Index: llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
+++ llvm/trunk/lib/Target/X86/X86ISelLowering.cpp
@@ -11542,6 +11542,23 @@
   if (!canWidenShuffleElements(Mask, WidenedMask))
     return SDValue();
 
+  SDValue Ops[2] = {DAG.getUNDEF(VT), DAG.getUNDEF(VT)};
+  // Insure elements came from the same Op.
+  int MaxOp1Index = VT.getVectorNumElements()/2 - 1;
+  for (int i = 0, Size = WidenedMask.size(); i < Size; ++i) {
+    if (WidenedMask[i] == SM_SentinelZero)
+      return SDValue();
+    if (WidenedMask[i] == SM_SentinelUndef)
+      continue;
+
+    SDValue Op = WidenedMask[i] > MaxOp1Index ? V2 : V1;
+    unsigned OpIndex = (i < Size/2) ? 0 : 1;
+    if (Ops[OpIndex].isUndef())
+      Ops[OpIndex] = Op;
+    else if (Ops[OpIndex] != Op)
+      return SDValue();
+  }
+
   // Form a 128-bit permutation.
   // Convert the 64-bit shuffle mask selection values into 128-bit selection
   // bits defined by a vshuf64x2 instruction's immediate control byte.
@@ -11549,15 +11566,12 @@
   unsigned ControlBitsNum = WidenedMask.size() / 2;
 
   for (int i = 0, Size = WidenedMask.size(); i < Size; ++i) {
-    if (WidenedMask[i] == SM_SentinelZero)
-      return SDValue();
-
     // Use first element in place of undef mask.
     Imm = (WidenedMask[i] == SM_SentinelUndef) ? 0 : WidenedMask[i];
     PermMask |= (Imm % WidenedMask.size()) << (i * ControlBitsNum);
   }
 
-  return DAG.getNode(X86ISD::SHUF128, DL, VT, V1, V2,
+  return DAG.getNode(X86ISD::SHUF128, DL, VT, Ops[0], Ops[1],
                      DAG.getConstant(PermMask, DL, MVT::i8));
 }
 
Index: llvm/trunk/test/CodeGen/X86/vector-shuffle-512-v8.ll
===================================================================
--- llvm/trunk/test/CodeGen/X86/vector-shuffle-512-v8.ll
+++ llvm/trunk/test/CodeGen/X86/vector-shuffle-512-v8.ll
@@ -2271,3 +2271,35 @@
   %res = shufflevector <16 x float> %x, <16 x float> %x1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 20, i32 21, i32 22, i32 23, i32 16, i32 17, i32 18, i32 19>
   ret <16 x float> %res
 }
+
+define <8 x double> @shuffle_v8f64_23014567(<8 x double> %a0, <8 x double> %a1) {
+; ALL-LABEL: shuffle_v8f64_23014567:
+; ALL:       # BB#0:
+; ALL-NEXT:    vshuff64x2 $225, %zmm1, %zmm1, %zmm0 # zmm0 = zmm1[2,3,0,1,4,5,6,7]
+  %1 = shufflevector <8 x double> %a1, <8 x double> undef, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 4, i32 5, i32 6, i32 7>
+  ret <8 x double> %1
+}
+
+define <8 x double> @shuffle_v8f64_2301uu67(<8 x double> %a0, <8 x double> %a1) {
+; ALL-LABEL: shuffle_v8f64_2301uu67:
+; ALL:       # BB#0:
+; ALL-NEXT:    vshuff64x2 $193, %zmm1, %zmm1, %zmm0 # zmm0 = zmm1[2,3,0,1,0,1,6,7]
+  %1 = shufflevector <8 x double> %a1, <8 x double> undef, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 undef, i32 undef, i32 6, i32 7>
+  ret <8 x double> %1
+}
+
+define <8 x double> @shuffle_v8f64_2301uuuu(<8 x double> %a0, <8 x double> %a1) {
+; ALL-LABEL: shuffle_v8f64_2301uuuu:
+; ALL:       # BB#0:
+; ALL-NEXT:    vshuff64x2 $1, %zmm0, %zmm1, %zmm0 # zmm0 = zmm1[2,3,0,1],zmm0[0,1,0,1]
+  %1 = shufflevector <8 x double> %a1, <8 x double> undef, <8 x i32> <i32 2, i32 3, i32 0, i32 1, i32 undef, i32 undef, i32 undef, i32 undef>
+  ret <8 x double> %1
+}
+
+define <8 x double> @shuffle_v8f64_uuu2301(<8 x double> %a0, <8 x double> %a1) {
+; ALL-LABEL: shuffle_v8f64_uuu2301:
+; ALL:       # BB#0:
+; ALL-NEXT:    vshuff64x2 $16, %zmm1, %zmm0, %zmm0 # zmm0 = zmm0[0,1,0,1],zmm1[2,3,0,1]
+  %1 = shufflevector <8 x double> %a1, <8 x double> undef, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 2, i32 3, i32 0, i32 1>
+  ret <8 x double> %1
+}