Index: llvm/lib/Target/X86/X86InstrFMA.td
===================================================================
--- llvm/lib/Target/X86/X86InstrFMA.td
+++ llvm/lib/Target/X86/X86InstrFMA.td
@@ -223,9 +223,12 @@
 multiclass fma3s<bits<8> opc132, bits<8> opc213, bits<8> opc231,
                  string OpStr, Intrinsic IntF32, Intrinsic IntF64,
                  SDNode OpNode> {
+  let ExeDomain = SSEPackedSingle in
   defm SS : fma3s_forms<opc132, opc213, opc231, OpStr, "ss", OpNode,
                         FR32, f32mem>,
             fma3s_int_forms<opc132, opc213, opc231, OpStr, "ss", VR128, ssmem>;
+
+  let ExeDomain = SSEPackedDouble in
   defm SD : fma3s_forms<opc132, opc213, opc231, OpStr, "sd", OpNode,
                         FR64, f64mem>,
             fma3s_int_forms<opc132, opc213, opc231, OpStr, "sd", VR128, sdmem>,
Index: llvm/test/CodeGen/X86/fma-commute-x86.ll
===================================================================
--- llvm/test/CodeGen/X86/fma-commute-x86.ll
+++ llvm/test/CodeGen/X86/fma-commute-x86.ll
@@ -8,8 +8,8 @@
 define <4 x float> @test_x86_fmadd_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_baa_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfmadd213ss %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
@@ -102,8 +102,8 @@
 define <2 x double> @test_x86_fmadd_baa_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_baa_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfmadd213sd %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
@@ -113,7 +113,7 @@
 define <2 x double> @test_x86_fmadd_aba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_aba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rcx), %xmm0
+; CHECK-NEXT: vmovapd	(%rcx), %xmm0
 ; CHECK-NEXT: vfmadd132sd (%rdx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
@@ -123,7 +123,7 @@
 define <2 x double> @test_x86_fmadd_bba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_bba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rdx), %xmm0
+; CHECK-NEXT: vmovapd	(%rdx), %xmm0
 ; CHECK-NEXT: vfmadd213sd (%rcx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
@@ -197,8 +197,8 @@
 define <4 x float> @test_x86_fnmadd_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_baa_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfnmadd213ss %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
@@ -291,8 +291,8 @@
 define <2 x double> @test_x86_fnmadd_baa_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_baa_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfnmadd213sd %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
@@ -302,7 +302,7 @@
 define <2 x double> @test_x86_fnmadd_aba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_aba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rcx), %xmm0
+; CHECK-NEXT: vmovapd	(%rcx), %xmm0
 ; CHECK-NEXT: vfnmadd132sd (%rdx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
@@ -312,7 +312,7 @@
 define <2 x double> @test_x86_fnmadd_bba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_bba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rdx), %xmm0
+; CHECK-NEXT: vmovapd	(%rdx), %xmm0
 ; CHECK-NEXT: vfnmadd213sd (%rcx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
@@ -386,8 +386,8 @@
 define <4 x float> @test_x86_fmsub_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_baa_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfmsub213ss %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
@@ -480,8 +480,8 @@
 define <2 x double> @test_x86_fmsub_baa_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_baa_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfmsub213sd %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
@@ -491,7 +491,7 @@
 define <2 x double> @test_x86_fmsub_aba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_aba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rcx), %xmm0
+; CHECK-NEXT: vmovapd	(%rcx), %xmm0
 ; CHECK-NEXT: vfmsub132sd (%rdx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
@@ -501,7 +501,7 @@
 define <2 x double> @test_x86_fmsub_bba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_bba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rdx), %xmm0
+; CHECK-NEXT: vmovapd	(%rdx), %xmm0
 ; CHECK-NEXT: vfmsub213sd (%rcx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
@@ -575,8 +575,8 @@
 define <4 x float> @test_x86_fnmsub_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_baa_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfnmsub213ss %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
@@ -669,8 +669,8 @@
 define <2 x double> @test_x86_fnmsub_baa_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_baa_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
+; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
 ; CHECK-NEXT: vfnmsub213sd %xmm1, %xmm1, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
@@ -680,7 +680,7 @@
 define <2 x double> @test_x86_fnmsub_aba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_aba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rcx), %xmm0
+; CHECK-NEXT: vmovapd	(%rcx), %xmm0
 ; CHECK-NEXT: vfnmsub132sd (%rdx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
@@ -690,7 +690,7 @@
 define <2 x double> @test_x86_fnmsub_bba_sd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_bba_sd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps	(%rdx), %xmm0
+; CHECK-NEXT: vmovapd	(%rdx), %xmm0
 ; CHECK-NEXT: vfnmsub213sd (%rcx), %xmm0, %xmm0
 ; CHECK-NEXT: retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
Index: llvm/test/CodeGen/X86/fma-intrinsics-x86.ll
===================================================================
--- llvm/test/CodeGen/X86/fma-intrinsics-x86.ll
+++ llvm/test/CodeGen/X86/fma-intrinsics-x86.ll
@@ -10,8 +10,8 @@
 ; CHECK-LABEL: test_x86_fma_vfmadd_ss:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfmadd132ss     (%rdx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0
