Index: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
@@ -32071,6 +32071,14 @@
     if (Src.getOpcode() == ISD::SCALAR_TO_VECTOR)
       return DAG.getNode(X86ISD::VBROADCAST, DL, VT, Src.getOperand(0));
 
+    // Share broadcast with the longest vector and extract low subvector (free).
+    for (SDNode *User : Src->uses())
+      if (User != N.getNode() && User->getOpcode() == X86ISD::VBROADCAST &&
+          User->getValueSizeInBits(0) > VT.getSizeInBits()) {
+        return extractSubVector(SDValue(User, 0), 0, DAG, DL,
+                                VT.getSizeInBits());
+      }
+
     return SDValue();
   }
   case X86ISD::PSHUFD:
Index: llvm/trunk/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll
+++ llvm/trunk/test/CodeGen/X86/broadcast-elm-cross-splat-vec.ll
@@ -357,17 +357,15 @@
 define <64 x i8> @f64i8_i32(<64 x i8> %a) {
 ; AVX-LABEL: f64i8_i32:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; AVX-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddb %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-NEXT:    vbroadcastss {{.*#+}} ymm2 = [3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37]
+; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-NEXT:    retl
@@ -390,17 +388,15 @@
 ;
 ; AVX-64-LABEL: f64i8_i32:
 ; AVX-64:       # %bb.0:
-; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-64-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; AVX-64-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm0, %xmm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-64-NEXT:    vbroadcastss {{.*#+}} ymm2 = [3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37,3.82047143E-37]
+; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-64-NEXT:    retq
@@ -426,21 +422,18 @@
 }
 
 
-; FIXME the load should be folded with the MOVDDUP with AVX1. PR39454
 define <64 x i8> @f64xi8_i64(<64 x i8> %a) {
 ; AVX-LABEL: f64xi8_i64:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; AVX-NEXT:    vmovddup {{.*#+}} xmm3 = xmm3[0,0]
-; AVX-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddb %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [7.9499288951273625E-275,7.9499288951273625E-275,7.9499288951273625E-275,7.9499288951273625E-275]
+; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-NEXT:    retl
@@ -463,17 +456,15 @@
 ;
 ; AVX-64-LABEL: f64xi8_i64:
 ; AVX-64:       # %bb.0:
-; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-64-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; AVX-64-NEXT:    vmovddup {{.*#+}} xmm3 = xmm3[0,0]
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm1, %xmm1
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddb %xmm3, %xmm0, %xmm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-64-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [7.9499288951273625E-275,7.9499288951273625E-275,7.9499288951273625E-275,7.9499288951273625E-275]
+; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm1, %xmm1
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-64-NEXT:    retq
@@ -498,6 +489,7 @@
   ret <64 x i8> %res2
 }
 
+
 define <64 x i8> @f64xi8_i128(<64 x i8> %a) {
 ; AVX-LABEL: f64xi8_i128:
 ; AVX:       # %bb.0:
@@ -843,17 +835,15 @@
 define <32 x i16> @f32xi16_i32(<32 x i16> %a) {
 ; AVX-LABEL: f32xi16_i32:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; AVX-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; AVX-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddw %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-NEXT:    vbroadcastss {{.*#+}} ymm2 = [9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41]
+; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-NEXT:    retl
@@ -876,17 +866,15 @@
 ;
 ; AVX-64-LABEL: f32xi16_i32:
 ; AVX-64:       # %bb.0:
-; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-64-NEXT:    vmovd {{.*#+}} xmm3 = mem[0],zero,zero,zero
-; AVX-64-NEXT:    vpshufd {{.*#+}} xmm3 = xmm3[0,0,0,0]
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm0, %xmm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-64-NEXT:    vbroadcastss {{.*#+}} ymm2 = [9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41,9.18354962E-41]
+; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm0, %xmm0
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-64-NEXT:    retq
@@ -912,21 +900,18 @@
 }
 
