[X86] Remove all of the avx512 masked packed fma intrinsics. Use llvm.fma or unmasked 512-bit intrinsics with rounding mode.

This upgrades all of the intrinsics to use fneg instructions to convert fma into fmsub/fnmsub/fnmadd/fmsubadd. And uses a select instruction for masking. This matches how clang uses the intrinsics these days. llvm-svn: 336409
author: Craig Topper <craig.topper@intel.com> 2018-07-06 03:42:09 +0000
committer: Craig Topper <craig.topper@intel.com> 2018-07-06 03:42:09 +0000
commit: 7b35585ff13d833e4736f8169c9ea200dd011aeb (patch)
tree: 98607c17c8067d09eea336be755773ba4e99a5da /llvm/test
parent: 4ea8949697828da0f702024b9fb2b23a19c69924 (diff)
download: bcm5719-llvm-7b35585ff13d833e4736f8169c9ea200dd011aeb.tar.gz
bcm5719-llvm-7b35585ff13d833e4736f8169c9ea200dd011aeb.zip
3 files changed, 1967 insertions, 356 deletions
diff --git a/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll b/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll
index f758aebaea5..ee30370ba95 100644
--- a/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll
@@ -11,10 +11,10 @@ define <16 x float> @test_x86_vfnmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <
 ; CHECK-NEXT:    vfnmadd213ps %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xac,0xc2]
 ; CHECK-NEXT:    # zmm0 = -(zmm1 * zmm0) + zmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %a2)
+  ret <16 x float> %2
 }
-declare <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
 
 define <16 x float> @test_mask_vfnmadd_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
 ; X86-LABEL: test_mask_vfnmadd_ps:
@@ -30,8 +30,11 @@ define <16 x float> @test_mask_vfnmadd_ps(<16 x float> %a0, <16 x float> %a1, <1
 ; X64-NEXT:    vfnmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9c,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) + zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %a2)
+  %3 = bitcast i16 %mask to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %a0
+  ret <16 x float> %4
 }
 
 define <8 x double> @test_x86_vfnmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
@@ -40,10 +43,10 @@ define <8 x double> @test_x86_vfnmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <
 ; CHECK-NEXT:    vfnmadd213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xac,0xc2]
 ; CHECK-NEXT:    # zmm0 = -(zmm1 * zmm0) + zmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %a2)
+  ret <8 x double> %2
 }
-declare <8 x double> @llvm.x86.avx512.mask.vfnmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
 
 define <8 x double> @test_mask_vfnmadd_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfnmadd_pd:
@@ -60,8 +63,11 @@ define <8 x double> @test_mask_vfnmadd_pd(<8 x double> %a0, <8 x double> %a1, <8
 ; X64-NEXT:    vfnmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9c,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) + zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %a2)
+  %3 = bitcast i8 %mask to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %a0
+  ret <8 x double> %4
 }
 
 define <16 x float> @test_x86_vfnmsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
@@ -70,10 +76,11 @@ define <16 x float> @test_x86_vfnmsubps_z(<16 x float> %a0, <16 x float> %a1, <1
 ; CHECK-NEXT:    vfnmsub213ps %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xae,0xc2]
 ; CHECK-NEXT:    # zmm0 = -(zmm1 * zmm0) - zmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 -1, i32 4) nounwind
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
+  %3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %2)
+  ret <16 x float> %3
 }
-declare <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
 
 define <16 x float> @test_mask_vfnmsub_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
 ; X86-LABEL: test_mask_vfnmsub_ps:
@@ -89,8 +96,12 @@ define <16 x float> @test_mask_vfnmsub_ps(<16 x float> %a0, <16 x float> %a1, <1
 ; X64-NEXT:    vfnmsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9e,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) - zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
+  %3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %a0, <16 x float> %1, <16 x float> %2)
+  %4 = bitcast i16 %mask to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %a0
+  ret <16 x float> %5
 }
 
 define <8 x double> @test_x86_vfnmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
@@ -99,10 +110,11 @@ define <8 x double> @test_x86_vfnmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8
 ; CHECK-NEXT:    vfnmsub213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xae,0xc2]
 ; CHECK-NEXT:    # zmm0 = -(zmm1 * zmm0) - zmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
+  ret <8 x double> %3
 }
-declare <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
 
 define <8 x double> @test_mask_vfnmsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfnmsub_pd:
@@ -119,8 +131,12 @@ define <8 x double> @test_mask_vfnmsub_pd(<8 x double> %a0, <8 x double> %a1, <8
 ; X64-NEXT:    vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) - zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
+  ret <8 x double> %5
 }
 
 define <16 x float> @test_x86_vfmaddsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
@@ -129,7 +145,7 @@ define <16 x float> @test_x86_vfmaddsubps_z(<16 x float> %a0, <16 x float> %a1,
 ; CHECK-NEXT:    vfmaddsub213ps %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x75,0x48,0xa6,0xc2]
 ; CHECK-NEXT:    # zmm0 = (zmm1 * zmm0) +/- zmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) nounwind
+  %res = call <16 x float> @llvm.x86.avx512.vfmaddsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i32 4) #2
   ret <16 x float> %res
 }
 
@@ -161,7 +177,7 @@ define <8 x double> @test_x86_vfmaddsubpd_z(<8 x double> %a0, <8 x double> %a1,
 ; CHECK-NEXT:    vfmaddsub213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xa6,0xc2]
 ; CHECK-NEXT:    # zmm0 = (zmm1 * zmm0) +/- zmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) #2
   ret <8 x double> %res
 }
 declare <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i32) nounwind readnone
@@ -181,7 +197,7 @@ define <8 x double> @test_mask_vfmaddsub_pd(<8 x double> %a0, <8 x double> %a1,
 ; X64-NEXT:    vfmaddsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x96,0xc1]
 ; X64-NEXT:    # zmm0 = (zmm0 * zmm1) +/- zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) nounwind
+  %res = call <8 x double> @llvm.x86.avx512.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i32 4) #2
   %bc = bitcast i8 %mask to <8 x i1>
   %sel = select <8 x i1> %bc, <8 x double> %res, <8 x double> %a0
   ret <8 x double> %sel
@@ -208,8 +224,6 @@ define <8 x double>@test_int_x86_avx512_mask_vfmaddsub_pd_512(<8 x double> %x0,
   ret <8 x double> %sel
 }
 
-declare <8 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
-
 define <8 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_512:
 ; X86:       # %bb.0:
@@ -227,12 +241,15 @@ define <8 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_512(<8 x double> %x0,
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) +/- zmm2
 ; X64-NEXT:    vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %2)
+  %4 = shufflevector <8 x double> %3, <8 x double> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %6 = select <8 x i1> %5, <8 x double> %4, <8 x double> %x2
+  ret <8 x double> %6
 }
 
-declare <8 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
-
 define <8 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_512:
 ; X86:       # %bb.0:
@@ -248,8 +265,13 @@ define <8 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_512(<8 x double> %x0,
 ; X64-NEXT:    vfmaddsub213pd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0xa6,0xc2]
 ; X64-NEXT:    # zmm0 = (zmm1 * zmm0) +/- zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %2)
+  %4 = shufflevector <8 x double> %3, <8 x double> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %6 = select <8 x i1> %5, <8 x double> %4, <8 x double> zeroinitializer
+  ret <8 x double> %6
 }
 
 define <16 x float>@test_int_x86_avx512_mask_vfmaddsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
@@ -272,8 +294,6 @@ define <16 x float>@test_int_x86_avx512_mask_vfmaddsub_ps_512(<16 x float> %x0,
   ret <16 x float> %sel
 }
 
-declare <16 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-
 define <16 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_512:
 ; X86:       # %bb.0:
@@ -290,12 +310,15 @@ define <16 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_512(<16 x float> %x0,
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) +/- zmm2
 ; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
+  %2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %2)
+  %4 = shufflevector <16 x float> %3, <16 x float> %1, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %5 = bitcast i16 %x3 to <16 x i1>
+  %6 = select <16 x i1> %5, <16 x float> %4, <16 x float> %x2
+  ret <16 x float> %6
 }
 
-declare <16 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-
 define <16 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_512:
 ; X86:       # %bb.0:
@@ -310,12 +333,15 @@ define <16 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_512(<16 x float> %x0,
 ; X64-NEXT:    vfmaddsub213ps %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0xa6,0xc2]
 ; X64-NEXT:    # zmm0 = (zmm1 * zmm0) +/- zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
+  %2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %2)
+  %4 = shufflevector <16 x float> %3, <16 x float> %1, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %5 = bitcast i16 %x3 to <16 x i1>
+  %6 = select <16 x i1> %5, <16 x float> %4, <16 x float> zeroinitializer
+  ret <16 x float> %6
 }
 
-declare <8 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
-
 define <8 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_512:
 ; X86:       # %bb.0:
@@ -333,12 +359,15 @@ define <8 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_512(<8 x double> %x0,
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) -/+ zmm2
 ; X64-NEXT:    vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %2)
+  %4 = shufflevector <8 x double> %1, <8 x double> %3, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %6 = select <8 x i1> %5, <8 x double> %4, <8 x double> %x2
+  ret <8 x double> %6
 }
 
-declare <16 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-
 define <16 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_512:
 ; X86:       # %bb.0:
@@ -355,8 +384,13 @@ define <16 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_512(<16 x float> %x0,
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) -/+ zmm2
 ; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
+  %2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %2)
+  %4 = shufflevector <16 x float> %1, <16 x float> %3, <16 x i32> <i32 0, i32 17, i32 2, i32 19, i32 4, i32 21, i32 6, i32 23, i32 8, i32 25, i32 10, i32 27, i32 12, i32 29, i32 14, i32 31>
+  %5 = bitcast i16 %x3 to <16 x i1>
+  %6 = select <16 x i1> %5, <16 x float> %4, <16 x float> %x2
+  ret <16 x float> %6
 }
 
 define <16 x float> @test_mask_round_vfmadd512_ps_rrb_rne(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
@@ -497,8 +531,6 @@ define <16 x float> @test_mask_round_vfmadd512_ps_rrbz_current(<16 x float> %a0,
   ret <16 x float> %res
 }
 
-declare <8 x double> @llvm.x86.avx512.mask3.vfmsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
-
 define <8 x double>@test_int_x86_avx512_mask3_vfmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_512:
 ; X86:       # %bb.0:
@@ -516,12 +548,13 @@ define <8 x double>@test_int_x86_avx512_mask3_vfmsub_pd_512(<8 x double> %x0, <8
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) - zmm2
 ; X64-NEXT:    vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask3.vfmsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %1)
+  %3 = bitcast i8 %x3 to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %x2
+  ret <8 x double> %4
 }
 
-declare <16 x float> @llvm.x86.avx512.mask3.vfmsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-
 define <16 x float>@test_int_x86_avx512_mask3_vfmsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_512:
 ; X86:       # %bb.0:
@@ -538,8 +571,11 @@ define <16 x float>@test_int_x86_avx512_mask3_vfmsub_ps_512(<16 x float> %x0, <1
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) - zmm2
 ; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask3.vfmsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %1)
+  %3 = bitcast i16 %x3 to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %x2
+  ret <16 x float> %4
 }
 
 define <8 x double> @test_mask_round_vfmadd512_pd_rrb_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
@@ -706,8 +742,6 @@ define <8 x double>@test_int_x86_avx512_mask_vfmadd_pd_512(<8 x double> %x0, <8
   ret <8 x double> %sel
 }
 
-declare <8 x double> @llvm.x86.avx512.mask3.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
-
 define <8 x double>@test_int_x86_avx512_mask3_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_512:
 ; X86:       # %bb.0:
@@ -725,12 +759,12 @@ define <8 x double>@test_int_x86_avx512_mask3_vfmadd_pd_512(<8 x double> %x0, <8
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) + zmm2
 ; X64-NEXT:    vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask3.vfmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x double> %1, <8 x double> %x2
+  ret <8 x double> %3
 }
 
-declare <8 x double> @llvm.x86.avx512.maskz.vfmadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
-
 define <8 x double>@test_int_x86_avx512_maskz_vfmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_512:
 ; X86:       # %bb.0:
@@ -746,8 +780,10 @@ define <8 x double>@test_int_x86_avx512_maskz_vfmadd_pd_512(<8 x double> %x0, <8
 ; X64-NEXT:    vfmadd213pd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0xa8,0xc2]
 ; X64-NEXT:    # zmm0 = (zmm1 * zmm0) + zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.maskz.vfmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x double> %1, <8 x double> zeroinitializer
+  ret <8 x double> %3
 }
 
 define <16 x float>@test_int_x86_avx512_mask_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
@@ -770,8 +806,6 @@ define <16 x float>@test_int_x86_avx512_mask_vfmadd_ps_512(<16 x float> %x0, <16
   ret <16 x float> %sel
 }
 
-declare <16 x float> @llvm.x86.avx512.mask3.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-
 define <16 x float>@test_int_x86_avx512_mask3_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_512:
 ; X86:       # %bb.0:
@@ -788,13 +822,13 @@ define <16 x float>@test_int_x86_avx512_mask3_vfmadd_ps_512(<16 x float> %x0, <1
 ; X64-NEXT:    # zmm2 = (zmm0 * zmm1) + zmm2
 ; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask3.vfmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
+  %2 = bitcast i16 %x3 to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x float> %1, <16 x float> %x2
+  ret <16 x float> %3
 }
 
-declare <16 x float> @llvm.x86.avx512.maskz.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-
-define <16 x float>@test_int_x86_avx512_maskz_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
+define <16 x float> @test_int_x86_avx512_maskz_vfmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_512:
 ; X86:       # %bb.0:
 ; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
@@ -808,11 +842,12 @@ define <16 x float>@test_int_x86_avx512_maskz_vfmadd_ps_512(<16 x float> %x0, <1
 ; X64-NEXT:    vfmadd213ps %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xc9,0xa8,0xc2]
 ; X64-NEXT:    # zmm0 = (zmm1 * zmm0) + zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.maskz.vfmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2)
+  %2 = bitcast i16 %x3 to <16 x i1>
+  %3 = select <16 x i1> %2, <16 x float> %1, <16 x float> zeroinitializer
+  ret <16 x float> %3
 }
 
-
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
 ; X86-LABEL: test_mask_round_vfnmsub512_pd_rrb_rne:
 ; X86:       # %bb.0:
@@ -826,8 +861,12 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rne(<8 x double> %a0, <8
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
 ; X64-NEXT:    vfnmsub132pd {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x19,0x9e,0xc1]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 0) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 0)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
+  ret <8 x double> %5
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
@@ -843,8 +882,12 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtn(<8 x double> %a0, <8
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
 ; X64-NEXT:    vfnmsub132pd {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x39,0x9e,0xc1]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 1) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 1)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
+  ret <8 x double> %5
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
@@ -860,8 +903,12 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtp(<8 x double> %a0, <8
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
 ; X64-NEXT:    vfnmsub132pd {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x59,0x9e,0xc1]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 2) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 2)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
+  ret <8 x double> %5
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
@@ -877,8 +924,12 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_rtz(<8 x double> %a0, <8
 ; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
 ; X64-NEXT:    vfnmsub132pd {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x79,0x9e,0xc1]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 3) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 3)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
+  ret <8 x double> %5
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
@@ -896,8 +947,12 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrb_current(<8 x double> %a0,
 ; X64-NEXT:    vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) - zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %a0
+  ret <8 x double> %5
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rne(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
@@ -905,8 +960,10 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rne(<8 x double> %a0, <8
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213pd {rn-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x18,0xae,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 0) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 0)
+  ret <8 x double> %3
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtn(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
@@ -914,8 +971,10 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtn(<8 x double> %a0, <8
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213pd {rd-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x38,0xae,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 1) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 1)
+  ret <8 x double> %3
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtp(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
@@ -923,8 +982,10 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtp(<8 x double> %a0, <8
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213pd {ru-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x58,0xae,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 2) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 2)
+  ret <8 x double> %3
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtz(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
@@ -932,8 +993,10 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_rtz(<8 x double> %a0, <8
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213pd {rz-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x78,0xae,0xc2]
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 3) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.x86.avx512.vfmadd.pd.512(<8 x double> %a0, <8 x double> %1, <8 x double> %2, i32 3)
+  ret <8 x double> %3
 }
 
 define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_current(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
@@ -942,8 +1005,10 @@ define <8 x double> @test_mask_round_vfnmsub512_pd_rrbz_current(<8 x double> %a0
 ; CHECK-NEXT:    vfnmsub213pd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0xf5,0x48,0xae,0xc2]
 ; CHECK-NEXT:    # zmm0 = -(zmm1 * zmm0) - zmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 -1, i32 4) nounwind
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %a0, <8 x double> %1, <8 x double> %2)
+  ret <8 x double> %3
 }
 
 define <8 x double>@test_int_x86_avx512_mask_vfnmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
@@ -961,12 +1026,14 @@ define <8 x double>@test_int_x86_avx512_mask_vfnmsub_pd_512(<8 x double> %x0, <8
 ; X64-NEXT:    vfnmsub132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9e,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) - zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x1
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %1, <8 x double> %2)
+  %4 = bitcast i8 %x3 to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %x0
+  ret <8 x double> %5
 }
 
-declare <8 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
-
 define <8 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_512:
 ; X86:       # %bb.0:
@@ -984,8 +1051,12 @@ define <8 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_512(<8 x double> %x0, <
 ; X64-NEXT:    # zmm2 = -(zmm0 * zmm1) - zmm2
 ; X64-NEXT:    vmovapd %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x0
+  %2 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <8 x double> @llvm.fma.v8f64(<8 x double> %1, <8 x double> %x1, <8 x double> %2)
+  %4 = bitcast i8 %x3 to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x double> %3, <8 x double> %x2
+  ret <8 x double> %5
 }
 
 define <16 x float>@test_int_x86_avx512_mask_vfnmsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
@@ -1002,12 +1073,14 @@ define <16 x float>@test_int_x86_avx512_mask_vfnmsub_ps_512(<16 x float> %x0, <1
 ; X64-NEXT:    vfnmsub132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9e,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) - zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfnmsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x1
+  %2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %1, <16 x float> %2)
+  %4 = bitcast i16 %x3 to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %x0
+  ret <16 x float> %5
 }
 
-declare <16 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
-
 define <16 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
 ; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_512:
 ; X86:       # %bb.0:
@@ -1024,8 +1097,12 @@ define <16 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_512(<16 x float> %x0, <
 ; X64-NEXT:    # zmm2 = -(zmm0 * zmm1) - zmm2
 ; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x0
+  %2 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <16 x float> @llvm.fma.v16f32(<16 x float> %1, <16 x float> %x1, <16 x float> %2)
+  %4 = bitcast i16 %x3 to <16 x i1>
+  %5 = select <16 x i1> %4, <16 x float> %3, <16 x float> %x2
+  ret <16 x float> %5
 }
 
 define <8 x double>@test_int_x86_avx512_mask_vfnmadd_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3){
@@ -1043,8 +1120,11 @@ define <8 x double>@test_int_x86_avx512_mask_vfnmadd_pd_512(<8 x double> %x0, <8
 ; X64-NEXT:    vfnmadd132pd %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x49,0x9c,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) + zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x double> @llvm.x86.avx512.mask.vfnmadd.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 4)
-  ret <8 x double> %res
+  %1 = fsub <8 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x1
+  %2 = call <8 x double> @llvm.fma.v8f64(<8 x double> %x0, <8 x double> %1, <8 x double> %x2)
+  %3 = bitcast i8 %x3 to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x double> %2, <8 x double> %x0
+  ret <8 x double> %4
 }
 
 define <16 x float>@test_int_x86_avx512_mask_vfnmadd_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3){
@@ -1061,6 +1141,12 @@ define <16 x float>@test_int_x86_avx512_mask_vfnmadd_ps_512(<16 x float> %x0, <1
 ; X64-NEXT:    vfnmadd132ps %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x49,0x9c,0xc1]
 ; X64-NEXT:    # zmm0 = -(zmm0 * zmm1) + zmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <16 x float> @llvm.x86.avx512.mask.vfnmadd.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 4)
-  ret <16 x float> %res
+  %1 = fsub <16 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x1
+  %2 = call <16 x float> @llvm.fma.v16f32(<16 x float> %x0, <16 x float> %1, <16 x float> %x2)
+  %3 = bitcast i16 %x3 to <16 x i1>
+  %4 = select <16 x i1> %3, <16 x float> %2, <16 x float> %x0
+  ret <16 x float> %4
 }
+
+declare <16 x float> @llvm.fma.v16f32(<16 x float>, <16 x float>, <16 x float>)
+declare <8 x double> @llvm.fma.v8f64(<8 x double>, <8 x double>, <8 x double>)
diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
index 76975f29f88..62fc47a97d6 100644
--- a/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics-upgrade.ll
@@ -12603,3 +12603,1380 @@ define <4 x i64>@test_int_x86_avx512_mask_pror_q_256(<4 x i64> %x0, i32 %x1, <4
   %res4 = add <4 x i64> %res3, %res2
   ret <4 x i64> %res4
 }
+
+declare <8 x float> @llvm.x86.avx512.mask.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_vfmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
+; CHECK-LABEL: test_vfmadd256_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
+; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 -1) nounwind
+  ret <8 x float> %res
+}
+
+define <8 x float> @test_mask_vfmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd256_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x98,0xc1]
+; X86-NEXT:    # ymm0 = (ymm0 * ymm1) + ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd256_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x98,0xc1]
+; X64-NEXT:    # ymm0 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) nounwind
+  ret <8 x float> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_vfmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
+; CHECK-LABEL: test_vfmadd128_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
+; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd128_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x98,0xc1]
+; X86-NEXT:    # xmm0 = (xmm0 * xmm1) + xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x98,0xc1]
+; X64-NEXT:    # xmm0 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
+  ret <4 x float> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double> @test_fmadd256_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
+; CHECK-LABEL: test_fmadd256_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
+; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %c, i8 -1)
+  ret <4 x double> %res
+}
+
+define <4 x double> @test_mask_fmadd256_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c, i8 %mask) {
+; X86-LABEL: test_mask_fmadd256_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x98,0xc1]
+; X86-NEXT:    # ymm0 = (ymm0 * ymm1) + ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_fmadd256_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x98,0xc1]
+; X64-NEXT:    # ymm0 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %c, i8 %mask)
+  ret <4 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double> @test_fmadd128_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
+; CHECK-LABEL: test_fmadd128_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
+; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 -1)
+  ret <2 x double> %res
+}
+
+define <2 x double> @test_mask_fmadd128_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
+; X86-LABEL: test_mask_fmadd128_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x98,0xc1]
+; X86-NEXT:    # xmm0 = (xmm0 * xmm1) + xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_fmadd128_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x98,0xc1]
+; X64-NEXT:    # xmm0 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask)
+  ret <2 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double>@test_int_x86_avx512_mask3_vfmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xb8,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) + xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xb8,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
+  ret <2 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.maskz.vfmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double>@test_int_x86_avx512_maskz_vfmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xa8,0xc2]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xa8,0xc2]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.maskz.vfmadd.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
+  ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask3.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double>@test_int_x86_avx512_mask3_vfmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xb8,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) + ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xb8,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmadd.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
+  ret <4 x double> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.maskz.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double>@test_int_x86_avx512_maskz_vfmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xa8,0xc2]
+; X86-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xa8,0xc2]
+; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.maskz.vfmadd.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
+  ret <4 x double> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_mask3_vfmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xb8,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) + xmm2
+; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xb8,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) + xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
+  ret <4 x float> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.maskz.vfmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_maskz_vfmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0xa8,0xc2]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0xa8,0xc2]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
+  ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask3.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_mask3_vfmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xb8,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) + ymm2
+; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xb8,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) + ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmadd.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
+  ret <8 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.maskz.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_maskz_vfmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0xa8,0xc2]
+; X86-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0xa8,0xc2]
+; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.maskz.vfmadd.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
+  ret <8 x float> %res
+}
+
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double>@test_int_x86_avx512_mask3_vfmsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsub231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xba,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) - xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsub231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xba,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
+  ret <2 x double> %res
+}
+
+
+declare <4 x double> @llvm.x86.avx512.mask3.vfmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double>@test_int_x86_avx512_mask3_vfmsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsub231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xba,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) - ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsub231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xba,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
+  ret <4 x double> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_mask3_vfmsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsub231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xba,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) - xmm2
+; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsub231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xba,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
+  ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask3.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_mask3_vfmsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsub231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xba,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) - ymm2
+; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsub231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xba,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
+  ret <8 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask.vfnmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_vfnmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
+; CHECK-LABEL: test_vfnmadd256_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xac,0xc2]
+; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 -1) nounwind
+  ret <8 x float> %res
+}
+
+define <8 x float> @test_mask_vfnmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmadd256_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmadd132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x9c,0xc1]
+; X86-NEXT:    # ymm0 = -(ymm0 * ymm1) + ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmadd256_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmadd132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x9c,0xc1]
+; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) nounwind
+  ret <8 x float> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.vfnmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_vfnmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
+; CHECK-LABEL: test_vfnmadd128_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xac,0xc2]
+; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfnmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmadd128_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmadd132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x9c,0xc1]
+; X86-NEXT:    # xmm0 = -(xmm0 * xmm1) + xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmadd128_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmadd132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x9c,0xc1]
+; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
+  ret <4 x float> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask.vfnmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_vfnmadd256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
+; CHECK-LABEL: test_vfnmadd256_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
+; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
+  ret <4 x double> %res
+}
+
+define <4 x double> @test_mask_vfnmadd256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmadd256_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmadd132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x9c,0xc1]
+; X86-NEXT:    # ymm0 = -(ymm0 * ymm1) + ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmadd256_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmadd132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x9c,0xc1]
+; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) + ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
+  ret <4 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.vfnmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_vfnmadd128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
+; CHECK-LABEL: test_vfnmadd128_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
+; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
+  ret <2 x double> %res
+}
+
+define <2 x double> @test_mask_vfnmadd128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmadd128_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmadd132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x9c,0xc1]
+; X86-NEXT:    # xmm0 = -(xmm0 * xmm1) + xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmadd128_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmadd132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x9c,0xc1]
+; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) + xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
+  ret <2 x double> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_vfnmsub256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
+; CHECK-LABEL: test_vfnmsub256_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xae,0xc2]
+; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 -1) nounwind
+  ret <8 x float> %res
+}
+
+define <8 x float> @test_mask_vfnmsub256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmsub256_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x9e,0xc1]
+; X86-NEXT:    # ymm0 = -(ymm0 * ymm1) - ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmsub256_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x9e,0xc1]
+; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) nounwind
+  ret <8 x float> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.vfnmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_vfnmsub128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
+; CHECK-LABEL: test_vfnmsub128_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xae,0xc2]
+; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmsub.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfnmsub128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmsub128_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x9e,0xc1]
+; X86-NEXT:    # xmm0 = -(xmm0 * xmm1) - xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmsub128_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x9e,0xc1]
+; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmsub.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
+  ret <4 x float> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask.vfnmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_vfnmsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
+; CHECK-LABEL: test_vfnmsub256_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
+; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
+  ret <4 x double> %res
+}
+
+define <4 x double> @test_mask_vfnmsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmsub256_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x9e,0xc1]
+; X86-NEXT:    # ymm0 = -(ymm0 * ymm1) - ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmsub256_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x9e,0xc1]
+; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
+  ret <4 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.vfnmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_vfnmsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
+; CHECK-LABEL: test_vfnmsub128_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
+; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
