Index: lib/Target/X86/X86FastISel.cpp
===================================================================
--- lib/Target/X86/X86FastISel.cpp
+++ lib/Target/X86/X86FastISel.cpp
@@ -434,6 +434,7 @@
                                    X86AddressMode &AM,
                                    MachineMemOperand *MMO, bool Aligned) {
   bool HasSSE2 = Subtarget->hasSSE2();
+  bool HasSSE4A = Subtarget->hasSSE4A();
   bool HasAVX = Subtarget->hasAVX();
   bool IsNonTemporal = MMO && MMO->isNonTemporal();
 
@@ -461,12 +462,22 @@
     Opc = (IsNonTemporal && HasSSE2) ? X86::MOVNTI_64mr : X86::MOV64mr;
     break;
   case MVT::f32:
-    Opc = X86ScalarSSEf32 ? 
-          (HasAVX ? X86::VMOVSSmr : X86::MOVSSmr) : X86::ST_Fp32m;
+    if (X86ScalarSSEf32) {
+      if (IsNonTemporal && HasSSE4A)
+        Opc = X86::MOVNTSS;
+      else
+        Opc = HasAVX ? X86::VMOVSSmr : X86::MOVSSmr;
+    } else
+      Opc = X86::ST_Fp32m;
     break;
   case MVT::f64:
-    Opc = X86ScalarSSEf64 ?
-          (HasAVX ? X86::VMOVSDmr : X86::MOVSDmr) : X86::ST_Fp64m;
+    if (X86ScalarSSEf32) {
+      if (IsNonTemporal && HasSSE4A)
+        Opc = X86::MOVNTSD;
+      else
+        Opc = HasAVX ? X86::VMOVSDmr : X86::MOVSDmr;
+    } else
+      Opc = X86::ST_Fp64m;
     break;
   case MVT::v4f32:
     if (Aligned) {
@@ -1937,7 +1948,7 @@
 
   const TargetRegisterClass *RC = TLI.getRegClassFor(RetVT);
   unsigned ResultReg;
-  
+
   if (Subtarget->hasAVX()) {
     const TargetRegisterClass *FR32 = &X86::FR32RegClass;
     const TargetRegisterClass *VR128 = &X86::VR128RegClass;
@@ -1951,7 +1962,7 @@
       (RetVT.SimpleTy == MVT::f32) ? X86::VCMPSSrr : X86::VCMPSDrr;
     unsigned BlendOpcode =
       (RetVT.SimpleTy == MVT::f32) ? X86::VBLENDVPSrr : X86::VBLENDVPDrr;
-    
+
     unsigned CmpReg = fastEmitInst_rri(CmpOpcode, FR32, CmpLHSReg, CmpLHSIsKill,
                                        CmpRHSReg, CmpRHSIsKill, CC);
     unsigned VBlendReg = fastEmitInst_rrr(BlendOpcode, VR128, RHSReg, RHSIsKill,
@@ -2289,7 +2300,7 @@
       BuildMI(*FuncInfo.MBB, FuncInfo.InsertPt, DbgLoc,
               TII.get(X86::VMOVPDI2DIrr), ResultReg)
           .addReg(InputReg, RegState::Kill);
-      
+
       // The result value is in the lower 16-bits of ResultReg.
       unsigned RegIdx = X86::sub_16bit;
       ResultReg = fastEmitInst_extractsubreg(MVT::i16, ResultReg, true, RegIdx);
@@ -3270,7 +3281,7 @@
     unsigned Reg = getRegForValue(I->getOperand(0));
     if (Reg == 0)
       return false;
-      
+
     // No instruction is needed for conversion. Reuse the register used by
     // the fist operand.
     updateValueMap(I, Reg);
Index: test/CodeGen/X86/fast-isel-nontemporal.ll
===================================================================
--- test/CodeGen/X86/fast-isel-nontemporal.ll
+++ test/CodeGen/X86/fast-isel-nontemporal.ll
@@ -1,4 +1,5 @@
-; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=SSE2
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse2 -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE2
+; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+sse4a -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=SSE --check-prefix=SSE4A
 ; RUN: llc -mtriple=x86_64-unknown-unknown -mattr=+avx -fast-isel -O0 < %s | FileCheck %s --check-prefix=ALL --check-prefix=AVX
 
 define void @test_nti32(i32* nocapture %ptr, i32 %X) {
@@ -21,12 +22,52 @@
   ret void
 }
 
