diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index be3ecd7e531b..94faae5b556b 100644
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -12589,6 +12589,29 @@ static SDValue PerformSTORECombine(SDNode *N, SelectionDAG &DAG,
   EVT VT = St->getValue().getValueType();
   EVT StVT = St->getMemoryVT();
   DebugLoc dl = St->getDebugLoc();
+  SDValue StoredVal = St->getOperand(1);
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+  // If we are saving a concatination of two XMM registers, perform two stores.
+  if (VT.getSizeInBits() == 256 &&
+    StoredVal.getNode()->getOpcode() == ISD::CONCAT_VECTORS &&
+    StoredVal.getNumOperands() == 2) {
+
+    SDValue Value0 = StoredVal.getOperand(0);
+    SDValue Value1 = StoredVal.getOperand(1);
+
+    SDValue Stride = DAG.getConstant(16, TLI.getPointerTy());
+    SDValue Ptr0 = St->getBasePtr();
+    SDValue Ptr1 = DAG.getNode(ISD::ADD, dl, Ptr0.getValueType(), Ptr0, Stride);
+
+    SDValue Ch0 = DAG.getStore(St->getChain(), dl, Value0, Ptr0,
+                                St->getPointerInfo(), St->isVolatile(),
+                                St->isNonTemporal(), St->getAlignment());
+    SDValue Ch1 = DAG.getStore(St->getChain(), dl, Value1, Ptr1,
+                                St->getPointerInfo(), St->isVolatile(),
+                                St->isNonTemporal(), St->getAlignment());
+    return DAG.getNode(ISD::TokenFactor, dl, MVT::Other, Ch0, Ch1);
+  }
 
   // Optimize trunc store (of multiple scalars) to shuffle and store.
   // First, pack all of the elements in one place. Next, store to memory
diff --git a/llvm/test/CodeGen/X86/avx-insert.ll b/llvm/test/CodeGen/X86/avx-insert.ll
new file mode 100644
index 000000000000..d9eae03eee75
--- /dev/null
+++ b/llvm/test/CodeGen/X86/avx-insert.ll
@@ -0,0 +1,17 @@
+; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7-avx -mattr=+avx | FileCheck %s
+
+; It is faster to make two saves, if the data is already in XMM registers. For
+; example, after making an integer operation.
+define void @double_save(<4 x i32>* %Ap, <4 x i32>* %Bp, <8 x i32>* %P) nounwind ssp {
+entry:
+  ; CHECK: movaps
+  ; CHECK: movaps
+  ; CHECK: movaps
+  ; CHECK: movaps
+  %A = load <4 x i32>* %Ap
+  %B = load <4 x i32>* %Bp
+  %Z = shufflevector <4 x i32>%A, <4 x i32>%B, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+  store <8 x i32> %Z, <8 x i32>* %P, align 16
+  ret void
+}
+