+  ret <2 x double> %res
+}
+
+define <2 x double> @test_mask_vfnmsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfnmsub128_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x9e,0xc1]
+; X86-NEXT:    # xmm0 = -(xmm0 * xmm1) - xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfnmsub128_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x9e,0xc1]
+; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
+  ret <2 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xbe,0xd1]
+; X86-NEXT:    # xmm2 = -(xmm0 * xmm1) - xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xbe,0xd1]
+; X64-NEXT:    # xmm2 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
+  ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xbe,0xd1]
+; X86-NEXT:    # ymm2 = -(ymm0 * ymm1) - ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xbe,0xd1]
+; X64-NEXT:    # ymm2 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
+  ret <4 x double> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xbe,0xd1]
+; X86-NEXT:    # xmm2 = -(xmm0 * xmm1) - xmm2
+; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xbe,0xd1]
+; X64-NEXT:    # xmm2 = -(xmm0 * xmm1) - xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
+  ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfnmsub231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xbe,0xd1]
+; X86-NEXT:    # ymm2 = -(ymm0 * ymm1) - ymm2
+; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfnmsub231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xbe,0xd1]
+; X64-NEXT:    # ymm2 = -(ymm0 * ymm1) - ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
+  ret <8 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_fmaddsub256_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
+; CHECK-LABEL: test_fmaddsub256_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
+; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %c, i8 -1)
+  ret <8 x float> %res
+}
+
+define <8 x float> @test_mask_fmaddsub256_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c, i8 %mask) {
+; X86-LABEL: test_mask_fmaddsub256_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x96,0xc1]
+; X86-NEXT:    # ymm0 = (ymm0 * ymm1) +/- ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_fmaddsub256_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x96,0xc1]
+; X64-NEXT:    # ymm0 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %c, i8 %mask)
+  ret <8 x float> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_fmaddsub128_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
+; CHECK-LABEL: test_fmaddsub128_ps:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
+; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.128(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 -1)
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_fmaddsub128_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
+; X86-LABEL: test_mask_fmaddsub128_ps:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x96,0xc1]
+; X86-NEXT:    # xmm0 = (xmm0 * xmm1) +/- xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_fmaddsub128_ps:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x96,0xc1]
+; X64-NEXT:    # xmm0 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.128(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask)
+  ret <4 x float> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_vfmaddsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
+; CHECK-LABEL: test_vfmaddsub256_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
+; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
+  ret <4 x double> %res
+}
+
+define <4 x double> @test_mask_vfmaddsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmaddsub256_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x96,0xc1]
+; X86-NEXT:    # ymm0 = (ymm0 * ymm1) +/- ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmaddsub256_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x96,0xc1]
+; X64-NEXT:    # ymm0 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
+  ret <4 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_vfmaddsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
+; CHECK-LABEL: test_vfmaddsub128_pd:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
+; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
+  ret <2 x double> %res
+}
+
+define <2 x double> @test_mask_vfmaddsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmaddsub128_pd:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x96,0xc1]
+; X86-NEXT:    # xmm0 = (xmm0 * xmm1) +/- xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmaddsub128_pd:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x96,0xc1]
+; X64-NEXT:    # xmm0 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
+  ret <2 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xb6,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) +/- xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xb6,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
+  ret <2 x double> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xa6,0xc2]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xa6,0xc2]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
+  ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xb6,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) +/- ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xb6,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
+  ret <4 x double> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xa6,0xc2]
+; X86-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xa6,0xc2]
+; X64-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
+  ret <4 x double> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xb6,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) +/- xmm2
+; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xb6,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) +/- xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
+  ret <4 x float> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0xa6,0xc2]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0xa6,0xc2]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
+  ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xb6,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) +/- ymm2
+; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xb6,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) +/- ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
+  ret <8 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0xa6,0xc2]
+; X86-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0xa6,0xc2]
+; X64-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
+  ret <8 x float> %res
+}
+
+declare <2 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsubadd231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xb7,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X86-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsubadd231pd %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x09,0xb7,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
+  ret <2 x double> %res
+}
+
+declare <4 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsubadd231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xb7,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X86-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsubadd231pd %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x29,0xb7,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
+  ret <4 x double> %res
+}
+
+declare <4 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
+
+define <4 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_128:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsubadd231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xb7,0xd1]
+; X86-NEXT:    # xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X86-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_128:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsubadd231ps %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x09,0xb7,0xd1]
+; X64-NEXT:    # xmm2 = (xmm0 * xmm1) -/+ xmm2
+; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
+  ret <4 x float> %res
+}
+
+declare <8 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
+
+define <8 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
+; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_256:
+; X86:       # %bb.0:
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x04]
+; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
+; X86-NEXT:    vfmsubadd231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xb7,0xd1]
+; X86-NEXT:    # ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X86-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_256:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %edi, %k1 # encoding: [0xc5,0xf8,0x92,0xcf]
+; X64-NEXT:    vfmsubadd231ps %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x29,0xb7,0xd1]
+; X64-NEXT:    # ymm2 = (ymm0 * ymm1) -/+ ymm2
+; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
+; X64-NEXT:    retq # encoding: [0xc3]
+  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
+  ret <8 x float> %res
+}
+
+
+define <4 x float> @test_mask_vfmadd128_ps_rmk(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmk:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
+; X86-NEXT:    vfmadd213ps (%eax), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmk:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
+; X64-NEXT:    vfmadd213ps (%rdi), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <4 x float>, <4 x float>* %ptr_a2
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_rmka(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmka:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
+; X86-NEXT:    vfmadd213ps (%eax), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmka:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
+; X64-NEXT:    vfmadd213ps (%rdi), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <4 x float>, <4 x float>* %ptr_a2, align 8
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_rmkz(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmkz:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vfmadd213ps (%eax), %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmkz:
+; X64:       # %bb.0:
+; X64-NEXT:    vfmadd213ps (%rdi), %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <4 x float>, <4 x float>* %ptr_a2
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_rmkza(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmkza:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vfmadd213ps (%eax), %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmkza:
+; X64:       # %bb.0:
+; X64-NEXT:    vfmadd213ps (%rdi), %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <4 x float>, <4 x float>* %ptr_a2, align 4
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_rmb(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmb:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
+; X86-NEXT:    vfmadd213ps (%eax){1to4}, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x19,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmb:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
+; X64-NEXT:    vfmadd213ps (%rdi){1to4}, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %q = load float, float* %ptr_a2
+  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
+  %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
+  %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
+  %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 %mask) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_rmba(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmba:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
+; X86-NEXT:    vfmadd213ps (%eax){1to4}, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x19,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmba:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
+; X64-NEXT:    vfmadd213ps (%rdi){1to4}, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %q = load float, float* %ptr_a2, align 4
+  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
+  %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
+  %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
+  %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 %mask) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_rmbz(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmbz:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vfmadd213ps (%eax){1to4}, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x18,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmbz:
+; X64:       # %bb.0:
+; X64-NEXT:    vfmadd213ps (%rdi){1to4}, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %q = load float, float* %ptr_a2
+  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
+  %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
+  %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
+  %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 -1) nounwind
+  ret <4 x float> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_rmbza(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2) {
+; X86-LABEL: test_mask_vfmadd128_ps_rmbza:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vfmadd213ps (%eax){1to4}, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x18,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_ps_rmbza:
+; X64:       # %bb.0:
+; X64-NEXT:    vfmadd213ps (%rdi){1to4}, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %q = load float, float* %ptr_a2, align 4
+  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
+  %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
+  %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
+  %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
+  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 -1) nounwind
+  ret <4 x float> %res
+}
+
+define <2 x double> @test_mask_vfmadd128_pd_rmk(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd128_pd_rmk:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
+; X86-NEXT:    vfmadd213pd (%eax), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_pd_rmk:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
+; X64-NEXT:    vfmadd213pd (%rdi), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <2 x double>, <2 x double>* %ptr_a2
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
+  ret <2 x double> %res
+}
+
+define <2 x double> @test_mask_vfmadd128_pd_rmkz(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2) {
+; X86-LABEL: test_mask_vfmadd128_pd_rmkz:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vfmadd213pd (%eax), %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0x00]
+; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd128_pd_rmkz:
+; X64:       # %bb.0:
+; X64-NEXT:    vfmadd213pd (%rdi), %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0x07]
+; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <2 x double>, <2 x double>* %ptr_a2
+  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
+  ret <2 x double> %res
+}
+
+define <4 x double> @test_mask_vfmadd256_pd_rmk(<4 x double> %a0, <4 x double> %a1, <4 x double>* %ptr_a2, i8 %mask) {
+; X86-LABEL: test_mask_vfmadd256_pd_rmk:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
+; X86-NEXT:    vfmadd213pd (%eax), %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xa8,0x00]
+; X86-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd256_pd_rmk:
+; X64:       # %bb.0:
+; X64-NEXT:    kmovw %esi, %k1 # encoding: [0xc5,0xf8,0x92,0xce]
+; X64-NEXT:    vfmadd213pd (%rdi), %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xa8,0x07]
+; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <4 x double>, <4 x double>* %ptr_a2
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
+  ret <4 x double> %res
+}
+
+define <4 x double> @test_mask_vfmadd256_pd_rmkz(<4 x double> %a0, <4 x double> %a1, <4 x double>* %ptr_a2) {
+; X86-LABEL: test_mask_vfmadd256_pd_rmkz:
+; X86:       # %bb.0:
+; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    vfmadd213pd (%eax), %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0x00]
+; X86-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
+; X86-NEXT:    retl # encoding: [0xc3]
+;
+; X64-LABEL: test_mask_vfmadd256_pd_rmkz:
+; X64:       # %bb.0:
+; X64-NEXT:    vfmadd213pd (%rdi), %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0x07]
+; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
+; X64-NEXT:    retq # encoding: [0xc3]
+  %a2 = load <4 x double>, <4 x double>* %ptr_a2
+  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
+  ret <4 x double> %res
+}
diff --git a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
index aa178a337b1..0a4cd936058 100644
--- a/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512vl-intrinsics.ll
@@ -5978,16 +5978,14 @@ define <4 x i64> @test_x86_avx512_maskz_psrav_q_256(<4 x i64> %a0, <4 x i64> %a1
 
 declare <4 x i64> @llvm.x86.avx512.psrav.q.256(<4 x i64>, <4 x i64>) nounwind readnone
 
-declare <8 x float> @llvm.x86.avx512.mask.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
-
 define <8 x float> @test_vfmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
 ; CHECK-LABEL: test_vfmadd256_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa8,0xc2]
 ; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 -1) nounwind
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
+  ret <8 x float> %1
 }
 
 define <8 x float> @test_mask_vfmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
@@ -6005,20 +6003,20 @@ define <8 x float> @test_mask_vfmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8
 ; X64-NEXT:    vfmadd132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x98,0xc1]
 ; X64-NEXT:    # ymm0 = (ymm0 * ymm1) + ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) nounwind
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x float> %1, <8 x float> %a0
+  ret <8 x float> %3
 }
 
-declare <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
-
 define <4 x float> @test_vfmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
 ; CHECK-LABEL: test_vfmadd128_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa8,0xc2]
 ; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
+  ret <4 x float> %1
 }
 
 define <4 x float> @test_mask_vfmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
@@ -6036,20 +6034,21 @@ define <4 x float> @test_mask_vfmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4
 ; X64-NEXT:    vfmadd132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x98,0xc1]
 ; X64-NEXT:    # xmm0 = (xmm0 * xmm1) + xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x float> %1, <4 x float> %a0
+  ret <4 x float> %3
 }
 
-declare <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double> @test_fmadd256_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c) {
 ; CHECK-LABEL: test_fmadd256_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa8,0xc2]
 ; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %c, i8 -1)
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c)
+  ret <4 x double> %1
 }
 
 define <4 x double> @test_mask_fmadd256_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c, i8 %mask) {
@@ -6067,20 +6066,21 @@ define <4 x double> @test_mask_fmadd256_pd(<4 x double> %a, <4 x double> %b, <4
 ; X64-NEXT:    vfmadd132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x98,0xc1]
 ; X64-NEXT:    # ymm0 = (ymm0 * ymm1) + ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a, <4 x double> %b, <4 x double> %c, i8 %mask)
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a, <4 x double> %b, <4 x double> %c)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x double> %1, <4 x double> %a
+  ret <4 x double> %3
 }
 
-declare <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double> @test_fmadd128_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c) {
 ; CHECK-LABEL: test_fmadd128_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa8,0xc2]
 ; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 -1)
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c)
+  ret <2 x double> %1
 }
 
 define <2 x double> @test_mask_fmadd128_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
@@ -6098,12 +6098,13 @@ define <2 x double> @test_mask_fmadd128_pd(<2 x double> %a, <2 x double> %b, <2
 ; X64-NEXT:    vfmadd132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x98,0xc1]
 ; X64-NEXT:    # xmm0 = (xmm0 * xmm1) + xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask)
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a, <2 x double> %b, <2 x double> %c)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
+  %3 = select <2 x i1> %extract, <2 x double> %1, <2 x double> %a
+  ret <2 x double> %3
 }
 
-declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double>@test_int_x86_avx512_mask3_vfmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_128:
 ; X86:       # %bb.0:
@@ -6121,12 +6122,13 @@ define <2 x double>@test_int_x86_avx512_mask3_vfmadd_pd_128(<2 x double> %x0, <2
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) + xmm2
 ; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
+  %3 = select <2 x i1> %extract, <2 x double> %1, <2 x double> %x2
+  ret <2 x double> %3
 }
 
-declare <2 x double> @llvm.x86.avx512.maskz.vfmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double>@test_int_x86_avx512_maskz_vfmadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_128:
 ; X86:       # %bb.0:
@@ -6142,12 +6144,13 @@ define <2 x double>@test_int_x86_avx512_maskz_vfmadd_pd_128(<2 x double> %x0, <2
 ; X64-NEXT:    vfmadd213pd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xa8,0xc2]
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.maskz.vfmadd.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
+  %3 = select <2 x i1> %extract, <2 x double> %1, <2 x double> zeroinitializer
+  ret <2 x double> %3
 }
 
-declare <4 x double> @llvm.x86.avx512.mask3.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double>@test_int_x86_avx512_mask3_vfmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_pd_256:
 ; X86:       # %bb.0:
@@ -6165,12 +6168,13 @@ define <4 x double>@test_int_x86_avx512_mask3_vfmadd_pd_256(<4 x double> %x0, <4
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) + ymm2
 ; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmadd.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x double> %1, <4 x double> %x2
+  ret <4 x double> %3
 }
 
-declare <4 x double> @llvm.x86.avx512.maskz.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double>@test_int_x86_avx512_maskz_vfmadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_pd_256:
 ; X86:       # %bb.0:
@@ -6186,12 +6190,13 @@ define <4 x double>@test_int_x86_avx512_maskz_vfmadd_pd_256(<4 x double> %x0, <4
 ; X64-NEXT:    vfmadd213pd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xa8,0xc2]
 ; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.maskz.vfmadd.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x double> %1, <4 x double> zeroinitializer
+  ret <4 x double> %3
 }
 
-declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
-
 define <4 x float>@test_int_x86_avx512_mask3_vfmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_128:
 ; X86:       # %bb.0:
@@ -6209,12 +6214,13 @@ define <4 x float>@test_int_x86_avx512_mask3_vfmadd_ps_128(<4 x float> %x0, <4 x
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) + xmm2
 ; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x float> %1, <4 x float> %x2
+  ret <4 x float> %3
 }
 
-declare <4 x float> @llvm.x86.avx512.maskz.vfmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
-
 define <4 x float>@test_int_x86_avx512_maskz_vfmadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_128:
 ; X86:       # %bb.0:
@@ -6230,12 +6236,13 @@ define <4 x float>@test_int_x86_avx512_maskz_vfmadd_ps_128(<4 x float> %x0, <4 x
 ; X64-NEXT:    vfmadd213ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0xa8,0xc2]
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x float> %1, <4 x float> zeroinitializer
+  ret <4 x float> %3
 }
 
-declare <8 x float> @llvm.x86.avx512.mask3.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
-
 define <8 x float>@test_int_x86_avx512_mask3_vfmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ps_256:
 ; X86:       # %bb.0:
@@ -6253,12 +6260,12 @@ define <8 x float>@test_int_x86_avx512_mask3_vfmadd_ps_256(<8 x float> %x0, <8 x
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) + ymm2
 ; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmadd.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x float> %1, <8 x float> %x2
+  ret <8 x float> %3
 }
 
-declare <8 x float> @llvm.x86.avx512.maskz.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
-
 define <8 x float>@test_int_x86_avx512_maskz_vfmadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ps_256:
 ; X86:       # %bb.0:
@@ -6274,13 +6281,12 @@ define <8 x float>@test_int_x86_avx512_maskz_vfmadd_ps_256(<8 x float> %x0, <8 x
 ; X64-NEXT:    vfmadd213ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0xa8,0xc2]
 ; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.maskz.vfmadd.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2)
+  %2 = bitcast i8 %x3 to <8 x i1>
+  %3 = select <8 x i1> %2, <8 x float> %1, <8 x float> zeroinitializer
+  ret <8 x float> %3
 }
 
-
-declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double>@test_int_x86_avx512_mask3_vfmsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_128:
 ; X86:       # %bb.0:
@@ -6298,13 +6304,14 @@ define <2 x double>@test_int_x86_avx512_mask3_vfmsub_pd_128(<2 x double> %x0, <2
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) - xmm2
 ; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
-  ret <2 x double> %res
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %x2
+  %2 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %1)
+  %3 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %3, <8 x i1> %3, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract, <2 x double> %2, <2 x double> %x2
+  ret <2 x double> %4
 }
 
-
-declare <4 x double> @llvm.x86.avx512.mask3.vfmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double>@test_int_x86_avx512_mask3_vfmsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_pd_256:
 ; X86:       # %bb.0:
@@ -6322,12 +6329,14 @@ define <4 x double>@test_int_x86_avx512_mask3_vfmsub_pd_256(<4 x double> %x0, <4
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) - ymm2
 ; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
-  ret <4 x double> %res
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %2 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %1)
+  %3 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %3, <8 x i1> %3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract, <4 x double> %2, <4 x double> %x2
+  ret <4 x double> %4
 }
 
-declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
-
 define <4 x float>@test_int_x86_avx512_mask3_vfmsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_128:
 ; X86:       # %bb.0:
@@ -6345,12 +6354,14 @@ define <4 x float>@test_int_x86_avx512_mask3_vfmsub_ps_128(<4 x float> %x0, <4 x
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) - xmm2
 ; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
-  ret <4 x float> %res
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %1)
+  %3 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %3, <8 x i1> %3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract, <4 x float> %2, <4 x float> %x2
+  ret <4 x float> %4
 }
 
-declare <8 x float> @llvm.x86.avx512.mask3.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
-
 define <8 x float>@test_int_x86_avx512_mask3_vfmsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ps_256:
 ; X86:       # %bb.0:
@@ -6368,20 +6379,22 @@ define <8 x float>@test_int_x86_avx512_mask3_vfmsub_ps_256(<8 x float> %x0, <8 x
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) - ymm2
 ; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
-  ret <8 x float> %res
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %2 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %1)
+  %3 = bitcast i8 %x3 to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> %x2
+  ret <8 x float> %4
 }
 
-declare <8 x float> @llvm.x86.avx512.mask.vfnmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
-
 define <8 x float> @test_vfnmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
 ; CHECK-LABEL: test_vfnmadd256_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmadd213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xac,0xc2]
 ; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 -1) nounwind
-  ret <8 x float> %res
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %1, <8 x float> %a2)
+  ret <8 x float> %2
 }
 
 define <8 x float> @test_mask_vfnmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
@@ -6399,20 +6412,22 @@ define <8 x float> @test_mask_vfnmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8
 ; X64-NEXT:    vfnmadd132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x9c,0xc1]
 ; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) + ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmadd.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) nounwind
-  ret <8 x float> %res
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %1, <8 x float> %a2)
+  %3 = bitcast i8 %mask to <8 x i1>
+  %4 = select <8 x i1> %3, <8 x float> %2, <8 x float> %a0
+  ret <8 x float> %4
 }
 
