Index: lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- lib/Target/X86/X86ISelLowering.cpp
+++ lib/Target/X86/X86ISelLowering.cpp
@@ -16740,10 +16740,8 @@
     // FIXME: The 'haddpd' instruction may be slower than 'movhlps + addsd'.
     Result = DAG.getNode(X86ISD::FHADD, dl, MVT::v2f64, Sub, Sub);
   } else {
-    SDValue S2F = DAG.getBitcast(MVT::v4i32, Sub);
-    SDValue Shuffle = DAG.getVectorShuffle(MVT::v4i32, dl, S2F, S2F, {2,3,0,1});
-    Result = DAG.getNode(ISD::FADD, dl, MVT::v2f64,
-                         DAG.getBitcast(MVT::v2f64, Shuffle), Sub);
+    SDValue Shuffle = DAG.getVectorShuffle(MVT::v2f64, dl, Sub, Sub, {1,-1});
+    Result = DAG.getNode(ISD::FADD, dl, MVT::v2f64, Shuffle, Sub);
   }
 
   return DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::f64, Result,
Index: test/CodeGen/X86/ftrunc.ll
===================================================================
--- test/CodeGen/X86/ftrunc.ll
+++ test/CodeGen/X86/ftrunc.ll
@@ -41,7 +41,8 @@
 ; SSE2-NEXT:    movq %rax, %xmm1
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
 ; SSE2-NEXT:    subpd {{.*}}(%rip), %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE2-NEXT:    movapd %xmm1, %xmm0
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE2-NEXT:    addpd %xmm1, %xmm0
 ; SSE2-NEXT:    retq
 ;
@@ -128,12 +129,14 @@
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    movapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
 ; SSE2-NEXT:    subpd %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE2-NEXT:    movapd %xmm1, %xmm0
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE2-NEXT:    addpd %xmm1, %xmm0
 ; SSE2-NEXT:    movq %rdx, %xmm1
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    subpd %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[2,3,0,1]
+; SSE2-NEXT:    movapd %xmm1, %xmm2
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm1[1]
 ; SSE2-NEXT:    addpd %xmm1, %xmm2
 ; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
 ; SSE2-NEXT:    retq
@@ -194,23 +197,27 @@
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    movapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
 ; SSE2-NEXT:    subpd %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE2-NEXT:    movapd %xmm1, %xmm0
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE2-NEXT:    addpd %xmm1, %xmm0
 ; SSE2-NEXT:    movq %rdi, %xmm1
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE2-NEXT:    subpd %xmm3, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[2,3,0,1]
+; SSE2-NEXT:    movapd %xmm1, %xmm4
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm4 = xmm4[1],xmm1[1]
 ; SSE2-NEXT:    addpd %xmm1, %xmm4
 ; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm4[0]
 ; SSE2-NEXT:    movq %rcx, %xmm4
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
 ; SSE2-NEXT:    subpd %xmm3, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[2,3,0,1]
+; SSE2-NEXT:    movapd %xmm4, %xmm1
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm4[1]
 ; SSE2-NEXT:    addpd %xmm4, %xmm1
 ; SSE2-NEXT:    movq %rax, %xmm4
 ; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
 ; SSE2-NEXT:    subpd %xmm3, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[2,3,0,1]
+; SSE2-NEXT:    movapd %xmm4, %xmm2
+; SSE2-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm4[1]
 ; SSE2-NEXT:    addpd %xmm4, %xmm2
 ; SSE2-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; SSE2-NEXT:    retq
@@ -356,7 +363,7 @@
   ret <4 x double> %r
 }
 
-; The fold may be guarded to allow existing code to continue 
+; The fold may be guarded to allow existing code to continue
 ; working based on its assumptions of float->int overflow.
