf2 = shufflevector <4 x i32> %src2, <4 x i32> undef, <2 x i32> <i32 1, i32 3>
   %out2 = zext <2 x i32> %shuf2 to <2 x i64>
   %out = mul <2 x i64> %out1, %out2
   ret <2 x i64> %out
 }
 
 
 define arm_aapcs_vfpcc <4 x i32> @sext_0246(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: sext_0246:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.s16 q1, q1
-; CHECK-NEXT:    vmovlb.s16 q0, q0
-; CHECK-NEXT:    vmul.i32 q0, q0, q1
+; CHECK-NEXT:    vmullb.s16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <8 x i16> %src1, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
   %out1 = sext <4 x i16> %shuf1 to <4 x i32>
   %shuf2 = shufflevector <8 x i16> %src2, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
   %out2 = sext <4 x i16> %shuf2 to <4 x i32>
   %out = mul <4 x i32> %out1, %out2
   ret <4 x i32> %out
 }
 
 define arm_aapcs_vfpcc <4 x i32> @sext_1357(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: sext_1357:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlt.s16 q1, q1
-; CHECK-NEXT:    vmovlt.s16 q0, q0
-; CHECK-NEXT:    vmul.i32 q0, q0, q1
+; CHECK-NEXT:    vmullt.s16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <8 x i16> %src1, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
   %out1 = sext <4 x i16> %shuf1 to <4 x i32>
   %shuf2 = shufflevector <8 x i16> %src2, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
   %out2 = sext <4 x i16> %shuf2 to <4 x i32>
   %out = mul <4 x i32> %out1, %out2
   ret <4 x i32> %out
 }
 
 define arm_aapcs_vfpcc <4 x i32> @zext_0246(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: zext_0246:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.u16 q1, q1
-; CHECK-NEXT:    vmovlb.u16 q0, q0
-; CHECK-NEXT:    vmul.i32 q0, q0, q1
+; CHECK-NEXT:    vmullb.u16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <8 x i16> %src1, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
   %out1 = zext <4 x i16> %shuf1 to <4 x i32>
   %shuf2 = shufflevector <8 x i16> %src2, <8 x i16> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
   %out2 = zext <4 x i16> %shuf2 to <4 x i32>
   %out = mul <4 x i32> %out1, %out2
   ret <4 x i32> %out
 }
 
 define arm_aapcs_vfpcc <4 x i32> @zext_1357(<8 x i16> %src1, <8 x i16> %src2) {
 ; CHECK-LABEL: zext_1357:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlt.u16 q1, q1
-; CHECK-NEXT:    vmovlt.u16 q0, q0
-; CHECK-NEXT:    vmul.i32 q0, q0, q1
+; CHECK-NEXT:    vmullt.u16 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <8 x i16> %src1, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
   %out1 = zext <4 x i16> %shuf1 to <4 x i32>
   %shuf2 = shufflevector <8 x i16> %src2, <8 x i16> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
   %out2 = zext <4 x i16> %shuf2 to <4 x i32>
   %out = mul <4 x i32> %out1, %out2
   ret <4 x i32> %out
 }
 
 define arm_aapcs_vfpcc <8 x i16> @sext_02468101214(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: sext_02468101214:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.s8 q1, q1
-; CHECK-NEXT:    vmovlb.s8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullb.s8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <16 x i8> %src1, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
   %out1 = sext <8 x i8> %shuf1 to <8 x i16>
   %shuf2 = shufflevector <16 x i8> %src2, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
   %out2 = sext <8 x i8> %shuf2 to <8 x i16>
   %out = mul <8 x i16> %out1, %out2
   ret <8 x i16> %out
 }
 
 define arm_aapcs_vfpcc <8 x i16> @sext_13579111315(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: sext_13579111315:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlt.s8 q1, q1
-; CHECK-NEXT:    vmovlt.s8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullt.s8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <16 x i8> %src1, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
   %out1 = sext <8 x i8> %shuf1 to <8 x i16>
   %shuf2 = shufflevector <16 x i8> %src2, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
   %out2 = sext <8 x i8> %shuf2 to <8 x i16>
   %out = mul <8 x i16> %out1, %out2
   ret <8 x i16> %out
 }
 
 define arm_aapcs_vfpcc <8 x i16> @zext_02468101214(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: zext_02468101214:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlb.u8 q1, q1
-; CHECK-NEXT:    vmovlb.u8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullb.u8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <16 x i8> %src1, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
   %out1 = zext <8 x i8> %shuf1 to <8 x i16>
   %shuf2 = shufflevector <16 x i8> %src2, <16 x i8> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
   %out2 = zext <8 x i8> %shuf2 to <8 x i16>
   %out = mul <8 x i16> %out1, %out2
   ret <8 x i16> %out
 }
 
 define arm_aapcs_vfpcc <8 x i16> @zext_13579111315(<16 x i8> %src1, <16 x i8> %src2) {
 ; CHECK-LABEL: zext_13579111315:
 ; CHECK:       @ %bb.0: @ %entry
-; CHECK-NEXT:    vmovlt.u8 q1, q1
-; CHECK-NEXT:    vmovlt.u8 q0, q0
-; CHECK-NEXT:    vmul.i16 q0, q0, q1
+; CHECK-NEXT:    vmullt.u8 q0, q0, q1
 ; CHECK-NEXT:    bx lr
 entry:
   %shuf1 = shufflevector <16 x i8> %src1, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
   %out1 = zext <8 x i8> %shuf1 to <8 x i16>
   %shuf2 = shufflevector <16 x i8> %src2, <16 x i8> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
   %out2 = zext <8 x i8> %shuf2 to <8 x i16>
   %out = mul <8 x i16> %out1, %out2
   ret <8 x i16> %out
 }