-declare <4 x float> @llvm.x86.avx512.mask.vfnmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
-
 define <4 x float> @test_vfnmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
 ; CHECK-LABEL: test_vfnmadd128_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmadd213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xac,0xc2]
 ; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
-  ret <4 x float> %res
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %1, <4 x float> %a2)
+  ret <4 x float> %2
 }
 
 define <4 x float> @test_mask_vfnmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
@@ -6430,20 +6445,23 @@ define <4 x float> @test_mask_vfnmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4
 ; X64-NEXT:    vfnmadd132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x9c,0xc1]
 ; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) + xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
-  ret <4 x float> %res
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %1, <4 x float> %a2)
+  %3 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %3, <8 x i1> %3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract, <4 x float> %2, <4 x float> %a0
+  ret <4 x float> %4
 }
 
-declare <4 x double> @llvm.x86.avx512.mask.vfnmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
-
 define <4 x double> @test_vfnmadd256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
 ; CHECK-LABEL: test_vfnmadd256_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmadd213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xac,0xc2]
 ; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) + ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
-  ret <4 x double> %res
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %1, <4 x double> %a2)
+  ret <4 x double> %2
 }
 
 define <4 x double> @test_mask_vfnmadd256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
@@ -6461,20 +6479,23 @@ define <4 x double> @test_mask_vfnmadd256_pd(<4 x double> %a0, <4 x double> %a1,
 ; X64-NEXT:    vfnmadd132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x9c,0xc1]
 ; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) + ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
-  ret <4 x double> %res
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %1, <4 x double> %a2)
+  %3 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %3, <8 x i1> %3, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %4 = select <4 x i1> %extract, <4 x double> %2, <4 x double> %a0
+  ret <4 x double> %4
 }
 
-declare <2 x double> @llvm.x86.avx512.mask.vfnmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
-
 define <2 x double> @test_vfnmadd128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
 ; CHECK-LABEL: test_vfnmadd128_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmadd213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xac,0xc2]
 ; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) + xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
-  ret <2 x double> %res
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %1, <2 x double> %a2)
+  ret <2 x double> %2
 }
 
 define <2 x double> @test_mask_vfnmadd128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
@@ -6492,20 +6513,24 @@ define <2 x double> @test_mask_vfnmadd128_pd(<2 x double> %a0, <2 x double> %a1,
 ; X64-NEXT:    vfnmadd132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x9c,0xc1]
 ; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) + xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
-  ret <2 x double> %res
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %1, <2 x double> %a2)
+  %3 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %3, <8 x i1> %3, <2 x i32> <i32 0, i32 1>
+  %4 = select <2 x i1> %extract, <2 x double> %2, <2 x double> %a0
+  ret <2 x double> %4
 }
 
-declare <8 x float> @llvm.x86.avx512.mask.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
-
 define <8 x float> @test_vfnmsub256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2) {
 ; CHECK-LABEL: test_vfnmsub256_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xae,0xc2]
 ; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 -1) nounwind
-  ret <8 x float> %res
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %1, <8 x float> %2)
+  ret <8 x float> %3
 }
 
 define <8 x float> @test_mask_vfnmsub256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
@@ -6523,20 +6548,24 @@ define <8 x float> @test_mask_vfnmsub256_ps(<8 x float> %a0, <8 x float> %a1, <8
 ; X64-NEXT:    vfnmsub132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x9e,0xc1]
 ; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) - ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfnmsub.ps.256(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) nounwind
-  ret <8 x float> %res
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a0, <8 x float> %1, <8 x float> %2)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> %a0
+  ret <8 x float> %5
 }
 
-declare <4 x float> @llvm.x86.avx512.mask.vfnmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
-
 define <4 x float> @test_vfnmsub128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
 ; CHECK-LABEL: test_vfnmsub128_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xae,0xc2]
 ; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmsub.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
-  ret <4 x float> %res
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %1, <4 x float> %2)
+  ret <4 x float> %3
 }
 
 define <4 x float> @test_mask_vfnmsub128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
@@ -6554,20 +6583,25 @@ define <4 x float> @test_mask_vfnmsub128_ps(<4 x float> %a0, <4 x float> %a1, <4
 ; X64-NEXT:    vfnmsub132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x9e,0xc1]
 ; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) - xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfnmsub.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
-  ret <4 x float> %res
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a1
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %a2
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %1, <4 x float> %2)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %4, <8 x i1> %4, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract, <4 x float> %3, <4 x float> %a0
+  ret <4 x float> %5
 }
 
-declare <4 x double> @llvm.x86.avx512.mask.vfnmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
-
 define <4 x double> @test_vfnmsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
 ; CHECK-LABEL: test_vfnmsub256_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xae,0xc2]
 ; CHECK-NEXT:    # ymm0 = -(ymm1 * ymm0) - ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
-  ret <4 x double> %res
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %1, <4 x double> %2)
+  ret <4 x double> %3
 }
 
 define <4 x double> @test_mask_vfnmsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
@@ -6585,20 +6619,25 @@ define <4 x double> @test_mask_vfnmsub256_pd(<4 x double> %a0, <4 x double> %a1,
 ; X64-NEXT:    vfnmsub132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x9e,0xc1]
 ; X64-NEXT:    # ymm0 = -(ymm0 * ymm1) - ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfnmsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
-  ret <4 x double> %res
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %1, <4 x double> %2)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %4, <8 x i1> %4, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract, <4 x double> %3, <4 x double> %a0
+  ret <4 x double> %5
 }
 
-declare <2 x double> @llvm.x86.avx512.mask.vfnmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
-
 define <2 x double> @test_vfnmsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
 ; CHECK-LABEL: test_vfnmsub128_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfnmsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xae,0xc2]
 ; CHECK-NEXT:    # xmm0 = -(xmm1 * xmm0) - xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
-  ret <2 x double> %res
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %1, <2 x double> %2)
+  ret <2 x double> %3
 }
 
 define <2 x double> @test_mask_vfnmsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
@@ -6616,12 +6655,15 @@ define <2 x double> @test_mask_vfnmsub128_pd(<2 x double> %a0, <2 x double> %a1,
 ; X64-NEXT:    vfnmsub132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x9e,0xc1]
 ; X64-NEXT:    # xmm0 = -(xmm0 * xmm1) - xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfnmsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
-  ret <2 x double> %res
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a1
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %1, <2 x double> %2)
+  %4 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %4, <8 x i1> %4, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract, <2 x double> %3, <2 x double> %a0
+  ret <2 x double> %5
 }
 
-declare <2 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_128:
 ; X86:       # %bb.0:
@@ -6639,12 +6681,15 @@ define <2 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_128(<2 x double> %x0, <
 ; X64-NEXT:    # xmm2 = -(xmm0 * xmm1) - xmm2
 ; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
-  ret <2 x double> %res
+  %1 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %x0
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %1, <2 x double> %x1, <2 x double> %2)
+  %4 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %4, <8 x i1> %4, <2 x i32> <i32 0, i32 1>
+  %5 = select <2 x i1> %extract, <2 x double> %3, <2 x double> %x2
+  ret <2 x double> %5
 }
 
-declare <4 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_pd_256:
 ; X86:       # %bb.0:
@@ -6662,12 +6707,15 @@ define <4 x double>@test_int_x86_avx512_mask3_vfnmsub_pd_256(<4 x double> %x0, <
 ; X64-NEXT:    # ymm2 = -(ymm0 * ymm1) - ymm2
 ; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask3.vfnmsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
-  ret <4 x double> %res
+  %1 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x0
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %1, <4 x double> %x1, <4 x double> %2)
+  %4 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %4, <8 x i1> %4, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract, <4 x double> %3, <4 x double> %x2
+  ret <4 x double> %5
 }
 
-declare <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
-
 define <4 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_128:
 ; X86:       # %bb.0:
@@ -6685,12 +6733,15 @@ define <4 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_128(<4 x float> %x0, <4
 ; X64-NEXT:    # xmm2 = -(xmm0 * xmm1) - xmm2
 ; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
-  ret <4 x float> %res
+  %1 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x0
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %1, <4 x float> %x1, <4 x float> %2)
+  %4 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %4, <8 x i1> %4, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %5 = select <4 x i1> %extract, <4 x float> %3, <4 x float> %x2
+  ret <4 x float> %5
 }
 
-declare <8 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
-
 define <8 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ps_256:
 ; X86:       # %bb.0:
@@ -6708,20 +6759,25 @@ define <8 x float>@test_int_x86_avx512_mask3_vfnmsub_ps_256(<8 x float> %x0, <8
 ; X64-NEXT:    # ymm2 = -(ymm0 * ymm1) - ymm2
 ; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask3.vfnmsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
-  ret <8 x float> %res
+  %1 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x0
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %1, <8 x float> %x1, <8 x float> %2)
+  %4 = bitcast i8 %x3 to <8 x i1>
+  %5 = select <8 x i1> %4, <8 x float> %3, <8 x float> %x2
+  ret <8 x float> %5
 }
 
-declare <8 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
-
 define <8 x float> @test_fmaddsub256_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c) {
 ; CHECK-LABEL: test_fmaddsub256_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x75,0xa6,0xc2]
 ; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %c, i8 -1)
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c)
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %2)
+  %4 = shufflevector <8 x float> %3, <8 x float> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  ret <8 x float> %4
 }
 
 define <8 x float> @test_mask_fmaddsub256_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c, i8 %mask) {
@@ -6739,20 +6795,26 @@ define <8 x float> @test_mask_fmaddsub256_ps(<8 x float> %a, <8 x float> %b, <8
 ; X64-NEXT:    vfmaddsub132ps %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x29,0x96,0xc1]
 ; X64-NEXT:    # ymm0 = (ymm0 * ymm1) +/- ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.256(<8 x float> %a, <8 x float> %b, <8 x float> %c, i8 %mask)
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %c)
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %a, <8 x float> %b, <8 x float> %2)
+  %4 = shufflevector <8 x float> %3, <8 x float> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %5 = bitcast i8 %mask to <8 x i1>
+  %6 = select <8 x i1> %5, <8 x float> %4, <8 x float> %a
+  ret <8 x float> %6
 }
 
