Index: llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
===================================================================
--- llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -5517,19 +5517,38 @@
                                 SDValue &Overflow, SelectionDAG &DAG) const {
   SDLoc dl(Node);
   EVT VT = Node->getValueType(0);
+  EVT SetCCVT = getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT);
+  SDValue LHS = Node->getOperand(0);
+  SDValue RHS = Node->getOperand(1);
+  bool isSigned = Node->getOpcode() == ISD::SMULO;
+
+  // For power-of-two multiplications we can use a simpler shift expansion.
+  if (ConstantSDNode *RHSC = isConstOrConstSplat(RHS)) {
+    const APInt &C = RHSC->getAPIntValue();
+    if (C.isPowerOf2()) {
+      EVT ShiftAmtTy = getShiftAmountTy(VT, DAG.getDataLayout());
+      SDValue ShiftAmt = DAG.getConstant(C.logBase2(), dl, ShiftAmtTy);
+      Result = DAG.getNode(ISD::SHL, dl, VT, LHS, ShiftAmt);
+
+      // Check whether ((X << Shift) >> Shift) == X.
+      SDValue ShiftedResult = DAG.getNode(
+          isSigned ? ISD::SRA : ISD::SRL, dl, VT, Result, ShiftAmt);
+      Overflow = DAG.getSetCC(
+          dl, SetCCVT, ShiftedResult, LHS, ISD::SETNE);
+      return true;
+    }
+  }
+
   EVT WideVT = EVT::getIntegerVT(*DAG.getContext(), VT.getScalarSizeInBits() * 2);
   if (VT.isVector())
     WideVT = EVT::getVectorVT(*DAG.getContext(), WideVT,
                               VT.getVectorNumElements());
 
-  SDValue LHS = Node->getOperand(0);
-  SDValue RHS = Node->getOperand(1);
   SDValue BottomHalf;
   SDValue TopHalf;
   static const unsigned Ops[2][3] =
       { { ISD::MULHU, ISD::UMUL_LOHI, ISD::ZERO_EXTEND },
         { ISD::MULHS, ISD::SMUL_LOHI, ISD::SIGN_EXTEND }};
-  bool isSigned = Node->getOpcode() == ISD::SMULO;
   if (isOperationLegalOrCustom(Ops[isSigned][0], VT)) {
     BottomHalf = DAG.getNode(ISD::MUL, dl, VT, LHS, RHS);
     TopHalf = DAG.getNode(Ops[isSigned][0], dl, VT, LHS, RHS);
@@ -5616,7 +5635,6 @@
     }
   }
 
-  EVT SetCCVT = getSetCCResultType(DAG.getDataLayout(), *DAG.getContext(), VT);
   Result = BottomHalf;
   if (isSigned) {
     SDValue ShiftAmt = DAG.getConstant(
Index: llvm/test/CodeGen/X86/mulo-pow2.ll
===================================================================
--- llvm/test/CodeGen/X86/mulo-pow2.ll
+++ llvm/test/CodeGen/X86/mulo-pow2.ll
@@ -19,15 +19,6 @@
 define <4 x i32> @umul_v4i32_1(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX-LABEL: umul_v4i32_1:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1]
-; AVX-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
-; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
     %x = call { <4 x i32>, <4 x i1> } @llvm.umul.with.overflow.v4i32(<4 x i32> %a, <4 x i32> <i32 1, i32 1, i32 1, i32 1>)
     %y = extractvalue { <4 x i32>, <4 x i1> } %x, 0
@@ -54,14 +45,8 @@
 define <4 x i32> @umul_v4i32_8(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX-LABEL: umul_v4i32_8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8]
-; AVX-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
-; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm2
+; AVX-NEXT:    vpcmpeqd %xmm0, %xmm2, %xmm2
 ; AVX-NEXT:    vpslld $3, %xmm0, %xmm0
 ; AVX-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
@@ -75,14 +60,8 @@
 define <4 x i32> @umul_v4i32_2pow31(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX-LABEL: umul_v4i32_2pow31:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; AVX-NEXT:    vpmuludq %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpmuludq %xmm3, %xmm0, %xmm3
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
-; AVX-NEXT:    vpxor %xmm3, %xmm3, %xmm3
-; AVX-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
+; AVX-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm2
+; AVX-NEXT:    vpcmpeqd %xmm0, %xmm2, %xmm2
 ; AVX-NEXT:    vpslld $31, %xmm0, %xmm0
 ; AVX-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
@@ -108,15 +87,6 @@
 define <4 x i32> @smul_v4i32_1(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX-LABEL: smul_v4i32_1:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [1,1,1,1]
-; AVX-NEXT:    vpmuldq %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpmuldq %xmm3, %xmm0, %xmm3
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
-; AVX-NEXT:    vpsrad $31, %xmm0, %xmm3
-; AVX-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
 ; AVX-NEXT:    retq
     %x = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> %a, <4 x i32> <i32 1, i32 1, i32 1, i32 1>)
     %y = extractvalue { <4 x i32>, <4 x i1> } %x, 0
@@ -148,16 +118,10 @@
 define <4 x i32> @smul_v4i32_8(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX-LABEL: smul_v4i32_8:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [8,8,8,8]
-; AVX-NEXT:    vpmuldq %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpmuldq %xmm3, %xmm0, %xmm3
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
-; AVX-NEXT:    vpslld $3, %xmm0, %xmm0
-; AVX-NEXT:    vpsrad $31, %xmm0, %xmm3
-; AVX-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpslld $3, %xmm0, %xmm2
+; AVX-NEXT:    vpsrad $3, %xmm2, %xmm3
+; AVX-NEXT:    vpcmpeqd %xmm0, %xmm3, %xmm0
+; AVX-NEXT:    vblendvps %xmm0, %xmm2, %xmm1, %xmm0
 ; AVX-NEXT:    retq
     %x = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> %a, <4 x i32> <i32 8, i32 8, i32 8, i32 8>)
     %y = extractvalue { <4 x i32>, <4 x i1> } %x, 0
@@ -169,16 +133,10 @@
 define <4 x i32> @smul_v4i32_2pow31(<4 x i32> %a, <4 x i32> %b) nounwind {
 ; AVX-LABEL: smul_v4i32_2pow31:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[1,1,3,3]
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm3 = [2147483648,2147483648,2147483648,2147483648]
-; AVX-NEXT:    vpmuldq %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpmuldq %xmm3, %xmm0, %xmm3
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[1,1,3,3]
-; AVX-NEXT:    vpblendw {{.*#+}} xmm2 = xmm3[0,1],xmm2[2,3],xmm3[4,5],xmm2[6,7]
-; AVX-NEXT:    vpslld $31, %xmm0, %xmm0
-; AVX-NEXT:    vpsrad $31, %xmm0, %xmm3
-; AVX-NEXT:    vpcmpeqd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vblendvps %xmm2, %xmm0, %xmm1, %xmm0
+; AVX-NEXT:    vpslld $31, %xmm0, %xmm2
+; AVX-NEXT:    vpsrad $31, %xmm2, %xmm3
+; AVX-NEXT:    vpcmpeqd %xmm0, %xmm3, %xmm0
+; AVX-NEXT:    vblendvps %xmm0, %xmm2, %xmm1, %xmm0
 ; AVX-NEXT:    retq
     %x = call { <4 x i32>, <4 x i1> } @llvm.smul.with.overflow.v4i32(<4 x i32> %a, <4 x i32> <i32 -2147483648, i32 -2147483648, i32 -2147483648, i32 -2147483648>)
     %y = extractvalue { <4 x i32>, <4 x i1> } %x, 0