[X86][FMA] Regenerate FMA tests

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@312871 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/test/CodeGen/X86/fma-commute-x86.ll b/test/CodeGen/X86/fma-commute-x86.ll
index 162a97ac025c511a0093b00ef50f08aaf7b51aa2..cb8d396a0e9f7fc7496fb3f0a04f8d15940beb65 100644 (file)
--- a/test/CodeGen/X86/fma-commute-x86.ll
+++ b/test/CodeGen/X86/fma-commute-x86.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-pc-win32 -mcpu=core-avx2 | FileCheck %s
 ; RUN: llc < %s -mtriple=x86_64-pc-win32 -mattr=+fma | FileCheck %s
 ; RUN: llc < %s -mcpu=bdver2 -mtriple=x86_64-pc-win32 -mattr=-fma4 | FileCheck %s
@@ -8,10 +9,10 @@ declare <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float>, <4 x float>, <4 x float
 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: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfmadd213ss %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; 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
   ret <4 x float> %res
 }
@@ -19,9 +20,9 @@ define <4 x float> @test_x86_fmadd_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmadd_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_aba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfmadd132ss (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfmadd132ss (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -29,9 +30,9 @@ define <4 x float> @test_x86_fmadd_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmadd_bba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_bba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfmadd213ss (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfmadd213ss (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -40,9 +41,9 @@ declare <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float>, <4 x float>, <4 x float
 define <4 x float> @test_x86_fmadd_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_baa_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfmadd132ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfmadd132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -50,9 +51,9 @@ define <4 x float> @test_x86_fmadd_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmadd_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_aba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfmadd231ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfmadd231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -60,9 +61,9 @@ define <4 x float> @test_x86_fmadd_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmadd_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_bba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfmadd213ps (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfmadd213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -71,9 +72,9 @@ declare <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x f
 define <8 x float> @test_x86_fmadd_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_baa_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfmadd132ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfmadd132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -81,9 +82,9 @@ define <8 x float> @test_x86_fmadd_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 define <8 x float> @test_x86_fmadd_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_aba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfmadd231ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfmadd231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -91,9 +92,9 @@ define <8 x float> @test_x86_fmadd_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 define <8 x float> @test_x86_fmadd_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_bba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %ymm0
-; CHECK-NEXT: vfmadd213ps (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-NEXT:    vfmadd213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -102,10 +103,10 @@ declare <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double>, <2 x double>, <2 x do
 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: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfmadd213sd %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; 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
   ret <2 x double> %res
 }
@@ -113,9 +114,9 @@ define <2 x double> @test_x86_fmadd_baa_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfmadd132sd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -123,9 +124,9 @@ define <2 x double> @test_x86_fmadd_aba_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfmadd213sd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -134,9 +135,9 @@ declare <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double>, <2 x double>, <2 x do
 define <2 x double> @test_x86_fmadd_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_baa_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfmadd132pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfmadd132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -144,9 +145,9 @@ define <2 x double> @test_x86_fmadd_baa_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fmadd_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_aba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfmadd231pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfmadd231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -154,9 +155,9 @@ define <2 x double> @test_x86_fmadd_aba_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fmadd_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_bba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfmadd213pd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-NEXT:    vfmadd213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -165,9 +166,9 @@ declare <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double>, <4 x double>, <4
 define <4 x double> @test_x86_fmadd_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_baa_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfmadd132pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfmadd132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -175,9 +176,9 @@ define <4 x double> @test_x86_fmadd_baa_pd_y(<4 x double> %a, <4 x double> %b) #
 define <4 x double> @test_x86_fmadd_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_aba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfmadd231pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfmadd231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -185,9 +186,9 @@ define <4 x double> @test_x86_fmadd_aba_pd_y(<4 x double> %a, <4 x double> %b) #
 define <4 x double> @test_x86_fmadd_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmadd_bba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %ymm0
-; CHECK-NEXT: vfmadd213pd (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-NEXT:    vfmadd213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -197,10 +198,10 @@ declare <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float>, <4 x float>, <4 x floa
 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: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfnmadd213ss %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; 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
   ret <4 x float> %res
 }
@@ -208,9 +209,9 @@ define <4 x float> @test_x86_fnmadd_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmadd_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_aba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfnmadd132ss (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfnmadd132ss (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -218,9 +219,9 @@ define <4 x float> @test_x86_fnmadd_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmadd_bba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_bba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfnmadd213ss (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfnmadd213ss (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -229,9 +230,9 @@ declare <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float>, <4 x float>, <4 x floa
 define <4 x float> @test_x86_fnmadd_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_baa_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfnmadd132ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfnmadd132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -239,9 +240,9 @@ define <4 x float> @test_x86_fnmadd_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmadd_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_aba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfnmadd231ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfnmadd231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -249,9 +250,9 @@ define <4 x float> @test_x86_fnmadd_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmadd_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_bba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfnmadd213ps (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfnmadd213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -260,9 +261,9 @@ declare <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float>, <8 x float>, <8 x
 define <8 x float> @test_x86_fnmadd_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_baa_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfnmadd132ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfnmadd132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -270,9 +271,9 @@ define <8 x float> @test_x86_fnmadd_baa_ps_y(<8 x float> %a, <8 x float> %b) #0
 define <8 x float> @test_x86_fnmadd_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_aba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfnmadd231ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfnmadd231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -280,9 +281,9 @@ define <8 x float> @test_x86_fnmadd_aba_ps_y(<8 x float> %a, <8 x float> %b) #0
 define <8 x float> @test_x86_fnmadd_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_bba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %ymm0
-; CHECK-NEXT: vfnmadd213ps (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-NEXT:    vfnmadd213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -291,10 +292,10 @@ declare <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double>, <2 x double>, <2 x d
 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: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfnmadd213sd %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; 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
   ret <2 x double> %res
 }
@@ -302,9 +303,9 @@ define <2 x double> @test_x86_fnmadd_baa_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfnmadd132sd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -312,9 +313,9 @@ define <2 x double> @test_x86_fnmadd_aba_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfnmadd213sd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -323,9 +324,9 @@ declare <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double>, <2 x double>, <2 x d
 define <2 x double> @test_x86_fnmadd_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_baa_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfnmadd132pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfnmadd132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -333,9 +334,9 @@ define <2 x double> @test_x86_fnmadd_baa_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fnmadd_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_aba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfnmadd231pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfnmadd231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -343,9 +344,9 @@ define <2 x double> @test_x86_fnmadd_aba_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fnmadd_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_bba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfnmadd213pd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-NEXT:    vfnmadd213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -354,9 +355,9 @@ declare <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double>, <4 x double>, <4
 define <4 x double> @test_x86_fnmadd_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_baa_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfnmadd132pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfnmadd132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -364,9 +365,9 @@ define <4 x double> @test_x86_fnmadd_baa_pd_y(<4 x double> %a, <4 x double> %b)
 define <4 x double> @test_x86_fnmadd_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_aba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfnmadd231pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfnmadd231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -374,9 +375,9 @@ define <4 x double> @test_x86_fnmadd_aba_pd_y(<4 x double> %a, <4 x double> %b)
 define <4 x double> @test_x86_fnmadd_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmadd_bba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %ymm0
-; CHECK-NEXT: vfnmadd213pd (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-NEXT:    vfnmadd213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -386,10 +387,10 @@ declare <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float>, <4 x float>, <4 x float
 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: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfmsub213ss %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; 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
   ret <4 x float> %res
 }
@@ -397,9 +398,9 @@ define <4 x float> @test_x86_fmsub_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmsub_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_aba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfmsub132ss (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfmsub132ss (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -407,9 +408,9 @@ define <4 x float> @test_x86_fmsub_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmsub_bba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_bba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfmsub213ss (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfmsub213ss (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -418,9 +419,9 @@ declare <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float>, <4 x float>, <4 x float
 define <4 x float> @test_x86_fmsub_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_baa_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfmsub132ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfmsub132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -428,9 +429,9 @@ define <4 x float> @test_x86_fmsub_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmsub_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_aba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfmsub231ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfmsub231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -438,9 +439,9 @@ define <4 x float> @test_x86_fmsub_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fmsub_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_bba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfmsub213ps (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfmsub213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -449,9 +450,9 @@ declare <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x f
 define <8 x float> @test_x86_fmsub_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_baa_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfmsub132ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfmsub132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -459,9 +460,9 @@ define <8 x float> @test_x86_fmsub_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 define <8 x float> @test_x86_fmsub_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_aba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfmsub231ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfmsub231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -469,9 +470,9 @@ define <8 x float> @test_x86_fmsub_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 define <8 x float> @test_x86_fmsub_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_bba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %ymm0
-; CHECK-NEXT: vfmsub213ps (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-NEXT:    vfmsub213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -480,10 +481,10 @@ declare <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double>, <2 x double>, <2 x do
 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: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfmsub213sd %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; 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
   ret <2 x double> %res
 }
@@ -491,9 +492,9 @@ define <2 x double> @test_x86_fmsub_baa_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfmsub132sd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -501,9 +502,9 @@ define <2 x double> @test_x86_fmsub_aba_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfmsub213sd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -512,9 +513,9 @@ declare <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double>, <2 x double>, <2 x do
 define <2 x double> @test_x86_fmsub_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_baa_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfmsub132pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfmsub132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -522,9 +523,9 @@ define <2 x double> @test_x86_fmsub_baa_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fmsub_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_aba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfmsub231pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfmsub231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -532,9 +533,9 @@ define <2 x double> @test_x86_fmsub_aba_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fmsub_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_bba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfmsub213pd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-NEXT:    vfmsub213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -543,9 +544,9 @@ declare <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double>, <4 x double>, <4
 define <4 x double> @test_x86_fmsub_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_baa_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfmsub132pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfmsub132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -553,9 +554,9 @@ define <4 x double> @test_x86_fmsub_baa_pd_y(<4 x double> %a, <4 x double> %b) #
 define <4 x double> @test_x86_fmsub_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_aba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfmsub231pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfmsub231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -563,9 +564,9 @@ define <4 x double> @test_x86_fmsub_aba_pd_y(<4 x double> %a, <4 x double> %b) #
 define <4 x double> @test_x86_fmsub_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fmsub_bba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %ymm0
-; CHECK-NEXT: vfmsub213pd (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-NEXT:    vfmsub213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -575,10 +576,10 @@ declare <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float>, <4 x float>, <4 x floa
 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: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfnmsub213ss %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; 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
   ret <4 x float> %res
 }
@@ -586,9 +587,9 @@ define <4 x float> @test_x86_fnmsub_baa_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmsub_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_aba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfnmsub132ss (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfnmsub132ss (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -596,9 +597,9 @@ define <4 x float> @test_x86_fnmsub_aba_ss(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmsub_bba_ss(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_bba_ss:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfnmsub213ss (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfnmsub213ss (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -607,9 +608,9 @@ declare <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float>, <4 x float>, <4 x floa
 define <4 x float> @test_x86_fnmsub_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_baa_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfnmsub132ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfnmsub132ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %b, <4 x float> %a, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -617,9 +618,9 @@ define <4 x float> @test_x86_fnmsub_baa_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmsub_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_aba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %xmm0
-; CHECK-NEXT: vfnmsub231ps (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-NEXT:    vfnmsub231ps (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %a, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -627,9 +628,9 @@ define <4 x float> @test_x86_fnmsub_aba_ps(<4 x float> %a, <4 x float> %b) #0 {
 define <4 x float> @test_x86_fnmsub_bba_ps(<4 x float> %a, <4 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_bba_ps:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %xmm0
-; CHECK-NEXT: vfnmsub213ps (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-NEXT:    vfnmsub213ps (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %b, <4 x float> %b, <4 x float> %a) nounwind
   ret <4 x float> %res
 }
@@ -638,9 +639,9 @@ declare <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x
 define <8 x float> @test_x86_fnmsub_baa_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_baa_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfnmsub132ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfnmsub132ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %b, <8 x float> %a, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -648,9 +649,9 @@ define <8 x float> @test_x86_fnmsub_baa_ps_y(<8 x float> %a, <8 x float> %b) #0
 define <8 x float> @test_x86_fnmsub_aba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_aba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rcx), %ymm0
-; CHECK-NEXT: vfnmsub231ps (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rcx), %ymm0
+; CHECK-NEXT:    vfnmsub231ps (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -658,9 +659,9 @@ define <8 x float> @test_x86_fnmsub_aba_ps_y(<8 x float> %a, <8 x float> %b) #0
 define <8 x float> @test_x86_fnmsub_bba_ps_y(<8 x float> %a, <8 x float> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_bba_ps_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovaps  (%rdx), %ymm0
-; CHECK-NEXT: vfnmsub213ps (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-NEXT:    vfnmsub213ps (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %b, <8 x float> %b, <8 x float> %a) nounwind
   ret <8 x float> %res
 }
@@ -669,10 +670,10 @@ declare <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double>, <2 x double>, <2 x d
 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: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-NEXT: vfnmsub213sd %xmm1, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; 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
   ret <2 x double> %res
 }
@@ -680,9 +681,9 @@ define <2 x double> @test_x86_fnmsub_baa_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfnmsub132sd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -690,9 +691,9 @@ define <2 x double> @test_x86_fnmsub_aba_sd(<2 x double> %a, <2 x double> %b) #0
 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: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfnmsub213sd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; 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
   ret <2 x double> %res
 }
@@ -701,9 +702,9 @@ declare <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double>, <2 x double>, <2 x d
 define <2 x double> @test_x86_fnmsub_baa_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_baa_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfnmsub132pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfnmsub132pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %b, <2 x double> %a, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -711,9 +712,9 @@ define <2 x double> @test_x86_fnmsub_baa_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fnmsub_aba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_aba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %xmm0
-; CHECK-NEXT: vfnmsub231pd (%rdx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-NEXT:    vfnmsub231pd (%rdx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %a, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -721,9 +722,9 @@ define <2 x double> @test_x86_fnmsub_aba_pd(<2 x double> %a, <2 x double> %b) #0
 define <2 x double> @test_x86_fnmsub_bba_pd(<2 x double> %a, <2 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_bba_pd:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %xmm0
-; CHECK-NEXT: vfnmsub213pd (%rcx), %xmm0, %xmm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-NEXT:    vfnmsub213pd (%rcx), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %b, <2 x double> %b, <2 x double> %a) nounwind
   ret <2 x double> %res
 }
@@ -732,9 +733,9 @@ declare <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double>, <4 x double>, <4
 define <4 x double> @test_x86_fnmsub_baa_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_baa_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfnmsub132pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfnmsub132pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %b, <4 x double> %a, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -742,9 +743,9 @@ define <4 x double> @test_x86_fnmsub_baa_pd_y(<4 x double> %a, <4 x double> %b)
 define <4 x double> @test_x86_fnmsub_aba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_aba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rcx), %ymm0
-; CHECK-NEXT: vfnmsub231pd (%rdx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rcx), %ymm0
+; CHECK-NEXT:    vfnmsub231pd (%rdx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }
@@ -752,9 +753,9 @@ define <4 x double> @test_x86_fnmsub_aba_pd_y(<4 x double> %a, <4 x double> %b)
 define <4 x double> @test_x86_fnmsub_bba_pd_y(<4 x double> %a, <4 x double> %b) #0 {
 ; CHECK-LABEL: test_x86_fnmsub_bba_pd_y:
 ; CHECK:       # BB#0:
-; CHECK-NEXT: vmovapd  (%rdx), %ymm0
-; CHECK-NEXT: vfnmsub213pd (%rcx), %ymm0, %ymm0
-; CHECK-NEXT: retq
+; CHECK-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-NEXT:    vfnmsub213pd (%rcx), %ymm0, %ymm0
+; CHECK-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %b, <4 x double> %b, <4 x double> %a) nounwind
   ret <4 x double> %res
 }

diff --git a/test/CodeGen/X86/fma-intrinsics-x86.ll b/test/CodeGen/X86/fma-intrinsics-x86.ll
index d30f1c9899e02530608b1388cb32b7c54aaf363d..68f39469a82c6c0fdfc8639d954e2d8f820ba9c8 100644 (file)
--- a/test/CodeGen/X86/fma-intrinsics-x86.ll
+++ b/test/CodeGen/X86/fma-intrinsics-x86.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=corei7-avx -mattr=+fma | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=core-avx2 -mattr=+fma,+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA
 ; RUN: llc < %s -mtriple=x86_64-pc-windows -mcpu=core-avx2 -mattr=+fma,+avx2 | FileCheck %s --check-prefix=CHECK --check-prefix=CHECK-FMA-WIN
@@ -7,143 +8,176 @@
 
 ; VFMADD
 define <4 x float> @test_x86_fma_vfmadd_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213ss     (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddss %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 
 define <4 x float> @test_x86_fma_vfmadd_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_bac_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213ss (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_bac_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213ss %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddss %xmm2, %xmm0, %xmm1, %xmm0
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_bac_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_bac_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddss %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfmadd_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddsd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 
 define <2 x double> @test_x86_fma_vfmadd_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_bac_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_bac_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovapd   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddsd %xmm2, %xmm0, %xmm1, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_bac_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_bac_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddsd %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>)
 
 define <4 x float> @test_x86_fma_vfmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_ps:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213ps (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ps:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddps %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ps:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213ps (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_ps:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddps %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfmadd.ps(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfmadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_pd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213pd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_pd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddpd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_pd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213pd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_pd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddpd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfmadd.pd(<2 x double>, <2 x double>, <2 x double>)
 
 define <8 x float> @test_x86_fma_vfmadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_ps_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213ps (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_ps_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmaddps %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_ps_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213ps (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_ps_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddps %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
   ret <8 x float> %res
 }
 declare <8 x float> @llvm.x86.fma.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>)
 
 define <4 x double> @test_x86_fma_vfmadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmadd_pd_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmadd213pd (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmadd_pd_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmaddpd %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmadd_pd_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmadd213pd (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmadd_pd_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddpd %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
   ret <4 x double> %res
 }
@@ -151,144 +185,176 @@ declare <4 x double> @llvm.x86.fma.vfmadd.pd.256(<4 x double>, <4 x double>, <4
 
 ; VFMSUB
 define <4 x float> @test_x86_fma_vfmsub_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213ss (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213ss %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubss %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 
 define <4 x float> @test_x86_fma_vfmsub_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_bac_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213ss (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_bac_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213ss %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubss %xmm2, %xmm0, %xmm1, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_bac_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_bac_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubss %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfmsub.ss(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfmsub_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213sd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubsd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 
 define <2 x double> @test_x86_fma_vfmsub_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_bac_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_bac_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovapd   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubsd %xmm2, %xmm0, %xmm1, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_bac_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_bac_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubsd %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfmsub.sd(<2 x double>, <2 x double>, <2 x double>)
 
 define <4 x float> @test_x86_fma_vfmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_ps:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213ps (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ps:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213ps %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubps %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ps:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213ps (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_ps:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubps %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfmsub.ps(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfmsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_pd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213pd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_pd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213pd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubpd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_pd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213pd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_pd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubpd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfmsub.pd(<2 x double>, <2 x double>, <2 x double>)
 
 define <8 x float> @test_x86_fma_vfmsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_ps_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213ps (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_ps_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213ps %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmsubps %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_ps_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213ps (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_ps_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubps %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
   ret <8 x float> %res
 }
 declare <8 x float> @llvm.x86.fma.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x float>)
 
 define <4 x double> @test_x86_fma_vfmsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsub_pd_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsub213pd (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsub_pd_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsub213pd %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmsubpd %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsub_pd_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmsub213pd (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsub_pd_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubpd %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
   ret <4 x double> %res
 }
@@ -296,144 +362,176 @@ declare <4 x double> @llvm.x86.fma.vfmsub.pd.256(<4 x double>, <4 x double>, <4
 
 ; VFNMADD
 define <4 x float> @test_x86_fma_vfnmadd_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213ss (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213ss %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 
 define <4 x float> @test_x86_fma_vfnmadd_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_bac_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213ss (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_bac_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213ss %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmaddss %xmm2, %xmm0, %xmm1, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_bac_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_bac_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddss %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfnmadd.ss(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfnmadd_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213sd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmaddsd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 
 define <2 x double> @test_x86_fma_vfnmadd_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_bac_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_bac_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovapd   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmaddsd %xmm2, %xmm0, %xmm1, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_bac_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_bac_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddsd %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfnmadd.sd(<2 x double>, <2 x double>, <2 x double>)
 
 define <4 x float> @test_x86_fma_vfnmadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_ps:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213ps (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ps:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmaddps %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ps:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213ps (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_ps:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddps %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfnmadd.ps(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfnmadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_pd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213pd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_pd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmaddpd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_pd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213pd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_pd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddpd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfnmadd.pd(<2 x double>, <2 x double>, <2 x double>)
 
 define <8 x float> @test_x86_fma_vfnmadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_ps_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213ps (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_ps_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_ps_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213ps (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_ps_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddps %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
   ret <8 x float> %res
 }
 declare <8 x float> @llvm.x86.fma.vfnmadd.ps.256(<8 x float>, <8 x float>, <8 x float>)
 
 define <4 x double> @test_x86_fma_vfnmadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmadd_pd_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmadd213pd (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmadd_pd_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfnmaddpd %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmadd_pd_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfnmadd213pd (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmadd_pd_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmaddpd %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
   ret <4 x double> %res
 }
@@ -441,144 +539,176 @@ declare <4 x double> @llvm.x86.fma.vfnmadd.pd.256(<4 x double>, <4 x double>, <4
 
 ; VFNMSUB
 define <4 x float> @test_x86_fma_vfnmsub_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213ss (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213ss %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmsubss %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubss %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 
 define <4 x float> @test_x86_fma_vfnmsub_bac_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_bac_ss:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovaps {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213ss (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_bac_ss:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213ss %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovaps   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmsubss %xmm2, %xmm0, %xmm1, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovaps %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_bac_ss:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213ss (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_bac_ss:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubss %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float> %a1, <4 x float> %a0, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfnmsub.ss(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfnmsub_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rcx\), %xmm0|\(%rdx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213sd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmsubsd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubsd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 
 define <2 x double> @test_x86_fma_vfnmsub_bac_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_bac_sd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vmovapd {{\(%rdx\), %xmm0|\(%rcx\), %xmm1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213sd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_bac_sd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213sd %xmm2, %xmm0, %xmm1
-; CHECK-FMA-NEXT:    vmovapd   %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmsubsd %xmm2, %xmm0, %xmm1, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    vmovapd %xmm1, %xmm0
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_bac_sd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213sd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_bac_sd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubsd %xmm2, %xmm0, %xmm1, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double> %a1, <2 x double> %a0, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfnmsub.sd(<2 x double>, <2 x double>, <2 x double>)
 
 define <4 x float> @test_x86_fma_vfnmsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_ps:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213ps (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ps:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmsubps %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ps:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213ps (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_ps:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubps %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfnmsub.ps(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfnmsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_pd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213pd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_pd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfnmsubpd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_pd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213pd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_pd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubpd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfnmsub.pd(<2 x double>, <2 x double>, <2 x double>)
 
 define <8 x float> @test_x86_fma_vfnmsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_ps_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213ps (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_ps_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfnmsubps %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_ps_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213ps (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_ps_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubps %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
   ret <8 x float> %res
 }
 declare <8 x float> @llvm.x86.fma.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>)
 
 define <4 x double> @test_x86_fma_vfnmsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfnmsub_pd_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfnmsub213pd (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfnmsub_pd_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfnmsubpd %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfnmsub_pd_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfnmsub213pd (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfnmsub_pd_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfnmsubpd %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
   ret <4 x double> %res
 }
@@ -586,72 +716,88 @@ declare <4 x double> @llvm.x86.fma.vfnmsub.pd.256(<4 x double>, <4 x double>, <4
 
 ; VFMADDSUB
 define <4 x float> @test_x86_fma_vfmaddsub_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmaddsub_ps:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmaddsub213ps (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_ps:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddsubps %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_ps:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213ps (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmaddsub_ps:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddsubps %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmaddsub.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfmaddsub.ps(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfmaddsub_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmaddsub_pd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmaddsub213pd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_pd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmaddsubpd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_pd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213pd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmaddsub_pd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddsubpd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmaddsub.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfmaddsub.pd(<2 x double>, <2 x double>, <2 x double>)
 
 define <8 x float> @test_x86_fma_vfmaddsub_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmaddsub_ps_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmaddsub213ps (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_ps_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmaddsubps %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_ps_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213ps (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmaddsub_ps_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddsubps %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmaddsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
   ret <8 x float> %res
 }
 declare <8 x float> @llvm.x86.fma.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>)
 
 define <4 x double> @test_x86_fma_vfmaddsub_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmaddsub_pd_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmaddsub213pd (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmaddsub_pd_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmaddsubpd %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmaddsub_pd_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmaddsub213pd (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmaddsub_pd_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmaddsubpd %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmaddsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
   ret <4 x double> %res
 }
@@ -659,72 +805,88 @@ declare <4 x double> @llvm.x86.fma.vfmaddsub.pd.256(<4 x double>, <4 x double>,
 
 ; VFMSUBADD
 define <4 x float> @test_x86_fma_vfmsubadd_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsubadd_ps:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsubadd213ps (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_ps:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsubadd213ps %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubaddps %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_ps:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213ps (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsubadd_ps:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubaddps %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x float> @llvm.x86.fma.vfmsubadd.ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
   ret <4 x float> %res
 }
 declare <4 x float> @llvm.x86.fma.vfmsubadd.ps(<4 x float>, <4 x float>, <4 x float>)
 
 define <2 x double> @test_x86_fma_vfmsubadd_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsubadd_pd:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %xmm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsubadd213pd (%r8), %xmm1, %xmm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_pd:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsubadd213pd %xmm2, %xmm1, %xmm0
-;
-; CHECK-FMA4-NEXT: vfmsubaddpd %xmm2, %xmm1, %xmm0, %xmm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_pd:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %xmm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %xmm0
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213pd (%r8), %xmm1, %xmm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsubadd_pd:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubaddpd %xmm2, %xmm1, %xmm0, %xmm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <2 x double> @llvm.x86.fma.vfmsubadd.pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
   ret <2 x double> %res
 }
 declare <2 x double> @llvm.x86.fma.vfmsubadd.pd(<2 x double>, <2 x double>, <2 x double>)
 
 define <8 x float> @test_x86_fma_vfmsubadd_ps_256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsubadd_ps_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovaps (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsubadd213ps (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_ps_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsubadd213ps %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmsubaddps %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_ps_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovaps (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213ps (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsubadd_ps_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubaddps %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <8 x float> @llvm.x86.fma.vfmsubadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
   ret <8 x float> %res
 }
 declare <8 x float> @llvm.x86.fma.vfmsubadd.ps.256(<8 x float>, <8 x float>, <8 x float>)
 
 define <4 x double> @test_x86_fma_vfmsubadd_pd_256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) #0 {
-; CHECK-LABEL: test_x86_fma_vfmsubadd_pd_256:
-; CHECK-NEXT:  # BB#0:
-;
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vmovapd (%{{(rcx|rdx)}}), %ymm{{0|1}}
-; CHECK-FMA-WIN-NEXT: vfmsubadd213pd (%r8), %ymm1, %ymm0
-;
+; CHECK-FMA-LABEL: test_x86_fma_vfmsubadd_pd_256:
+; CHECK-FMA:       # BB#0:
 ; CHECK-FMA-NEXT:    vfmsubadd213pd %ymm2, %ymm1, %ymm0
-;
-; CHECK-FMA4-NEXT: vfmsubaddpd %ymm2, %ymm1, %ymm0, %ymm0
-;
-; CHECK-NEXT: retq
+; CHECK-FMA-NEXT:    retq
+;
+; CHECK-FMA-WIN-LABEL: test_x86_fma_vfmsubadd_pd_256:
+; CHECK-FMA-WIN:       # BB#0:
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rcx), %ymm1
+; CHECK-FMA-WIN-NEXT:    vmovapd (%rdx), %ymm0
+; CHECK-FMA-WIN-NEXT:    vfmsubadd213pd (%r8), %ymm1, %ymm0
+; CHECK-FMA-WIN-NEXT:    retq
+;
+; CHECK-FMA4-LABEL: test_x86_fma_vfmsubadd_pd_256:
+; CHECK-FMA4:       # BB#0:
+; CHECK-FMA4-NEXT:    vfmsubaddpd %ymm2, %ymm1, %ymm0, %ymm0
+; CHECK-FMA4-NEXT:    retq
   %res = call <4 x double> @llvm.x86.fma.vfmsubadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
   ret <4 x double> %res
 }

diff --git a/test/CodeGen/X86/fma4-intrinsics-x86_64-folded-load.ll b/test/CodeGen/X86/fma4-intrinsics-x86_64-folded-load.ll
index df4b48b769b03273ad8c109174b459fa9055ccda..5b86a7d7eeacf2ccadb42375f3e14ab5d52630a6 100644 (file)
--- a/test/CodeGen/X86/fma4-intrinsics-x86_64-folded-load.ll
+++ b/test/CodeGen/X86/fma4-intrinsics-x86_64-folded-load.ll
@@ -1,16 +1,23 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=corei7-avx -mattr=+fma4 | FileCheck %s
 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -mattr=+avx,-fma | FileCheck %s
 
 ; VFMADD
 define < 4 x float > @test_x86_fma_vfmadd_ss_load(< 4 x float > %a0, < 4 x float > %a1, float* %a2) {
-  ; CHECK: vfmaddss (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_ss_load:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddss (%rdi), %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load float , float *%a2
   %y = insertelement <4 x float> undef, float %x, i32 0
   %res = call < 4 x float > @llvm.x86.fma.vfmadd.ss(< 4 x float > %a0, < 4 x float > %a1, < 4 x float > %y)
   ret < 4 x float > %res
 }
 define < 4 x float > @test_x86_fma_vfmadd_ss_load2(< 4 x float > %a0, float* %a1, < 4 x float > %a2) {
-  ; CHECK: vfmaddss %{{.*}}, (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_ss_load2:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddss %xmm1, (%rdi), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load float , float *%a1
   %y = insertelement <4 x float> undef, float %x, i32 0
   %res = call < 4 x float > @llvm.x86.fma.vfmadd.ss(< 4 x float > %a0, < 4 x float > %y, < 4 x float > %a2)
@@ -20,14 +27,20 @@ define < 4 x float > @test_x86_fma_vfmadd_ss_load2(< 4 x float > %a0, float* %a1
 declare < 4 x float > @llvm.x86.fma.vfmadd.ss(< 4 x float >, < 4 x float >, < 4 x float >) nounwind readnone
 
 define < 2 x double > @test_x86_fma_vfmadd_sd_load(< 2 x double > %a0, < 2 x double > %a1, double* %a2) {
-  ; CHECK: vfmaddsd (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_sd_load:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddsd (%rdi), %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load double , double *%a2
   %y = insertelement <2 x double> undef, double %x, i32 0
   %res = call < 2 x double > @llvm.x86.fma.vfmadd.sd(< 2 x double > %a0, < 2 x double > %a1, < 2 x double > %y)
   ret < 2 x double > %res
 }
 define < 2 x double > @test_x86_fma_vfmadd_sd_load2(< 2 x double > %a0, double* %a1, < 2 x double > %a2) {
-  ; CHECK: vfmaddsd %{{.*}}, (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_sd_load2:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddsd %xmm1, (%rdi), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load double , double *%a1
   %y = insertelement <2 x double> undef, double %x, i32 0
   %res = call < 2 x double > @llvm.x86.fma.vfmadd.sd(< 2 x double > %a0, < 2 x double > %y, < 2 x double > %a2)
@@ -35,13 +48,19 @@ define < 2 x double > @test_x86_fma_vfmadd_sd_load2(< 2 x double > %a0, double*
 }
 declare < 2 x double > @llvm.x86.fma.vfmadd.sd(< 2 x double >, < 2 x double >, < 2 x double >) nounwind readnone
 define < 4 x float > @test_x86_fma_vfmadd_ps_load(< 4 x float > %a0, < 4 x float > %a1, < 4 x float >* %a2) {
-  ; CHECK: vfmaddps (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_ps_load:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddps (%rdi), %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load <4 x float>, <4 x float>* %a2
   %res = call < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float > %a0, < 4 x float > %a1, < 4 x float > %x)
   ret < 4 x float > %res
 }
 define < 4 x float > @test_x86_fma_vfmadd_ps_load2(< 4 x float > %a0, < 4 x float >* %a1, < 4 x float > %a2) {
-  ; CHECK: vfmaddps %{{.*}}, (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_ps_load2:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddps %xmm1, (%rdi), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load <4 x float>, <4 x float>* %a1
   %res = call < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float > %a0, < 4 x float > %x, < 4 x float > %a2)
   ret < 4 x float > %res
@@ -50,8 +69,11 @@ declare < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float >, < 4 x float >, < 4
 
 ; To test execution dependency
 define < 4 x float > @test_x86_fma_vfmadd_ps_load3(< 4 x float >* %a0, < 4 x float >* %a1, < 4 x float > %a2) {
-  ; CHECK: vmovaps
-  ; CHECK: vfmaddps %{{.*}}, (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_ps_load3:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vmovaps (%rdi), %xmm1
+; CHECK-NEXT:    vfmaddps %xmm0, (%rsi), %xmm1, %xmm0
+; CHECK-NEXT:    retq
   %x = load <4 x float>, <4 x float>* %a0
   %y = load <4 x float>, <4 x float>* %a1
   %res = call < 4 x float > @llvm.x86.fma.vfmadd.ps(< 4 x float > %x, < 4 x float > %y, < 4 x float > %a2)
@@ -59,13 +81,19 @@ define < 4 x float > @test_x86_fma_vfmadd_ps_load3(< 4 x float >* %a0, < 4 x flo
 }
 
 define < 2 x double > @test_x86_fma_vfmadd_pd_load(< 2 x double > %a0, < 2 x double > %a1, < 2 x double >* %a2) {
-  ; CHECK: vfmaddpd (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_pd_load:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddpd (%rdi), %xmm1, %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load <2 x double>, <2 x double>* %a2
   %res = call < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double > %a0, < 2 x double > %a1, < 2 x double > %x)
   ret < 2 x double > %res
 }
 define < 2 x double > @test_x86_fma_vfmadd_pd_load2(< 2 x double > %a0, < 2 x double >* %a1, < 2 x double > %a2) {
-  ; CHECK: vfmaddpd %{{.*}}, (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_pd_load2:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vfmaddpd %xmm1, (%rdi), %xmm0, %xmm0
+; CHECK-NEXT:    retq
   %x = load <2 x double>, <2 x double>* %a1
   %res = call < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double > %a0, < 2 x double > %x, < 2 x double > %a2)
   ret < 2 x double > %res
@@ -74,8 +102,11 @@ declare < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double >, < 2 x double >, <
 
 ; To test execution dependency
 define < 2 x double > @test_x86_fma_vfmadd_pd_load3(< 2 x double >* %a0, < 2 x double >* %a1, < 2 x double > %a2) {
-  ; CHECK: vmovapd
-  ; CHECK: vfmaddpd %{{.*}}, (%{{.*}})
+; CHECK-LABEL: test_x86_fma_vfmadd_pd_load3:
+; CHECK:       # BB#0:
+; CHECK-NEXT:    vmovapd (%rdi), %xmm1
+; CHECK-NEXT:    vfmaddpd %xmm0, (%rsi), %xmm1, %xmm0
+; CHECK-NEXT:    retq
   %x = load <2 x double>, <2 x double>* %a0
   %y = load <2 x double>, <2 x double>* %a1
   %res = call < 2 x double > @llvm.x86.fma.vfmadd.pd(< 2 x double > %x, < 2 x double > %y, < 2 x double > %a2)

