Index: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
@@ -38534,6 +38534,33 @@
   return SDValue();
 }
 
+// Simplify PMULDQ and PMULUDQ operations.
+static SDValue combinePMULDQ(SDNode *N, SelectionDAG &DAG,
+                             TargetLowering::DAGCombinerInfo &DCI) {
+  SDValue LHS = N->getOperand(0);
+  SDValue RHS = N->getOperand(1);
+
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+  TargetLowering::TargetLoweringOpt TLO(DAG, !DCI.isBeforeLegalize(),
+                                        !DCI.isBeforeLegalizeOps());
+  APInt DemandedMask(APInt::getLowBitsSet(64, 32));
+
+  // PMULQDQ/PMULUDQ only uses lower 32 bits from each vector element.
+  KnownBits LHSKnown;
+  if (TLI.SimplifyDemandedBits(LHS, DemandedMask, LHSKnown, TLO)) {
+    DCI.CommitTargetLoweringOpt(TLO);
+    return SDValue(N, 0);
+  }
+
+  KnownBits RHSKnown;
+  if (TLI.SimplifyDemandedBits(RHS, DemandedMask, RHSKnown, TLO)) {
+    DCI.CommitTargetLoweringOpt(TLO);
+    return SDValue(N, 0);
+  }
+
+  return SDValue();
+}
+
 SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
                                              DAGCombinerInfo &DCI) const {
   SelectionDAG &DAG = DCI.DAG;
@@ -38655,6 +38682,8 @@
   case ISD::MSCATTER:       return combineGatherScatter(N, DAG, DCI, Subtarget);
   case X86ISD::PCMPEQ:
   case X86ISD::PCMPGT:      return combineVectorCompare(N, DAG, Subtarget);
+  case X86ISD::PMULDQ:
+  case X86ISD::PMULUDQ:     return combinePMULDQ(N, DAG, DCI);
   }
 
   return SDValue();
Index: llvm/trunk/test/CodeGen/X86/vector-mul.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/vector-mul.ll
+++ llvm/trunk/test/CodeGen/X86/vector-mul.ll
@@ -692,17 +692,16 @@
 define <2 x i64> @mul_v2i64_neg_15_63(<2 x i64> %a0) nounwind {
 ; X86-LABEL: mul_v2i64_neg_15_63:
 ; X86:       # %bb.0:
-; X86-NEXT:    movdqa %xmm0, %xmm1
-; X86-NEXT:    psrlq $32, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm2 = [4294967281,4294967295,4294967233,4294967295]
-; X86-NEXT:    pmuludq %xmm2, %xmm1
-; X86-NEXT:    movdqa %xmm2, %xmm3
+; X86-NEXT:    movdqa {{.*#+}} xmm1 = [4294967281,4294967295,4294967233,4294967295]
+; X86-NEXT:    movdqa %xmm0, %xmm2
+; X86-NEXT:    pmuludq %xmm1, %xmm2
+; X86-NEXT:    movdqa %xmm0, %xmm3
 ; X86-NEXT:    psrlq $32, %xmm3
-; X86-NEXT:    pmuludq %xmm0, %xmm3
-; X86-NEXT:    paddq %xmm1, %xmm3
-; X86-NEXT:    psllq $32, %xmm3
-; X86-NEXT:    pmuludq %xmm2, %xmm0
+; X86-NEXT:    pmuludq %xmm1, %xmm3
+; X86-NEXT:    pmuludq {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    paddq %xmm3, %xmm0
+; X86-NEXT:    psllq $32, %xmm0
+; X86-NEXT:    paddq %xmm2, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: mul_v2i64_neg_15_63:
@@ -737,17 +736,16 @@
 define <2 x i64> @mul_v2i64_neg_17_65(<2 x i64> %a0) nounwind {
 ; X86-LABEL: mul_v2i64_neg_17_65:
 ; X86:       # %bb.0:
-; X86-NEXT:    movdqa %xmm0, %xmm1
-; X86-NEXT:    psrlq $32, %xmm1
-; X86-NEXT:    movdqa {{.*#+}} xmm2 = [4294967279,4294967295,4294967231,4294967295]
-; X86-NEXT:    pmuludq %xmm2, %xmm1
-; X86-NEXT:    movdqa %xmm2, %xmm3
+; X86-NEXT:    movdqa {{.*#+}} xmm1 = [4294967279,4294967295,4294967231,4294967295]
+; X86-NEXT:    movdqa %xmm0, %xmm2
+; X86-NEXT:    pmuludq %xmm1, %xmm2
+; X86-NEXT:    movdqa %xmm0, %xmm3
 ; X86-NEXT:    psrlq $32, %xmm3
-; X86-NEXT:    pmuludq %xmm0, %xmm3
-; X86-NEXT:    paddq %xmm1, %xmm3
-; X86-NEXT:    psllq $32, %xmm3
-; X86-NEXT:    pmuludq %xmm2, %xmm0
+; X86-NEXT:    pmuludq %xmm1, %xmm3
+; X86-NEXT:    pmuludq {{\.LCPI.*}}, %xmm0
 ; X86-NEXT:    paddq %xmm3, %xmm0
+; X86-NEXT:    psllq $32, %xmm0
+; X86-NEXT:    paddq %xmm2, %xmm0
 ; X86-NEXT:    retl
 ;
 ; X64-LABEL: mul_v2i64_neg_17_65: