Index: llvm/trunk/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
===================================================================
--- llvm/trunk/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
+++ llvm/trunk/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
@@ -1179,16 +1179,41 @@
     switch (II->getIntrinsicID()) {
     default: break;
 
+    // Unary vector operations that work column-wise.
+    case Intrinsic::x86_sse_rcp_ss:
+    case Intrinsic::x86_sse_rsqrt_ss:
+    case Intrinsic::x86_sse_sqrt_ss:
+    case Intrinsic::x86_sse2_sqrt_sd:
+    case Intrinsic::x86_xop_vfrcz_ss:
+    case Intrinsic::x86_xop_vfrcz_sd:
+      TmpV = SimplifyDemandedVectorElts(II->getArgOperand(0), DemandedElts,
+                                        UndefElts, Depth + 1);
+      if (TmpV) { II->setArgOperand(0, TmpV); MadeChange = true; }
+
+      // If lowest element of a scalar op isn't used then use Arg0.
+      if (DemandedElts.getLoBits(1) != 1)
+        return II->getArgOperand(0);
+      // TODO: If only low elt lower SQRT to FSQRT (with rounding/exceptions checks).
+      break;
+
     // Binary vector operations that work column-wise.  A dest element is a
     // function of the corresponding input elements from the two inputs.
+    case Intrinsic::x86_sse_add_ss:
     case Intrinsic::x86_sse_sub_ss:
     case Intrinsic::x86_sse_mul_ss:
+    case Intrinsic::x86_sse_div_ss:
     case Intrinsic::x86_sse_min_ss:
     case Intrinsic::x86_sse_max_ss:
+    case Intrinsic::x86_sse_cmp_ss:
+    case Intrinsic::x86_sse2_add_sd:
     case Intrinsic::x86_sse2_sub_sd:
     case Intrinsic::x86_sse2_mul_sd:
+    case Intrinsic::x86_sse2_div_sd:
     case Intrinsic::x86_sse2_min_sd:
     case Intrinsic::x86_sse2_max_sd:
+    case Intrinsic::x86_sse2_cmp_sd:
+    case Intrinsic::x86_sse41_round_ss:
+    case Intrinsic::x86_sse41_round_sd:
       TmpV = SimplifyDemandedVectorElts(II->getArgOperand(0), DemandedElts,
                                         UndefElts, Depth + 1);
       if (TmpV) { II->setArgOperand(0, TmpV); MadeChange = true; }
@@ -1201,11 +1226,14 @@
       if (DemandedElts == 1) {
         switch (II->getIntrinsicID()) {
         default: break;
+        case Intrinsic::x86_sse_add_ss:
         case Intrinsic::x86_sse_sub_ss:
         case Intrinsic::x86_sse_mul_ss:
+        case Intrinsic::x86_sse2_add_sd:
         case Intrinsic::x86_sse2_sub_sd:
         case Intrinsic::x86_sse2_mul_sd:
-          // TODO: Lower MIN/MAX/ABS/etc
+          // TODO: Lower MIN/MAX/etc.
+          // TODO: Lower DIV (with rounding/exceptions checks).
           Value *LHS = II->getArgOperand(0);
           Value *RHS = II->getArgOperand(1);
           // Extract the element as scalars.
@@ -1216,6 +1244,11 @@
 
           switch (II->getIntrinsicID()) {
           default: llvm_unreachable("Case stmts out of sync!");
+          case Intrinsic::x86_sse_add_ss:
+          case Intrinsic::x86_sse2_add_sd:
+            TmpV = InsertNewInstWith(BinaryOperator::CreateFAdd(LHS, RHS,
+                                                        II->getName()), *II);
+            break;
           case Intrinsic::x86_sse_sub_ss:
           case Intrinsic::x86_sse2_sub_sd:
             TmpV = InsertNewInstWith(BinaryOperator::CreateFSub(LHS, RHS,
@@ -1238,6 +1271,10 @@
         }
       }
 
+      // If lowest element of a scalar op isn't used then use Arg0.
+      if (DemandedElts.getLoBits(1) != 1)
+        return II->getArgOperand(0);
+
       // Output elements are undefined if both are undefined.  Consider things
       // like undef&0.  The result is known zero, not undef.
       UndefElts &= UndefElts2;
Index: llvm/trunk/test/Transforms/InstCombine/x86-sse.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/x86-sse.ll
+++ llvm/trunk/test/Transforms/InstCombine/x86-sse.ll
@@ -5,12 +5,9 @@
 define float @test_rcp_ss_0(float %a) {
 ; CHECK-LABEL: @test_rcp_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 0
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; CHECK-NEXT:    ret float [[TMP3]]
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -23,13 +20,7 @@
 
 define float @test_rcp_ss_1(float %a) {
 ; CHECK-LABEL: @test_rcp_ss_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.sse.rcp.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 1
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    ret float 1.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -43,12 +34,9 @@
 define float @test_sqrt_ss_0(float %a) {
 ; CHECK-LABEL: @test_sqrt_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 0
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; CHECK-NEXT:    ret float [[TMP3]]
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -61,13 +49,7 @@
 
 define float @test_sqrt_ss_2(float %a) {
 ; CHECK-LABEL: @test_sqrt_ss_2(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.sse.sqrt.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 2
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    ret float 2.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -81,12 +63,9 @@
 define float @test_rsqrt_ss_0(float %a) {
 ; CHECK-LABEL: @test_rsqrt_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 0
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; CHECK-NEXT:    ret float [[TMP3]]
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -99,13 +78,7 @@
 
 define float @test_rsqrt_ss_3(float %a) {
 ; CHECK-LABEL: @test_rsqrt_ss_3(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.sse.rsqrt.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 3
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    ret float 3.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -130,14 +103,8 @@
 
 define float @test_add_ss_0(float %a, float %b) {
 ; CHECK-LABEL: @test_add_ss_0(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = tail call <4 x float> @llvm.x86.sse.add.ss(<4 x float> [[TMP4]], <4 x float> [[TMP5]])
-; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x float> [[TMP6]], i32 0
-; CHECK-NEXT:    ret float [[R]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fadd float %a, %b
+; CHECK-NEXT:    ret float [[TMP1]]
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -154,14 +121,7 @@
 
 define float @test_add_ss_1(float %a, float %b) {
 ; CHECK-LABEL: @test_add_ss_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = tail call <4 x float> @llvm.x86.sse.add.ss(<4 x float> [[TMP4]], <4 x float> [[TMP5]])
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP6]], i32 1
-; CHECK-NEXT:    ret float [[TMP7]]
+; CHECK-NEXT:    ret float 1.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -205,9 +165,7 @@
 
 define float @test_sub_ss_2(float %a, float %b) {
 ; CHECK-LABEL: @test_sub_ss_2(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.sse.sub.ss(<4 x float> <float undef, float undef, float 2.000000e+00, float undef>, <4 x float> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
-; CHECK-NEXT:    ret float [[TMP2]]
+; CHECK-NEXT:    ret float 2.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -251,9 +209,7 @@
 
 define float @test_mul_ss_3(float %a, float %b) {
 ; CHECK-LABEL: @test_mul_ss_3(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.sse.mul.ss(<4 x float> <float undef, float undef, float undef, float 3.000000e+00>, <4 x float> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[TMP1]], i32 3
-; CHECK-NEXT:    ret float [[TMP2]]
+; CHECK-NEXT:    ret float 3.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -280,12 +236,9 @@
 define float @test_div_ss_0(float %a, float %b) {
 ; CHECK-LABEL: @test_div_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = tail call <4 x float> @llvm.x86.sse.div.ss(<4 x float> [[TMP4]], <4 x float> [[TMP5]])
-; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x float> [[TMP6]], i32 0
+; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> undef, float %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call <4 x float> @llvm.x86.sse.div.ss(<4 x float> [[TMP1]], <4 x float> [[TMP2]])
+; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x float> [[TMP3]], i32 0
 ; CHECK-NEXT:    ret float [[R]]
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
@@ -303,14 +256,7 @@
 
 define float @test_div_ss_1(float %a, float %b) {
 ; CHECK-LABEL: @test_div_ss_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = tail call <4 x float> @llvm.x86.sse.div.ss(<4 x float> [[TMP4]], <4 x float> [[TMP5]])
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP6]], i32 1
-; CHECK-NEXT:    ret float [[TMP7]]
+; CHECK-NEXT:    ret float 1.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -357,9 +303,7 @@
 
 define float @test_min_ss_2(float %a, float %b) {
 ; CHECK-LABEL: @test_min_ss_2(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.sse.min.ss(<4 x float> <float undef, float undef, float 2.000000e+00, float undef>, <4 x float> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[TMP1]], i32 2
-; CHECK-NEXT:    ret float [[TMP2]]
+; CHECK-NEXT:    ret float 2.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -406,9 +350,7 @@
 
 define float @test_max_ss_3(float %a, float %b) {
 ; CHECK-LABEL: @test_max_ss_3(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <4 x float> @llvm.x86.sse.max.ss(<4 x float> <float undef, float undef, float undef, float 3.000000e+00>, <4 x float> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <4 x float> [[TMP1]], i32 3
-; CHECK-NEXT:    ret float [[TMP2]]
+; CHECK-NEXT:    ret float 3.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -435,12 +377,9 @@
 define float @test_cmp_ss_0(float %a, float %b) {
 ; CHECK-LABEL: @test_cmp_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = tail call <4 x float> @llvm.x86.sse.cmp.ss(<4 x float> [[TMP4]], <4 x float> [[TMP5]], i8 0)
-; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x float> [[TMP6]], i32 0
+; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> undef, float %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call <4 x float> @llvm.x86.sse.cmp.ss(<4 x float> [[TMP1]], <4 x float> [[TMP2]], i8 0)
+; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x float> [[TMP3]], i32 0
 ; CHECK-NEXT:    ret float [[R]]
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
@@ -458,14 +397,7 @@
 
 define float @test_cmp_ss_1(float %a, float %b) {
 ; CHECK-LABEL: @test_cmp_ss_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP6:%.*]] = tail call <4 x float> @llvm.x86.sse.cmp.ss(<4 x float> [[TMP4]], <4 x float> [[TMP5]], i8 0)
-; CHECK-NEXT:    [[TMP7:%.*]] = extractelement <4 x float> [[TMP6]], i32 1
-; CHECK-NEXT:    ret float [[TMP7]]
+; CHECK-NEXT:    ret float 1.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
Index: llvm/trunk/test/Transforms/InstCombine/x86-sse2.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/x86-sse2.ll
+++ llvm/trunk/test/Transforms/InstCombine/x86-sse2.ll
@@ -5,10 +5,9 @@
 define double @test_sqrt_sd_0(double %a) {
 ; CHECK-LABEL: @test_sqrt_sd_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = tail call <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double> [[TMP2]])
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
-; CHECK-NEXT:    ret double [[TMP4]]
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[TMP2]], i32 0
+; CHECK-NEXT:    ret double [[TMP3]]
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -19,11 +18,7 @@
 
 define double @test_sqrt_sd_1(double %a) {
 ; CHECK-LABEL: @test_sqrt_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = tail call <2 x double> @llvm.x86.sse2.sqrt.sd(<2 x double> [[TMP2]])
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
-; CHECK-NEXT:    ret double [[TMP4]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -44,12 +39,8 @@
 
 define double @test_add_sd_0(double %a, double %b) {
 ; CHECK-LABEL: @test_add_sd_0(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = tail call <2 x double> @llvm.x86.sse2.add.sd(<2 x double> [[TMP2]], <2 x double> [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 0
-; CHECK-NEXT:    ret double [[TMP5]]
+; CHECK-NEXT:    [[TMP1:%.*]] = fadd double %a, %b
+; CHECK-NEXT:    ret double [[TMP1]]
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -62,12 +53,7 @@
 
 define double @test_add_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_add_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = tail call <2 x double> @llvm.x86.sse2.add.sd(<2 x double> [[TMP2]], <2 x double> [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 1
-; CHECK-NEXT:    ret double [[TMP5]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -104,9 +90,7 @@
 
 define double @test_sub_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_sub_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.sse2.sub.sd(<2 x double> <double undef, double 1.000000e+00>, <2 x double> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[TMP1]], i32 1
-; CHECK-NEXT:    ret double [[TMP2]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -143,9 +127,7 @@
 