 
 define float @trunc_unsigned_f32_disable_via_attr(float %x) #1 {
Index: test/CodeGen/X86/scalar-int-to-fp.ll
===================================================================
--- test/CodeGen/X86/scalar-int-to-fp.ll
+++ test/CodeGen/X86/scalar-int-to-fp.ll
@@ -636,7 +636,8 @@
 ; SSE2_32-NEXT:    movsd {{.*#+}} xmm0 = mem[0],zero
 ; SSE2_32-NEXT:    unpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
 ; SSE2_32-NEXT:    subpd {{\.LCPI.*}}, %xmm0
-; SSE2_32-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
+; SSE2_32-NEXT:    movapd %xmm0, %xmm1
+; SSE2_32-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
 ; SSE2_32-NEXT:    addpd %xmm0, %xmm1
 ; SSE2_32-NEXT:    movlpd %xmm1, (%esp)
 ; SSE2_32-NEXT:    fldl (%esp)
@@ -649,7 +650,8 @@
 ; SSE2_64-NEXT:    movq %rdi, %xmm1
 ; SSE2_64-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],mem[0],xmm1[1],mem[1]
 ; SSE2_64-NEXT:    subpd {{.*}}(%rip), %xmm1
-; SSE2_64-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE2_64-NEXT:    movapd %xmm1, %xmm0
+; SSE2_64-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE2_64-NEXT:    addpd %xmm1, %xmm0
 ; SSE2_64-NEXT:    retq
 ;
Index: test/CodeGen/X86/vec_int_to_fp.ll
===================================================================
--- test/CodeGen/X86/vec_int_to_fp.ll
+++ test/CodeGen/X86/vec_int_to_fp.ll
@@ -412,18 +412,21 @@
 define <2 x double> @uitofp_2i64_to_2f64(<2 x i64> %a) {
 ; SSE-LABEL: uitofp_2i64_to_2f64:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm1 = [1127219200,1160773632,0,0]
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE-NEXT:    movapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
-; SSE-NEXT:    subpd %xmm3, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
-; SSE-NEXT:    addpd %xmm4, %xmm0
-; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE-NEXT:    subpd %xmm3, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,3,0,1]
-; SSE-NEXT:    addpd %xmm2, %xmm1
-; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [1127219200,1160773632,0,0]
+; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
+; SSE-NEXT:    movapd {{.*#+}} xmm4 = [4.503600e+15,1.934281e+25]
+; SSE-NEXT:    subpd %xmm4, %xmm0
+; SSE-NEXT:    movapd %xmm0, %xmm1
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
+; SSE-NEXT:    addpd %xmm0, %xmm1
+; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
+; SSE-NEXT:    subpd %xmm4, %xmm3
+; SSE-NEXT:    movapd %xmm3, %xmm0
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm3[1]
+; SSE-NEXT:    addpd %xmm3, %xmm0
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm0[0]
+; SSE-NEXT:    movapd %xmm1, %xmm0
 ; SSE-NEXT:    retq
 ;
 ; VEX-LABEL: uitofp_2i64_to_2f64:
@@ -691,28 +694,34 @@
 define <4 x double> @uitofp_4i64_to_4f64(<4 x i64> %a) {
 ; SSE-LABEL: uitofp_4i64_to_4f64:
 ; SSE:       # %bb.0:
-; SSE-NEXT:    movdqa {{.*#+}} xmm2 = [1127219200,1160773632,0,0]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm0[2,3,0,1]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE-NEXT:    movapd {{.*#+}} xmm4 = [4.503600e+15,1.934281e+25]
-; SSE-NEXT:    subpd %xmm4, %xmm0
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,3,0,1]
-; SSE-NEXT:    addpd %xmm5, %xmm0
-; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
-; SSE-NEXT:    subpd %xmm4, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm3[2,3,0,1]
-; SSE-NEXT:    addpd %xmm3, %xmm5
-; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm5[0]
-; SSE-NEXT:    pshufd {{.*#+}} xmm3 = xmm1[2,3,0,1]
-; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
-; SSE-NEXT:    subpd %xmm4, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm5 = xmm1[2,3,0,1]
-; SSE-NEXT:    addpd %xmm5, %xmm1
-; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
-; SSE-NEXT:    subpd %xmm4, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm3[2,3,0,1]
-; SSE-NEXT:    addpd %xmm3, %xmm2
-; SSE-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; SSE-NEXT:    movdqa %xmm0, %xmm2
+; SSE-NEXT:    movdqa {{.*#+}} xmm3 = [1127219200,1160773632,0,0]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
+; SSE-NEXT:    movapd {{.*#+}} xmm5 = [4.503600e+15,1.934281e+25]
+; SSE-NEXT:    subpd %xmm5, %xmm2
+; SSE-NEXT:    movapd %xmm2, %xmm0
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm2[1]
+; SSE-NEXT:    addpd %xmm2, %xmm0
+; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; SSE-NEXT:    subpd %xmm5, %xmm4
+; SSE-NEXT:    movapd %xmm4, %xmm2
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm4[1]
+; SSE-NEXT:    addpd %xmm4, %xmm2
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm2[0]
+; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm1[2,3,0,1]
+; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
+; SSE-NEXT:    subpd %xmm5, %xmm1
+; SSE-NEXT:    movapd %xmm1, %xmm2
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm1[1]
+; SSE-NEXT:    addpd %xmm1, %xmm2
+; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
+; SSE-NEXT:    subpd %xmm5, %xmm4
+; SSE-NEXT:    movapd %xmm4, %xmm1
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm4[1]
+; SSE-NEXT:    addpd %xmm4, %xmm1
+; SSE-NEXT:    unpcklpd {{.*#+}} xmm2 = xmm2[0],xmm1[0]
+; SSE-NEXT:    movapd %xmm2, %xmm1
 ; SSE-NEXT:    retq
 ;
 ; VEX-LABEL: uitofp_4i64_to_4f64:
@@ -2797,11 +2806,13 @@
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm2[0],xmm1[1],xmm2[1]
 ; SSE-NEXT:    movapd {{.*#+}} xmm4 = [4.503600e+15,1.934281e+25]
 ; SSE-NEXT:    subpd %xmm4, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE-NEXT:    movapd %xmm1, %xmm0
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE-NEXT:    addpd %xmm1, %xmm0
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm3 = xmm3[0],xmm2[0],xmm3[1],xmm2[1]
 ; SSE-NEXT:    subpd %xmm4, %xmm3
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm3[2,3,0,1]
+; SSE-NEXT:    movapd %xmm3, %xmm1
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm3[1]
 ; SSE-NEXT:    addpd %xmm3, %xmm1
 ; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; SSE-NEXT:    retq
@@ -2963,21 +2974,25 @@
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
 ; SSE-NEXT:    movapd {{.*#+}} xmm5 = [4.503600e+15,1.934281e+25]
 ; SSE-NEXT:    subpd %xmm5, %xmm1
-; SSE-NEXT:    pshufd {{.*#+}} xmm0 = xmm1[2,3,0,1]
+; SSE-NEXT:    movapd %xmm1, %xmm0
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
 ; SSE-NEXT:    addpd %xmm1, %xmm0
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
 ; SSE-NEXT:    subpd %xmm5, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[2,3,0,1]
+; SSE-NEXT:    movapd %xmm4, %xmm1
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm4[1]
 ; SSE-NEXT:    addpd %xmm4, %xmm1
 ; SSE-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
 ; SSE-NEXT:    pshufd {{.*#+}} xmm4 = xmm2[2,3,0,1]
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
 ; SSE-NEXT:    subpd %xmm5, %xmm2
-; SSE-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,3,0,1]
+; SSE-NEXT:    movapd %xmm2, %xmm1
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm2[1]
 ; SSE-NEXT:    addpd %xmm2, %xmm1
 ; SSE-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm3[0],xmm4[1],xmm3[1]
 ; SSE-NEXT:    subpd %xmm5, %xmm4
-; SSE-NEXT:    pshufd {{.*#+}} xmm2 = xmm4[2,3,0,1]
+; SSE-NEXT:    movapd %xmm4, %xmm2
+; SSE-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm4[1]
 ; SSE-NEXT:    addpd %xmm4, %xmm2
 ; SSE-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
 ; SSE-NEXT:    retq