diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index f4329d3bf9d5..2d84860610c9 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -4977,6 +4977,18 @@ static SDValue EltsFromConsecutiveLoads(EVT VT, SmallVectorImpl<SDValue> &Elts,
                                 LDBase->getAlignment(),
                                 false/*isVolatile*/, true/*ReadMem*/,
                                 false/*WriteMem*/);
+
+    // Make sure the newly-created LOAD is in the same position as LDBase in
+    // terms of dependency. We create a TokenFactor for LDBase and ResNode, and
+    // update uses of LDBase's output chain to use the TokenFactor.
+    if (LDBase->hasAnyUseOfValue(1)) {
+      SDValue NewChain = DAG.getNode(ISD::TokenFactor, DL, MVT::Other,
+                             SDValue(LDBase, 1), SDValue(ResNode.getNode(), 1));
+      DAG.ReplaceAllUsesOfValueWith(SDValue(LDBase, 1), NewChain);
+      DAG.UpdateNodeOperands(NewChain.getNode(), SDValue(LDBase, 1),
+                             SDValue(ResNode.getNode(), 1));
+    }
+
     return DAG.getNode(ISD::BITCAST, DL, VT, ResNode);
   }
   return SDValue();
diff --git a/llvm/test/CodeGen/X86/vec_shuffle-26.ll b/llvm/test/CodeGen/X86/vec_shuffle-26.ll
index 086af6bb114b..93c553001f81 100644
--- a/llvm/test/CodeGen/X86/vec_shuffle-26.ll
+++ b/llvm/test/CodeGen/X86/vec_shuffle-26.ll
@@ -1,6 +1,4 @@
-; RUN: llc < %s -march=x86 -mattr=sse41 -o %t
-; RUN: grep unpcklps %t | count 1
-; RUN: grep unpckhps %t | count 3
+; RUN: llc < %s -march=x86 -mattr=sse41 | FileCheck %s
 
 ; Transpose example using the more generic vector shuffle. Return float8
 ; instead of float16
@@ -14,6 +12,11 @@ target triple = "i386-apple-cl.1.0"
 
 define <8 x float> @__transpose2(<4 x float> %p0, <4 x float> %p1, <4 x float> %p2, <4 x float> %p3) nounwind {
 entry:
+; CHECK: transpose2
+; CHECK: unpckhps
+; CHECK: unpckhps
+; CHECK: unpcklps
+; CHECK: unpckhps
 	%unpcklps = shufflevector <4 x float> %p0, <4 x float> %p2, <4 x i32> < i32 0, i32 4, i32 1, i32 5 >		; <<4 x float>> [#uses=2]
 	%unpckhps = shufflevector <4 x float> %p0, <4 x float> %p2, <4 x i32> < i32 2, i32 6, i32 3, i32 7 >		; <<4 x float>> [#uses=2]
 	%unpcklps8 = shufflevector <4 x float> %p1, <4 x float> %p3, <4 x i32> < i32 0, i32 4, i32 1, i32 5 >		; <<4 x float>> [#uses=2]
@@ -27,3 +30,28 @@ entry:
 ;       %r3 = shufflevector <8 x float> %r1,  <8 x float> %r2,  <16 x i32> < i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15 >; 
 	ret <8 x float> %r2
 }
+
+define <2 x i64> @lo_hi_shift(float* nocapture %x, float* nocapture %y) nounwind {
+entry:
+; movhps should happen before extractps to assure it gets the correct value.
+; CHECK: lo_hi_shift
+; CHECK: movhps ([[BASEREG:%[a-z]+]]),
+; CHECK: extractps ${{[0-9]+}}, %xmm{{[0-9]+}}, {{[0-9]*}}([[BASEREG]])
+; CHECK: extractps ${{[0-9]+}}, %xmm{{[0-9]+}}, {{[0-9]*}}([[BASEREG]])
+  %v.i = bitcast float* %y to <4 x float>*
+  %0 = load <4 x float>* %v.i, align 1
+  %1 = bitcast float* %x to <1 x i64>*
+  %.val = load <1 x i64>* %1, align 1
+  %2 = bitcast <1 x i64> %.val to <2 x float>
+  %shuffle.i = shufflevector <2 x float> %2, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 undef, i32 undef>
+  %shuffle1.i = shufflevector <4 x float> %0, <4 x float> %shuffle.i, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
+  %cast.i = bitcast <4 x float> %0 to <2 x i64>
+  %extract.i = extractelement <2 x i64> %cast.i, i32 1
+  %3 = bitcast float* %x to i64*
+  store i64 %extract.i, i64* %3, align 4
+  %4 = bitcast <4 x float> %0 to <16 x i8>
+  %5 = bitcast <4 x float> %shuffle1.i to <16 x i8>
+  %palignr = shufflevector <16 x i8> %5, <16 x i8> %4, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
+  %6 = bitcast <16 x i8> %palignr to <2 x i64>
+  ret <2 x i64> %6
+}