diff --git a/test/CodeGen/X86/fmaddsub-combine.ll b/test/CodeGen/X86/fmaddsub-combine.ll
index f3b13cd053b486c654133e7005cc250fce4c3be9..3ce5f132277f069d42ff365070c20036a4965194 100644 (file)
--- a/test/CodeGen/X86/fmaddsub-combine.ll
+++ b/test/CodeGen/X86/fmaddsub-combine.ll
@@ -1,3 +1,4 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
 ; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+fma | FileCheck -check-prefix=FMA3 -check-prefix=FMA3_256 %s
 ; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+fma,+avx512f | FileCheck -check-prefix=FMA3 -check-prefix=FMA3_512 %s
 ; RUN: llc < %s -mtriple=x86_64-unknown-linux-gnu -mattr=+fma4 | FileCheck -check-prefix=FMA4 %s
@@ -7,13 +8,13 @@
 define <2 x double> @mul_addsub_pd128(<2 x double> %A, <2 x double> %B,  <2 x double> %C) #0 {
 ; FMA3-LABEL: mul_addsub_pd128:
 ; FMA3:       # BB#0: # %entry
-; FMA3-NEXT:  vfmaddsub213pd  %xmm2, %xmm1, %xmm0
-; FMA3-NEXT:  retq
+; FMA3-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0
+; FMA3-NEXT:    retq
 ;
 ; FMA4-LABEL: mul_addsub_pd128:
 ; FMA4:       # BB#0: # %entry
-; FMA4-NEXT:  vfmaddsubpd     %xmm2, %xmm1, %xmm0, %xmm0
-; FMA4-NEXT:  retq
+; FMA4-NEXT:    vfmaddsubpd %xmm2, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    retq
 entry:
   %AB = fmul <2 x double> %A, %B
   %Sub = fsub <2 x double> %AB, %C
@@ -25,13 +26,13 @@ entry:
 define <4 x float> @mul_addsub_ps128(<4 x float> %A, <4 x float> %B, <4 x float> %C) #0 {
 ; FMA3-LABEL: mul_addsub_ps128:
 ; FMA3:       # BB#0: # %entry
-; FMA3-NEXT:  vfmaddsub213ps  %xmm2, %xmm1, %xmm0
-; FMA3-NEXT:  retq
+; FMA3-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0
+; FMA3-NEXT:    retq
 ;
 ; FMA4-LABEL: mul_addsub_ps128:
 ; FMA4:       # BB#0: # %entry
-; FMA4-NEXT:  vfmaddsubps     %xmm2, %xmm1, %xmm0, %xmm0
-; FMA4-NEXT:  retq
+; FMA4-NEXT:    vfmaddsubps %xmm2, %xmm1, %xmm0, %xmm0
+; FMA4-NEXT:    retq
 entry:
   %AB = fmul <4 x float> %A, %B
   %Sub = fsub <4 x float> %AB, %C
@@ -43,13 +44,13 @@ entry:
 define <4 x double> @mul_addsub_pd256(<4 x double> %A, <4 x double> %B, <4 x double> %C) #0 {
 ; FMA3-LABEL: mul_addsub_pd256:
 ; FMA3:       # BB#0: # %entry
-; FMA3-NEXT:  vfmaddsub213pd  %ymm2, %ymm1, %ymm0
-; FMA3-NEXT:  retq
+; FMA3-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0
+; FMA3-NEXT:    retq
 ;
 ; FMA4-LABEL: mul_addsub_pd256:
 ; FMA4:       # BB#0: # %entry
-; FMA4-NEXT:  vfmaddsubpd     %ymm2, %ymm1, %ymm0, %ymm0
-; FMA4-NEXT:  retq
+; FMA4-NEXT:    vfmaddsubpd %ymm2, %ymm1, %ymm0, %ymm0
+; FMA4-NEXT:    retq
 entry:
   %AB = fmul <4 x double> %A, %B
   %Sub = fsub <4 x double> %AB, %C
@@ -61,13 +62,13 @@ entry:
 define <8 x float> @mul_addsub_ps256(<8 x float> %A, <8 x float> %B, <8 x float> %C) #0 {
 ; FMA3-LABEL: mul_addsub_ps256:
 ; FMA3:       # BB#0: # %entry
-; FMA3-NEXT:  vfmaddsub213ps  %ymm2, %ymm1, %ymm0
-; FMA3-NEXT:  retq
+; FMA3-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0
+; FMA3-NEXT:    retq
 ;
 ; FMA4-LABEL: mul_addsub_ps256:
 ; FMA4:       # BB#0: # %entry
-; FMA4-NEXT:  vfmaddsubps     %ymm2, %ymm1, %ymm0, %ymm0
-; FMA4-NEXT:  retq
+; FMA4-NEXT:    vfmaddsubps %ymm2, %ymm1, %ymm0, %ymm0
+; FMA4-NEXT:    retq
 entry:
   %AB = fmul <8 x float> %A, %B
   %Sub = fsub <8 x float> %AB, %C
@@ -79,20 +80,20 @@ entry:
 define <8 x double> @mul_addsub_pd512(<8 x double> %A, <8 x double> %B, <8 x double> %C) #0 {
 ; FMA3_256-LABEL: mul_addsub_pd512:
 ; FMA3_256:       # BB#0: # %entry
-; FMA3_256-NEXT:  vfmaddsub213pd  %ymm4, %ymm2, %ymm0
-; FMA3_256-NEXT:  vfmaddsub213pd  %ymm5, %ymm3, %ymm1
-; FMA3_256-NEXT:  retq
+; FMA3_256-NEXT:    vfmaddsub213pd %ymm4, %ymm2, %ymm0
+; FMA3_256-NEXT:    vfmaddsub213pd %ymm5, %ymm3, %ymm1
+; FMA3_256-NEXT:    retq
 ;
 ; FMA3_512-LABEL: mul_addsub_pd512:
 ; FMA3_512:       # BB#0: # %entry
-; FMA3_512-NEXT:  vfmaddsub213pd  %zmm2, %zmm1, %zmm0
-; FMA3_512-NEXT:  retq
+; FMA3_512-NEXT:    vfmaddsub213pd %zmm2, %zmm1, %zmm0
+; FMA3_512-NEXT:    retq
 ;
 ; FMA4-LABEL: mul_addsub_pd512:
 ; FMA4:       # BB#0: # %entry
-; FMA4-NEXT:  vfmaddsubpd     %ymm4, %ymm2, %ymm0, %ymm0
-; FMA4-NEXT:  vfmaddsubpd     %ymm5, %ymm3, %ymm1, %ymm1
-; FMA4-NEXT:  retq
+; FMA4-NEXT:    vfmaddsubpd %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfmaddsubpd %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-NEXT:    retq
 entry:
   %AB = fmul <8 x double> %A, %B
   %Sub = fsub <8 x double> %AB, %C
@@ -104,20 +105,20 @@ entry:
 define <16 x float> @mul_addsub_ps512(<16 x float> %A, <16 x float> %B, <16 x float> %C) #0 {
 ; FMA3_256-LABEL: mul_addsub_ps512:
 ; FMA3_256:       # BB#0: # %entry
-; FMA3_256-NEXT:  vfmaddsub213ps  %ymm4, %ymm2, %ymm0
-; FMA3_256-NEXT:  vfmaddsub213ps  %ymm5, %ymm3, %ymm1
-; FMA3_256-NEXT:  retq
+; FMA3_256-NEXT:    vfmaddsub213ps %ymm4, %ymm2, %ymm0
+; FMA3_256-NEXT:    vfmaddsub213ps %ymm5, %ymm3, %ymm1
+; FMA3_256-NEXT:    retq
 ;
 ; FMA3_512-LABEL: mul_addsub_ps512:
 ; FMA3_512:       # BB#0: # %entry
-; FMA3_512-NEXT:  vfmaddsub213ps  %zmm2, %zmm1, %zmm0
-; FMA3_512-NEXT:  retq
+; FMA3_512-NEXT:    vfmaddsub213ps %zmm2, %zmm1, %zmm0
+; FMA3_512-NEXT:    retq
 ;
 ; FMA4-LABEL: mul_addsub_ps512:
 ; FMA4:       # BB#0: # %entry
-; FMA4-NEXT:  vfmaddsubps     %ymm4, %ymm2, %ymm0, %ymm0
-; FMA4-NEXT:  vfmaddsubps     %ymm5, %ymm3, %ymm1, %ymm1
-; FMA4-NEXT:  retq
+; FMA4-NEXT:    vfmaddsubps %ymm4, %ymm2, %ymm0, %ymm0
+; FMA4-NEXT:    vfmaddsubps %ymm5, %ymm3, %ymm1, %ymm1
+; FMA4-NEXT:    retq
 entry:
   %AB = fmul <16 x float> %A, %B
   %Sub = fsub <16 x float> %AB, %C