 define double @test_mul_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_mul_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.sse2.mul.sd(<2 x double> <double undef, double 1.000000e+00>, <2 x double> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[TMP1]], i32 1
-; CHECK-NEXT:    ret double [[TMP2]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -169,11 +151,10 @@
 define double @test_div_sd_0(double %a, double %b) {
 ; CHECK-LABEL: @test_div_sd_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = tail call <2 x double> @llvm.x86.sse2.div.sd(<2 x double> [[TMP2]], <2 x double> [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 0
-; CHECK-NEXT:    ret double [[TMP5]]
+; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> undef, double %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call <2 x double> @llvm.x86.sse2.div.sd(<2 x double> [[TMP1]], <2 x double> [[TMP2]])
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
+; CHECK-NEXT:    ret double [[TMP4]]
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -186,12 +167,7 @@
 
 define double @test_div_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_div_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = tail call <2 x double> @llvm.x86.sse2.div.sd(<2 x double> [[TMP2]], <2 x double> [[TMP3]])
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 1
-; CHECK-NEXT:    ret double [[TMP5]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -231,9 +207,7 @@
 
 define double @test_min_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_min_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.sse2.min.sd(<2 x double> <double undef, double 1.000000e+00>, <2 x double> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[TMP1]], i32 1
-; CHECK-NEXT:    ret double [[TMP2]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -273,9 +247,7 @@
 
 define double @test_max_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_max_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = tail call <2 x double> @llvm.x86.sse2.max.sd(<2 x double> <double undef, double 1.000000e+00>, <2 x double> undef)
-; CHECK-NEXT:    [[TMP2:%.*]] = extractelement <2 x double> [[TMP1]], i32 1
-; CHECK-NEXT:    ret double [[TMP2]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -299,11 +271,10 @@
 define double @test_cmp_sd_0(double %a, double %b) {
 ; CHECK-LABEL: @test_cmp_sd_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = tail call <2 x double> @llvm.x86.sse2.cmp.sd(<2 x double> [[TMP2]], <2 x double> [[TMP3]], i8 0)
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 0
-; CHECK-NEXT:    ret double [[TMP5]]
+; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> undef, double %b, i32 0
+; CHECK-NEXT:    [[TMP3:%.*]] = tail call <2 x double> @llvm.x86.sse2.cmp.sd(<2 x double> [[TMP1]], <2 x double> [[TMP2]], i8 0)
+; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
+; CHECK-NEXT:    ret double [[TMP4]]
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -316,12 +287,7 @@
 
 define double @test_cmp_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_cmp_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP4:%.*]] = tail call <2 x double> @llvm.x86.sse2.cmp.sd(<2 x double> [[TMP2]], <2 x double> [[TMP3]], i8 0)
-; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <2 x double> [[TMP4]], i32 1
-; CHECK-NEXT:    ret double [[TMP5]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
Index: llvm/trunk/test/Transforms/InstCombine/x86-sse41.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/x86-sse41.ll
+++ llvm/trunk/test/Transforms/InstCombine/x86-sse41.ll
@@ -16,7 +16,7 @@
 define double @test_round_sd_0(double %a, double %b) {
 ; CHECK-LABEL: @test_round_sd_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = tail call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> <double undef, double 1.000000e+00>, <2 x double> [[TMP1]], i32 10)
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> undef, <2 x double> [[TMP1]], i32 10)
 ; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[TMP2]], i32 0
 ; CHECK-NEXT:    ret double [[TMP3]]
 ;
@@ -31,10 +31,7 @@
 
 define double @test_round_sd_1(double %a, double %b) {
 ; CHECK-LABEL: @test_round_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %b, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = tail call <2 x double> @llvm.x86.sse41.round.sd(<2 x double> <double undef, double 1.000000e+00>, <2 x double> [[TMP1]], i32 10)
-; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[TMP2]], i32 1
-; CHECK-NEXT:    ret double [[TMP3]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -63,7 +60,7 @@
 define float @test_round_ss_0(float %a, float %b) {
 ; CHECK-LABEL: @test_round_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = tail call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> <float undef, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00>, <4 x float> [[TMP1]], i32 10)
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> undef, <4 x float> [[TMP1]], i32 10)
 ; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
 ; CHECK-NEXT:    ret float [[R]]
 ;
@@ -82,10 +79,7 @@
 