@@ -27,8 +27,8 @@
 ; CHECK-LABEL: test_x86_fma_vfmadd_bac_ss:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfmadd132ss (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfmadd213ss %xmm2, %xmm0, %xmm1
@@ -45,8 +45,8 @@
 ; CHECK-LABEL: test_x86_fma_vfmadd_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfmadd132sd (%rdx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0
@@ -62,12 +62,12 @@
 ; CHECK-LABEL: test_x86_fma_vfmadd_bac_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfmadd132sd (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfmadd213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps	%xmm1, %xmm0
+; CHECK-FMA-NEXT:    vmovapd	%xmm1, %xmm0
 ;
 ; CHECK-FMA4-NEXT: vfmaddsd %xmm2, %xmm0, %xmm1, %xmm0
 ;
@@ -171,8 +171,8 @@
 ; CHECK-LABEL: test_x86_fma_vfmsub_bac_ss:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfmsub132ss (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfmsub213ss %xmm2, %xmm0, %xmm1
@@ -190,8 +190,8 @@
 ; CHECK-LABEL: test_x86_fma_vfmsub_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfmsub132sd (%rdx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0
@@ -207,12 +207,12 @@
 ; CHECK-LABEL: test_x86_fma_vfmsub_bac_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfmsub132sd (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfmsub213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps	%xmm1, %xmm0
+; CHECK-FMA-NEXT:    vmovapd	%xmm1, %xmm0
 ;
 ; CHECK-FMA4-NEXT: vfmsubsd %xmm2, %xmm0, %xmm1, %xmm0
 ;
@@ -299,8 +299,8 @@
 ; CHECK-LABEL: test_x86_fma_vfnmadd_ss:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmadd132ss (%rdx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0
@@ -316,8 +316,8 @@
 ; CHECK-LABEL: test_x86_fma_vfnmadd_bac_ss:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmadd132ss (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmadd213ss %xmm2, %xmm0, %xmm1
@@ -335,8 +335,8 @@
 ; CHECK-LABEL: test_x86_fma_vfnmadd_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmadd132sd (%rdx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0
@@ -352,12 +352,12 @@
 ; CHECK-LABEL: test_x86_fma_vfnmadd_bac_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmadd132sd (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmadd213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps	%xmm1, %xmm0
+; CHECK-FMA-NEXT:    vmovapd	%xmm1, %xmm0
 ;
 ; CHECK-FMA4-NEXT: vfnmaddsd %xmm2, %xmm0, %xmm1, %xmm0
 ;
@@ -444,8 +444,8 @@
 ; CHECK-LABEL: test_x86_fma_vfnmsub_ss:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmsub132ss (%rdx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0
@@ -461,8 +461,8 @@
 ; CHECK-LABEL: test_x86_fma_vfnmsub_bac_ss:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmsub132ss (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmsub213ss %xmm2, %xmm0, %xmm1
@@ -480,8 +480,8 @@
 ; CHECK-LABEL: test_x86_fma_vfnmsub_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmsub132sd (%rdx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0
@@ -497,12 +497,12 @@
 ; CHECK-LABEL: test_x86_fma_vfnmsub_bac_sd:
 ; CHECK-NEXT:  # BB#0:
 ;
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovap{{s|d}} {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
+; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%r8\), %xmm1}}
 ; CHECK-FMA-WIN-NEXT: vfnmsub132sd (%rcx), %xmm1, %xmm0
 ;
 ; CHECK-FMA-NEXT:    vfnmsub213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps	%xmm1, %xmm0
+; CHECK-FMA-NEXT:    vmovapd	%xmm1, %xmm0
 ;
 ; CHECK-FMA4-NEXT: vfnmsubsd %xmm2, %xmm0, %xmm1, %xmm0
 ;
Index: llvm/test/CodeGen/X86/fma-scalar-memfold.ll
===================================================================
--- llvm/test/CodeGen/X86/fma-scalar-memfold.ll
+++ llvm/test/CodeGen/X86/fma-scalar-memfold.ll
@@ -216,7 +216,7 @@
 ; CHECK-LABEL: fmadd_aab_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfmadd213sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
@@ -237,7 +237,7 @@
 ; CHECK-LABEL: fmadd_aba_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfmadd132sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
@@ -258,7 +258,7 @@
 ; CHECK-LABEL: fmsub_aab_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfmsub213sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
@@ -279,7 +279,7 @@
 ; CHECK-LABEL: fmsub_aba_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfmsub132sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
@@ -300,7 +300,7 @@
 ; CHECK-LABEL: fnmadd_aab_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfnmadd213sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
@@ -321,7 +321,7 @@
 ; CHECK-LABEL: fnmadd_aba_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfnmadd132sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
@@ -342,7 +342,7 @@
 ; CHECK-LABEL: fnmsub_aab_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfnmsub213sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
@@ -363,7 +363,7 @@
 ; CHECK-LABEL: fnmsub_aba_sd:
 ; CHECK:      vmovsd (%rcx), %[[XMM:xmm[0-9]+]]
 ; CHECK-NEXT: vfnmsub132sd (%rdx), %[[XMM]], %[[XMM]]
-; CHECK-NEXT: vmovlps %[[XMM]], (%rcx)
+; CHECK-NEXT: vmovlpd %[[XMM]], (%rcx)
 ; CHECK-NEXT: ret
   %a.val = load double, double* %a
   %av0 = insertelement <2 x double> undef, double %a.val, i32 0
Index: llvm/test/CodeGen/X86/fma_patterns.ll
===================================================================
--- llvm/test/CodeGen/X86/fma_patterns.ll
+++ llvm/test/CodeGen/X86/fma_patterns.ll
@@ -1114,7 +1114,7 @@
 define double @test_f64_fneg_fmul(double %x, double %y) #0 {
 ; FMA-LABEL: test_f64_fneg_fmul:
 ; FMA:       # BB#0:
-; FMA-NEXT:    vxorps %xmm2, %xmm2, %xmm2
+; FMA-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
 ; FMA-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0
 ; FMA-NEXT:    retq
 ;