-declare <4 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
-
 define <4 x float> @test_fmaddsub128_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c) {
 ; CHECK-LABEL: test_fmaddsub128_ps:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0x71,0xa6,0xc2]
 ; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.128(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 -1)
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c)
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %2)
+  %4 = shufflevector <4 x float> %3, <4 x float> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x float> %4
 }
 
 define <4 x float> @test_mask_fmaddsub128_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
@@ -6770,20 +6832,27 @@ define <4 x float> @test_mask_fmaddsub128_ps(<4 x float> %a, <4 x float> %b, <4
 ; X64-NEXT:    vfmaddsub132ps %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0x6d,0x09,0x96,0xc1]
 ; X64-NEXT:    # xmm0 = (xmm0 * xmm1) +/- xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmaddsub.ps.128(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask)
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %c)
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %c
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a, <4 x float> %b, <4 x float> %2)
+  %4 = shufflevector <4 x float> %3, <4 x float> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x float> %4, <4 x float> %a
+  ret <4 x float> %6
 }
 
-declare <4 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
-
 define <4 x double> @test_vfmaddsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
 ; CHECK-LABEL: test_vfmaddsub256_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf5,0xa6,0xc2]
 ; CHECK-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %2)
+  %4 = shufflevector <4 x double> %3, <4 x double> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  ret <4 x double> %4
 }
 
 define <4 x double> @test_mask_vfmaddsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
@@ -6801,20 +6870,27 @@ define <4 x double> @test_mask_vfmaddsub256_pd(<4 x double> %a0, <4 x double> %a
 ; X64-NEXT:    vfmaddsub132pd %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf2,0xed,0x29,0x96,0xc1]
 ; X64-NEXT:    # ymm0 = (ymm0 * ymm1) +/- ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %2)
+  %4 = shufflevector <4 x double> %3, <4 x double> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x double> %4, <4 x double> %a0
+  ret <4 x double> %6
 }
 
-declare <2 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
-
 define <2 x double> @test_vfmaddsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
 ; CHECK-LABEL: test_vfmaddsub128_pd:
 ; CHECK:       # %bb.0:
 ; CHECK-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc4,0xe2,0xf1,0xa6,0xc2]
 ; CHECK-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
 ; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %2)
+  %4 = shufflevector <2 x double> %3, <2 x double> %1, <2 x i32> <i32 0, i32 3>
+  ret <2 x double> %4
 }
 
 define <2 x double> @test_mask_vfmaddsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
@@ -6832,12 +6908,16 @@ define <2 x double> @test_mask_vfmaddsub128_pd(<2 x double> %a0, <2 x double> %a
 ; X64-NEXT:    vfmaddsub132pd %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf2,0xed,0x09,0x96,0xc1]
 ; X64-NEXT:    # xmm0 = (xmm0 * xmm1) +/- xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfmaddsub.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %a2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %2)
+  %4 = shufflevector <2 x double> %3, <2 x double> %1, <2 x i32> <i32 0, i32 3>
+  %5 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
+  %6 = select <2 x i1> %extract, <2 x double> %4, <2 x double> %a0
+  ret <2 x double> %6
 }
 
-declare <2 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_128:
 ; X86:       # %bb.0:
@@ -6855,12 +6935,16 @@ define <2 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_128(<2 x double> %x0,
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) +/- xmm2
 ; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2)
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %2)
+  %4 = shufflevector <2 x double> %3, <2 x double> %1, <2 x i32> <i32 0, i32 3>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
+  %6 = select <2 x i1> %extract, <2 x double> %4, <2 x double> %x2
+  ret <2 x double> %6
 }
 
-declare <2 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_128:
 ; X86:       # %bb.0:
@@ -6876,12 +6960,16 @@ define <2 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_128(<2 x double> %x0,
 ; X64-NEXT:    vfmaddsub213pd %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0x89,0xa6,0xc2]
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2)
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %2)
+  %4 = shufflevector <2 x double> %3, <2 x double> %1, <2 x i32> <i32 0, i32 3>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
+  %6 = select <2 x i1> %extract, <2 x double> %4, <2 x double> zeroinitializer
+  ret <2 x double> %6
 }
 
-declare <4 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_pd_256:
 ; X86:       # %bb.0:
@@ -6899,12 +6987,16 @@ define <4 x double>@test_int_x86_avx512_mask3_vfmaddsub_pd_256(<4 x double> %x0,
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) +/- ymm2
 ; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmaddsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2)
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %2)
+  %4 = shufflevector <4 x double> %3, <4 x double> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x double> %4, <4 x double> %x2
+  ret <4 x double> %6
 }
 
-declare <4 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_pd_256:
 ; X86:       # %bb.0:
@@ -6920,12 +7012,16 @@ define <4 x double>@test_int_x86_avx512_maskz_vfmaddsub_pd_256(<4 x double> %x0,
 ; X64-NEXT:    vfmaddsub213pd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xa9,0xa6,0xc2]
 ; X64-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.maskz.vfmaddsub.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2)
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %2)
+  %4 = shufflevector <4 x double> %3, <4 x double> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x double> %4, <4 x double> zeroinitializer
+  ret <4 x double> %6
 }
 
-declare <4 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
-
 define <4 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_128:
 ; X86:       # %bb.0:
@@ -6943,12 +7039,16 @@ define <4 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_128(<4 x float> %x0, <
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) +/- xmm2
 ; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2)
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %2)
+  %4 = shufflevector <4 x float> %3, <4 x float> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x float> %4, <4 x float> %x2
+  ret <4 x float> %6
 }
 
-declare <4 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
-
 define <4 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_128:
 ; X86:       # %bb.0:
@@ -6964,12 +7064,16 @@ define <4 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_128(<4 x float> %x0, <
 ; X64-NEXT:    vfmaddsub213ps %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0x89,0xa6,0xc2]
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) +/- xmm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2)
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %2)
+  %4 = shufflevector <4 x float> %3, <4 x float> %1, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x float> %4, <4 x float> zeroinitializer
+  ret <4 x float> %6
 }
 
-declare <8 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
-
 define <8 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ps_256:
 ; X86:       # %bb.0:
@@ -6987,12 +7091,15 @@ define <8 x float>@test_int_x86_avx512_mask3_vfmaddsub_ps_256(<8 x float> %x0, <
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) +/- ymm2
 ; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmaddsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2)
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %2)
+  %4 = shufflevector <8 x float> %3, <8 x float> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %6 = select <8 x i1> %5, <8 x float> %4, <8 x float> %x2
+  ret <8 x float> %6
 }
 
-declare <8 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
-
 define <8 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ps_256:
 ; X86:       # %bb.0:
@@ -7008,12 +7115,15 @@ define <8 x float>@test_int_x86_avx512_maskz_vfmaddsub_ps_256(<8 x float> %x0, <
 ; X64-NEXT:    vfmaddsub213ps %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x75,0xa9,0xa6,0xc2]
 ; X64-NEXT:    # ymm0 = (ymm1 * ymm0) +/- ymm2
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.maskz.vfmaddsub.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2)
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %2)
+  %4 = shufflevector <8 x float> %3, <8 x float> %1, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %6 = select <8 x i1> %5, <8 x float> %4, <8 x float> zeroinitializer
+  ret <8 x float> %6
 }
 
-declare <2 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
-
 define <2 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_128:
 ; X86:       # %bb.0:
@@ -7031,12 +7141,16 @@ define <2 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_128(<2 x double> %x0,
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) -/+ xmm2
 ; X64-NEXT:    vmovapd %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf9,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.128(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3)
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2)
+  %2 = fsub <2 x double> <double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <2 x double> @llvm.fma.v2f64(<2 x double> %x0, <2 x double> %x1, <2 x double> %2)
+  %4 = shufflevector <2 x double> %1, <2 x double> %3, <2 x i32> <i32 0, i32 3>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <2 x i32> <i32 0, i32 1>
+  %6 = select <2 x i1> %extract, <2 x double> %4, <2 x double> %x2
+  ret <2 x double> %6
 }
 
-declare <4 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
-
 define <4 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_pd_256:
 ; X86:       # %bb.0:
@@ -7054,12 +7168,16 @@ define <4 x double>@test_int_x86_avx512_mask3_vfmsubadd_pd_256(<4 x double> %x0,
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) -/+ ymm2
 ; X64-NEXT:    vmovapd %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfd,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x double> @llvm.x86.avx512.mask3.vfmsubadd.pd.256(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2, i8 %x3)
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %x2)
+  %2 = fsub <4 x double> <double -0.000000e+00, double -0.000000e+00, double -0.000000e+00, double -0.000000e+00>, %x2
+  %3 = call <4 x double> @llvm.fma.v4f64(<4 x double> %x0, <4 x double> %x1, <4 x double> %2)
+  %4 = shufflevector <4 x double> %1, <4 x double> %3, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x double> %4, <4 x double> %x2
+  ret <4 x double> %6
 }
 
-declare <4 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8)
-
 define <4 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_128:
 ; X86:       # %bb.0:
@@ -7077,12 +7195,16 @@ define <4 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_128(<4 x float> %x0, <
 ; X64-NEXT:    # xmm2 = (xmm0 * xmm1) -/+ xmm2
 ; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.128(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3)
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2)
+  %2 = fsub <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <4 x float> @llvm.fma.v4f32(<4 x float> %x0, <4 x float> %x1, <4 x float> %2)
+  %4 = shufflevector <4 x float> %1, <4 x float> %3, <4 x i32> <i32 0, i32 5, i32 2, i32 7>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %extract = shufflevector <8 x i1> %5, <8 x i1> %5, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %6 = select <4 x i1> %extract, <4 x float> %4, <4 x float> %x2
+  ret <4 x float> %6
 }
 
