diff --git a/lib/Target/X86/X86InstrSSE.td b/lib/Target/X86/X86InstrSSE.td
index 959ae36fc59..1276bda524a 100644
--- a/lib/Target/X86/X86InstrSSE.td
+++ b/lib/Target/X86/X86InstrSSE.td
@@ -1019,6 +1019,46 @@ let Predicates = [HasAVX] in {
             (VMOVUPSYmr addr:$dst, VR256:$src)>;
   def : Pat<(store (v32i8 VR256:$src), addr:$dst),
             (VMOVUPSYmr addr:$dst, VR256:$src)>;
+
+  // Special patterns for storing subvector extracts of lower 128-bits
+  // Its cheaper to just use VMOVAPS/VMOVUPS instead of VEXTRACTF128mr
+  def : Pat<(alignedstore (v2f64 (extract_subvector
+                                  (v4f64 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVAPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(alignedstore (v4f32 (extract_subvector
+                                  (v8f32 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVAPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(alignedstore (v2i64 (extract_subvector
+                                  (v4i64 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVAPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(alignedstore (v4i32 (extract_subvector
+                                  (v8i32 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVAPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(alignedstore (v8i16 (extract_subvector
+                                  (v16i16 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(alignedstore (v16i8 (extract_subvector
+                                  (v32i8 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVAPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+
+  def : Pat<(store (v2f64 (extract_subvector
+                           (v4f64 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVUPDmr addr:$dst, (v2f64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(store (v4f32 (extract_subvector
+                           (v8f32 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVUPSmr addr:$dst, (v4f32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(store (v2i64 (extract_subvector
+                           (v4i64 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVUPDmr addr:$dst, (v2i64 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(store (v4i32 (extract_subvector
+                           (v8i32 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVUPSmr addr:$dst, (v4i32 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(store (v8i16 (extract_subvector
+                           (v16i16 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVAPSmr addr:$dst, (v8i16 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
+  def : Pat<(store (v16i8 (extract_subvector
+                           (v32i8 VR256:$src), (i32 0))), addr:$dst),
+            (VMOVUPSmr addr:$dst, (v16i8 (EXTRACT_SUBREG VR256:$src,sub_xmm)))>;
 }
 
 // Use movaps / movups for SSE integer load / store (one byte shorter).
diff --git a/test/CodeGen/X86/avx-vextractf128.ll b/test/CodeGen/X86/avx-vextractf128.ll
index a900eeb0609..ff56a454996 100644
--- a/test/CodeGen/X86/avx-vextractf128.ll
+++ b/test/CodeGen/X86/avx-vextractf128.ll
@@ -19,12 +19,12 @@ entry:
 }
 
 ; CHECK: @t0
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
+; CHECK-NOT: vextractf128 $1, %ymm0, %xmm0
 ; CHECK-NOT: vmovaps %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
+; CHECK: vextractf128 $1, %ymm0, (%rdi)
 define void @t0(float* nocapture %addr, <8 x float> %a) nounwind uwtable ssp {
 entry:
-  %0 = tail call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a, i8 0)
+  %0 = tail call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a, i8 1)
   %1 = bitcast float* %addr to <4 x float>*
   store <4 x float> %0, <4 x float>* %1, align 16
   ret void
@@ -33,12 +33,12 @@ entry:
 declare <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float>, i8) nounwind readnone
 
 ; CHECK: @t2
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
+; CHECK-NOT: vextractf128 $1, %ymm0, %xmm0
 ; CHECK-NOT: vmovaps %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
+; CHECK: vextractf128 $1, %ymm0, (%rdi)
 define void @t2(double* nocapture %addr, <4 x double> %a) nounwind uwtable ssp {
 entry:
-  %0 = tail call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a, i8 0)
+  %0 = tail call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a, i8 1)
   %1 = bitcast double* %addr to <2 x double>*
   store <2 x double> %0, <2 x double>* %1, align 16
   ret void
@@ -47,10 +47,43 @@ entry:
 declare <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double>, i8) nounwind readnone
 
 ; CHECK: @t4
-; CHECK-NOT: vextractf128 $0, %ymm0, %xmm0
+; CHECK-NOT: vextractf128 $1, %ymm0, %xmm0
 ; CHECK-NOT: vmovaps %xmm0, (%rdi)
-; CHECK: vextractf128 $0, %ymm0, (%rdi)
+; CHECK: vextractf128 $1, %ymm0, (%rdi)
 define void @t4(<2 x i64>* nocapture %addr, <4 x i64> %a) nounwind uwtable ssp {
+entry:
+  %0 = bitcast <4 x i64> %a to <8 x i32>
+  %1 = tail call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %0, i8 1)
+  %2 = bitcast <4 x i32> %1 to <2 x i64>
+  store <2 x i64> %2, <2 x i64>* %addr, align 16
+  ret void
+}
+
+declare <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32>, i8) nounwind readnone
+
+; CHECK: @t5
+; CHECK: vmovaps %xmm0, (%rdi)
+define void @t5(float* nocapture %addr, <8 x float> %a) nounwind uwtable ssp {
+entry:
+  %0 = tail call <4 x float> @llvm.x86.avx.vextractf128.ps.256(<8 x float> %a, i8 0)
+  %1 = bitcast float* %addr to <4 x float>*
+  store <4 x float> %0, <4 x float>* %1, align 16
+  ret void
+}
+
+; CHECK: @t6
+; CHECK: vmovaps %xmm0, (%rdi)
+define void @t6(double* nocapture %addr, <4 x double> %a) nounwind uwtable ssp {
+entry:
+  %0 = tail call <2 x double> @llvm.x86.avx.vextractf128.pd.256(<4 x double> %a, i8 0)
+  %1 = bitcast double* %addr to <2 x double>*
+  store <2 x double> %0, <2 x double>* %1, align 16
+  ret void
+}
+
+; CHECK: @t7
+; CHECK: vmovaps %xmm0, (%rdi)
+define void @t7(<2 x i64>* nocapture %addr, <4 x i64> %a) nounwind uwtable ssp {
 entry:
   %0 = bitcast <4 x i64> %a to <8 x i32>
   %1 = tail call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %0, i8 0)
@@ -59,4 +92,13 @@ entry:
   ret void
 }
 
-declare <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32>, i8) nounwind readnone
+; CHECK: @t8
+; CHECK: vmovups %xmm0, (%rdi)
+define void @t8(<2 x i64>* nocapture %addr, <4 x i64> %a) nounwind uwtable ssp {
+entry:
+  %0 = bitcast <4 x i64> %a to <8 x i32>
+  %1 = tail call <4 x i32> @llvm.x86.avx.vextractf128.si.256(<8 x i32> %0, i8 0)
+  %2 = bitcast <4 x i32> %1 to <2 x i64>
+  store <2 x i64> %2, <2 x i64>* %addr, align 1
+  ret void
+}