 define float @test_round_ss_2(float %a, float %b) {
 ; CHECK-LABEL: @test_round_ss_2(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %b, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = tail call <4 x float> @llvm.x86.sse41.round.ss(<4 x float> <float undef, float 1.000000e+00, float 2.000000e+00, float 3.000000e+00>, <4 x float> [[TMP1]], i32 10)
-; CHECK-NEXT:    [[R:%.*]] = extractelement <4 x float> [[TMP2]], i32 2
-; CHECK-NEXT:    ret float [[R]]
+; CHECK-NEXT:    ret float 2.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
Index: llvm/trunk/test/Transforms/InstCombine/x86-xop.ll
===================================================================
--- llvm/trunk/test/Transforms/InstCombine/x86-xop.ll
+++ llvm/trunk/test/Transforms/InstCombine/x86-xop.ll
@@ -4,10 +4,9 @@
 define double @test_vfrcz_sd_0(double %a) {
 ; CHECK-LABEL: @test_vfrcz_sd_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = tail call <2 x double> @llvm.x86.xop.vfrcz.sd(<2 x double> [[TMP2]])
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 0
-; CHECK-NEXT:    ret double [[TMP4]]
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <2 x double> @llvm.x86.xop.vfrcz.sd(<2 x double> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <2 x double> [[TMP2]], i32 0
+; CHECK-NEXT:    ret double [[TMP3]]
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -18,11 +17,7 @@
 
 define double @test_vfrcz_sd_1(double %a) {
 ; CHECK-LABEL: @test_vfrcz_sd_1(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> undef, double %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <2 x double> [[TMP1]], double 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = tail call <2 x double> @llvm.x86.xop.vfrcz.sd(<2 x double> [[TMP2]])
-; CHECK-NEXT:    [[TMP4:%.*]] = extractelement <2 x double> [[TMP3]], i32 1
-; CHECK-NEXT:    ret double [[TMP4]]
+; CHECK-NEXT:    ret double 1.000000e+00
 ;
   %1 = insertelement <2 x double> undef, double %a, i32 0
   %2 = insertelement <2 x double> %1, double 1.000000e+00, i32 1
@@ -34,12 +29,9 @@
 define float @test_vfrcz_ss_0(float %a) {
 ; CHECK-LABEL: @test_vfrcz_ss_0(
 ; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.xop.vfrcz.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 0
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    [[TMP2:%.*]] = tail call <4 x float> @llvm.x86.xop.vfrcz.ss(<4 x float> [[TMP1]])
+; CHECK-NEXT:    [[TMP3:%.*]] = extractelement <4 x float> [[TMP2]], i32 0
+; CHECK-NEXT:    ret float [[TMP3]]
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1
@@ -52,13 +44,7 @@
 
 define float @test_vfrcz_ss_3(float %a) {
 ; CHECK-LABEL: @test_vfrcz_ss_3(
-; CHECK-NEXT:    [[TMP1:%.*]] = insertelement <4 x float> undef, float %a, i32 0
-; CHECK-NEXT:    [[TMP2:%.*]] = insertelement <4 x float> [[TMP1]], float 1.000000e+00, i32 1
-; CHECK-NEXT:    [[TMP3:%.*]] = insertelement <4 x float> [[TMP2]], float 2.000000e+00, i32 2
-; CHECK-NEXT:    [[TMP4:%.*]] = insertelement <4 x float> [[TMP3]], float 3.000000e+00, i32 3
-; CHECK-NEXT:    [[TMP5:%.*]] = tail call <4 x float> @llvm.x86.xop.vfrcz.ss(<4 x float> [[TMP4]])
-; CHECK-NEXT:    [[TMP6:%.*]] = extractelement <4 x float> [[TMP5]], i32 3
-; CHECK-NEXT:    ret float [[TMP6]]
+; CHECK-NEXT:    ret float 3.000000e+00
 ;
   %1 = insertelement <4 x float> undef, float %a, i32 0
   %2 = insertelement <4 x float> %1, float 1.000000e+00, i32 1