-declare <8 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8)
-
 define <8 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3) {
 ; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ps_256:
 ; X86:       # %bb.0:
@@ -7100,17 +7222,21 @@ define <8 x float>@test_int_x86_avx512_mask3_vfmsubadd_ps_256(<8 x float> %x0, <
 ; X64-NEXT:    # ymm2 = (ymm0 * ymm1) -/+ ymm2
 ; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
 ; X64-NEXT:    retq # encoding: [0xc3]
-  %res = call <8 x float> @llvm.x86.avx512.mask3.vfmsubadd.ps.256(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2, i8 %x3)
-  ret <8 x float> %res
+  %1 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %x2)
+  %2 = fsub <8 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, %x2
+  %3 = call <8 x float> @llvm.fma.v8f32(<8 x float> %x0, <8 x float> %x1, <8 x float> %2)
+  %4 = shufflevector <8 x float> %1, <8 x float> %3, <8 x i32> <i32 0, i32 9, i32 2, i32 11, i32 4, i32 13, i32 6, i32 15>
+  %5 = bitcast i8 %x3 to <8 x i1>
+  %6 = select <8 x i1> %5, <8 x float> %4, <8 x float> %x2
+  ret <8 x float> %6
 }
 
-
 define <4 x float> @test_mask_vfmadd128_ps_rmk(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfmadd128_ps_rmk:
 ; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
-; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
 ; X86-NEXT:    vfmadd213ps (%eax), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0xa8,0x00]
 ; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X86-NEXT:    retl # encoding: [0xc3]
@@ -7122,16 +7248,19 @@ define <4 x float> @test_mask_vfmadd128_ps_rmk(<4 x float> %a0, <4 x float> %a1,
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x float> %1, <4 x float> %a0
+  ret <4 x float> %3
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmka(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfmadd128_ps_rmka:
 ; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
-; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
 ; X86-NEXT:    vfmadd213ps (%eax), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x09,0xa8,0x00]
 ; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X86-NEXT:    retl # encoding: [0xc3]
@@ -7143,8 +7272,11 @@ define <4 x float> @test_mask_vfmadd128_ps_rmka(<4 x float> %a0, <4 x float> %a1
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2, align 8
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x float> %1, <4 x float> %a0
+  ret <4 x float> %3
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmkz(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2) {
@@ -7161,8 +7293,8 @@ define <4 x float> @test_mask_vfmadd128_ps_rmkz(<4 x float> %a0, <4 x float> %a1
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
+  ret <4 x float> %1
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmkza(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2) {
@@ -7179,16 +7311,16 @@ define <4 x float> @test_mask_vfmadd128_ps_rmkza(<4 x float> %a0, <4 x float> %a
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <4 x float>, <4 x float>* %ptr_a2, align 4
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 -1) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2)
+  ret <4 x float> %1
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmb(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfmadd128_ps_rmb:
 ; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
-; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
 ; X86-NEXT:    vfmadd213ps (%eax){1to4}, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x19,0xa8,0x00]
 ; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X86-NEXT:    retl # encoding: [0xc3]
@@ -7204,16 +7336,19 @@ define <4 x float> @test_mask_vfmadd128_ps_rmb(<4 x float> %a0, <4 x float> %a1,
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 %mask) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x float> %1, <4 x float> %a0
+  ret <4 x float> %3
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmba(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfmadd128_ps_rmba:
 ; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
-; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
 ; X86-NEXT:    vfmadd213ps (%eax){1to4}, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x75,0x19,0xa8,0x00]
 ; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X86-NEXT:    retl # encoding: [0xc3]
@@ -7229,8 +7364,11 @@ define <4 x float> @test_mask_vfmadd128_ps_rmba(<4 x float> %a0, <4 x float> %a1
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 %mask) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x float> %1, <4 x float> %a0
+  ret <4 x float> %3
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmbz(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2) {
@@ -7251,8 +7389,8 @@ define <4 x float> @test_mask_vfmadd128_ps_rmbz(<4 x float> %a0, <4 x float> %a1
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 -1) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i)
+  ret <4 x float> %1
 }
 
 define <4 x float> @test_mask_vfmadd128_ps_rmbza(<4 x float> %a0, <4 x float> %a1, float* %ptr_a2) {
@@ -7273,16 +7411,16 @@ define <4 x float> @test_mask_vfmadd128_ps_rmbza(<4 x float> %a0, <4 x float> %a
   %vecinit2.i = insertelement <4 x float> %vecinit.i, float %q, i32 1
   %vecinit4.i = insertelement <4 x float> %vecinit2.i, float %q, i32 2
   %vecinit6.i = insertelement <4 x float> %vecinit4.i, float %q, i32 3
-  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ps.128(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i, i8 -1) nounwind
-  ret <4 x float> %res
+  %1 = call <4 x float> @llvm.fma.v4f32(<4 x float> %a0, <4 x float> %a1, <4 x float> %vecinit6.i)
+  ret <4 x float> %1
 }
 
 define <2 x double> @test_mask_vfmadd128_pd_rmk(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfmadd128_pd_rmk:
 ; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
-; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
 ; X86-NEXT:    vfmadd213pd (%eax), %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x09,0xa8,0x00]
 ; X86-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X86-NEXT:    retl # encoding: [0xc3]
@@ -7294,8 +7432,11 @@ define <2 x double> @test_mask_vfmadd128_pd_rmk(<2 x double> %a0, <2 x double> %
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <2 x double>, <2 x double>* %ptr_a2
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) nounwind
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <2 x i32> <i32 0, i32 1>
+  %3 = select <2 x i1> %extract, <2 x double> %1, <2 x double> %a0
+  ret <2 x double> %3
 }
 
 define <2 x double> @test_mask_vfmadd128_pd_rmkz(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2) {
@@ -7312,16 +7453,16 @@ define <2 x double> @test_mask_vfmadd128_pd_rmkz(<2 x double> %a0, <2 x double>
 ; X64-NEXT:    # xmm0 = (xmm1 * xmm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <2 x double>, <2 x double>* %ptr_a2
-  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.pd.128(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 -1) nounwind
-  ret <2 x double> %res
+  %1 = call <2 x double> @llvm.fma.v2f64(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2)
+  ret <2 x double> %1
 }
 
 define <4 x double> @test_mask_vfmadd256_pd_rmk(<4 x double> %a0, <4 x double> %a1, <4 x double>* %ptr_a2, i8 %mask) {
 ; X86-LABEL: test_mask_vfmadd256_pd_rmk:
 ; X86:       # %bb.0:
-; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax # encoding: [0x0f,0xb6,0x44,0x24,0x08]
-; X86-NEXT:    kmovw %eax, %k1 # encoding: [0xc5,0xf8,0x92,0xc8]
 ; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
+; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %ecx # encoding: [0x0f,0xb6,0x4c,0x24,0x08]
+; X86-NEXT:    kmovw %ecx, %k1 # encoding: [0xc5,0xf8,0x92,0xc9]
 ; X86-NEXT:    vfmadd213pd (%eax), %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0xf5,0x29,0xa8,0x00]
 ; X86-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
 ; X86-NEXT:    retl # encoding: [0xc3]
@@ -7333,8 +7474,11 @@ define <4 x double> @test_mask_vfmadd256_pd_rmk(<4 x double> %a0, <4 x double> %
 ; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <4 x double>, <4 x double>* %ptr_a2
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) nounwind
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
+  %2 = bitcast i8 %mask to <8 x i1>
+  %extract = shufflevector <8 x i1> %2, <8 x i1> %2, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
+  %3 = select <4 x i1> %extract, <4 x double> %1, <4 x double> %a0
+  ret <4 x double> %3
 }
 
 define <4 x double> @test_mask_vfmadd256_pd_rmkz(<4 x double> %a0, <4 x double> %a1, <4 x double>* %ptr_a2) {
@@ -7351,7 +7495,11 @@ define <4 x double> @test_mask_vfmadd256_pd_rmkz(<4 x double> %a0, <4 x double>
 ; X64-NEXT:    # ymm0 = (ymm1 * ymm0) + mem
 ; X64-NEXT:    retq # encoding: [0xc3]
   %a2 = load <4 x double>, <4 x double>* %ptr_a2
-  %res = call <4 x double> @llvm.x86.avx512.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
-  ret <4 x double> %res
+  %1 = call <4 x double> @llvm.fma.v4f64(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2)
+  ret <4 x double> %1
 }
 
+declare <8 x float> @llvm.fma.v8f32(<8 x float>, <8 x float>, <8 x float>)
+declare <4 x float> @llvm.fma.v4f32(<4 x float>, <4 x float>, <4 x float>)
+declare <4 x double> @llvm.fma.v4f64(<4 x double>, <4 x double>, <4 x double>)
+declare <2 x double> @llvm.fma.v2f64(<2 x double>, <2 x double>, <2 x double>)
author	Craig Topper <craig.topper@intel.com>	2018-07-06 03:42:09 +0000
committer	Craig Topper <craig.topper@intel.com>	2018-07-06 03:42:09 +0000
commit	7b35585ff13d833e4736f8169c9ea200dd011aeb (patch)
tree	98607c17c8067d09eea336be755773ba4e99a5da /llvm/test
parent	4ea8949697828da0f702024b9fb2b23a19c69924 (diff)
download	bcm5719-llvm-7b35585ff13d833e4736f8169c9ea200dd011aeb.tar.gz bcm5719-llvm-7b35585ff13d833e4736f8169c9ea200dd011aeb.zip