-define void @test_nt4xfloat(<4 x float>* nocapture %ptr, <4 x float> %X) {
-; SSE2-LABEL: test_nt4xfloat:
+define void @test_ntfloat(float* nocapture %ptr, float %X) {
+; SSE2-LABEL: test_ntfloat:
+; SSE2:       # BB#0: # %entry
+; SSE2-NEXT:    movss %xmm0, (%rdi)
+; SSE2-NEXT:    retq
+;
+; SSE4A-LABEL: test_ntfloat:
+; SSE4A:       # BB#0: # %entry
+; SSE4A-NEXT:    movntss %xmm0, (%rdi)
+; SSE4A-NEXT:    retq
+;
+; AVX-LABEL: test_ntfloat:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovss %xmm0, (%rdi)
+; AVX-NEXT:    retq
+entry:
+  store float %X, float* %ptr, align 4, !nontemporal !1
+  ret void
+}
+
+define void @test_ntdouble(double* nocapture %ptr, double %X) {
+; SSE2-LABEL: test_ntdouble:
 ; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    movntps %xmm0, (%rdi)
+; SSE2-NEXT:    movsd %xmm0, (%rdi)
 ; SSE2-NEXT:    retq
 ;
+; SSE4A-LABEL: test_ntdouble:
+; SSE4A:       # BB#0: # %entry
+; SSE4A-NEXT:    movntsd %xmm0, (%rdi)
+; SSE4A-NEXT:    retq
+;
+; AVX-LABEL: test_ntdouble:
+; AVX:       # BB#0: # %entry
+; AVX-NEXT:    vmovsd %xmm0, (%rdi)
+; AVX-NEXT:    retq
+entry:
+  store double %X, double* %ptr, align 8, !nontemporal !1
+  ret void
+}
+
+define void @test_nt4xfloat(<4 x float>* nocapture %ptr, <4 x float> %X) {
+; SSE-LABEL: test_nt4xfloat:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntps %xmm0, (%rdi)
+; SSE-NEXT:    retq
+;
 ; AVX-LABEL: test_nt4xfloat:
 ; AVX:       # BB#0: # %entry
 ; AVX-NEXT:    vmovntps %xmm0, (%rdi)
@@ -37,10 +78,10 @@
 }
 
 define void @test_nt2xdouble(<2 x double>* nocapture %ptr, <2 x double> %X) {
-; SSE2-LABEL: test_nt2xdouble:
-; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    movntpd %xmm0, (%rdi)
-; SSE2-NEXT:    retq
+; SSE-LABEL: test_nt2xdouble:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntpd %xmm0, (%rdi)
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_nt2xdouble:
 ; AVX:       # BB#0: # %entry
@@ -52,10 +93,10 @@
 }
 
 define void @test_nt2xi64(<2 x i64>* nocapture %ptr, <2 x i64> %X) {
-; SSE2-LABEL: test_nt2xi64:
-; SSE2:       # BB#0: # %entry
-; SSE2-NEXT:    movntdq %xmm0, (%rdi)
-; SSE2-NEXT:    retq
+; SSE-LABEL: test_nt2xi64:
+; SSE:       # BB#0: # %entry
+; SSE-NEXT:    movntdq %xmm0, (%rdi)
+; SSE-NEXT:    retq
 ;
 ; AVX-LABEL: test_nt2xi64:
 ; AVX:       # BB#0: # %entry