 
-; FIXME the load should be folded with the MOVDDUP with AVX1. PR39454
 define <32 x i16> @f32xi16_i64(<32 x i16> %a) {
 ; AVX-LABEL: f32xi16_i64:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; AVX-NEXT:    vmovddup {{.*#+}} xmm3 = xmm3[0,0]
-; AVX-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddw %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4.1720559249406128E-309,4.1720559249406128E-309,4.1720559249406128E-309,4.1720559249406128E-309]
+; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddw %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-NEXT:    retl
@@ -949,17 +934,15 @@
 ;
 ; AVX-64-LABEL: f32xi16_i64:
 ; AVX-64:       # %bb.0:
-; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-64-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; AVX-64-NEXT:    vmovddup {{.*#+}} xmm3 = xmm3[0,0]
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm1, %xmm1
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddw %xmm3, %xmm0, %xmm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-64-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [4.1720559249406128E-309,4.1720559249406128E-309,4.1720559249406128E-309,4.1720559249406128E-309]
+; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm1, %xmm1
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddw %xmm2, %xmm0, %xmm0
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-64-NEXT:    retq
@@ -1133,7 +1116,6 @@
 }
 
 
-
 define <4 x i32> @f4xi32_i64(<4 x i32> %a) {
 ; AVX-LABEL: f4xi32_i64:
 ; AVX:       # %bb.0:
@@ -1254,21 +1236,18 @@
 }
 
 
-; FIXME the load should be folded with the MOVDDUP with AVX1. PR39454
 define <16 x i32> @f16xi32_i64(<16 x i32> %a) {
 ; AVX-LABEL: f16xi32_i64:
 ; AVX:       # %bb.0:
-; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; AVX-NEXT:    vmovddup {{.*#+}} xmm3 = xmm3[0,0]
-; AVX-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vpaddd %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [2.1219957909652723E-314,2.1219957909652723E-314,2.1219957909652723E-314,2.1219957909652723E-314]
+; AVX-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-NEXT:    vpaddd %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-NEXT:    vpaddd %xmm2, %xmm3, %xmm3
+; AVX-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
+; AVX-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-NEXT:    retl
@@ -1291,17 +1270,15 @@
 ;
 ; AVX-64-LABEL: f16xi32_i64:
 ; AVX-64:       # %bb.0:
-; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm2
-; AVX-64-NEXT:    vmovsd {{.*#+}} xmm3 = mem[0],zero
-; AVX-64-NEXT:    vmovddup {{.*#+}} xmm3 = xmm3[0,0]
-; AVX-64-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddd %xmm3, %xmm1, %xmm1
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm1, %ymm1
-; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm2
-; AVX-64-NEXT:    vpaddd %xmm3, %xmm2, %xmm2
-; AVX-64-NEXT:    vpaddd %xmm3, %xmm0, %xmm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
-; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm3, %ymm2
+; AVX-64-NEXT:    vbroadcastsd {{.*#+}} ymm2 = [2.1219957909652723E-314,2.1219957909652723E-314,2.1219957909652723E-314,2.1219957909652723E-314]
+; AVX-64-NEXT:    vextractf128 $1, %ymm1, %xmm3
+; AVX-64-NEXT:    vpaddd %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddd %xmm2, %xmm1, %xmm1
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm1, %ymm1
+; AVX-64-NEXT:    vextractf128 $1, %ymm0, %xmm3
+; AVX-64-NEXT:    vpaddd %xmm2, %xmm3, %xmm3
+; AVX-64-NEXT:    vpaddd %xmm2, %xmm0, %xmm0
+; AVX-64-NEXT:    vinsertf128 $1, %xmm3, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm0, %ymm0
 ; AVX-64-NEXT:    vandps %ymm2, %ymm1, %ymm1
 ; AVX-64-NEXT:    retq
@@ -2045,7 +2022,6 @@
 }
 
 
-
 define <8 x i16> @f8xi16_i32_NaN(<8 x i16> %a) {
 ; AVX-LABEL: f8xi16_i32_NaN:
 ; AVX:       # %bb.0:
Index: llvm/trunk/test/CodeGen/X86/oddshuffles.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/oddshuffles.ll
+++ llvm/trunk/test/CodeGen/X86/oddshuffles.ll
@@ -1743,10 +1743,10 @@
 ;
 ; AVX2-LABEL: wrongorder:
 ; AVX2:       # %bb.0:
-; AVX2-NEXT:    vbroadcastsd %xmm0, %ymm1
-; AVX2-NEXT:    vmovaps %ymm1, 32(%rdi)
-; AVX2-NEXT:    vmovaps %ymm1, (%rdi)
-; AVX2-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
+; AVX2-NEXT:    vbroadcastsd %xmm0, %ymm0
+; AVX2-NEXT:    vmovaps %ymm0, 32(%rdi)
+; AVX2-NEXT:    vmovaps %ymm0, (%rdi)
+; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
 ;