AVX-512: Added FMA instructions, intrinsics an tests for KNL and SKX targets

by Asaf Badouh

http://reviews.llvm.org/D6456

llvm-svn: 224764

2 files changed, 456 insertions, 0 deletions

diff --git a/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll b/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll
index 366d32427b2..116531d5af9 100644
--- a/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512-fma-intrinsics.ll
@@ -8,6 +8,13 @@ define <16 x float> @test_x86_vfmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <1
 }
 declare <16 x float> @llvm.x86.fma.mask.vfmadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
 
+define <16 x float> @test_mask_vfmadd_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd_ps
+  ; CHECK: vfmadd213ps %zmm
+  %res = call <16 x float> @llvm.x86.fma.mask.vfmadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
+  ret <16 x float> %res
+}
+
 define <8 x double> @test_x86_vfmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
   ; CHECK-LABEL: test_x86_vfmadd_pd_z
   ; CHECK: vfmadd213pd %zmm
@@ -32,6 +39,13 @@ define <16 x float> @test_x86_vfmsubps_z(<16 x float> %a0, <16 x float> %a1, <16
 }
 declare <16 x float> @llvm.x86.fma.mask.vfmsub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
 
+define <16 x float> @test_mask_vfmsub_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsub_ps
+  ; CHECK: vfmsub213ps %zmm
+  %res = call <16 x float> @llvm.x86.fma.mask.vfmsub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
+  ret <16 x float> %res
+}
+
 define <8 x double> @test_x86_vfmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
   ; CHECK-LABEL: test_x86_vfmsubpd_z
   ; CHECK: vfmsub213pd %zmm
@@ -40,6 +54,13 @@ define <8 x double> @test_x86_vfmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8
 }
 declare <8 x double> @llvm.x86.fma.mask.vfmsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
 
+define <8 x double> @test_mask_vfmsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsub_pd
+  ; CHECK: vfmsub213pd %zmm
+  %res = call <8 x double> @llvm.x86.fma.mask.vfmsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
+  ret <8 x double> %res
+}
+
 define <16 x float> @test_x86_vfnmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
   ; CHECK-LABEL: test_x86_vfnmadd_ps_z
   ; CHECK: vfnmadd213ps %zmm
@@ -48,6 +69,13 @@ define <16 x float> @test_x86_vfnmadd_ps_z(<16 x float> %a0, <16 x float> %a1, <
 }
 declare <16 x float> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfnmadd_ps
+  ; CHECK: vfnmadd213ps %zmm
+  %res = call <16 x float> @llvm.x86.fma.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
+}
+
 define <8 x double> @test_x86_vfnmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
   ; CHECK-LABEL: test_x86_vfnmadd_pd_z
   ; CHECK: vfnmadd213pd %zmm
@@ -56,6 +84,13 @@ define <8 x double> @test_x86_vfnmadd_pd_z(<8 x double> %a0, <8 x double> %a1, <
 }
 declare <8 x double> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfnmadd_pd
+  ; CHECK: vfnmadd213pd %zmm
+  %res = call <8 x double> @llvm.x86.fma.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
+}
+
 define <16 x float> @test_x86_vfnmsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
   ; CHECK-LABEL: test_x86_vfnmsubps_z
   ; CHECK: vfnmsub213ps %zmm
@@ -64,6 +99,13 @@ define <16 x float> @test_x86_vfnmsubps_z(<16 x float> %a0, <16 x float> %a1, <1
 }
 declare <16 x float> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfnmsub_ps
+  ; CHECK: vfnmsub213ps %zmm
+  %res = call <16 x float> @llvm.x86.fma.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
+}
+
 define <8 x double> @test_x86_vfnmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
   ; CHECK-LABEL: test_x86_vfnmsubpd_z
   ; CHECK: vfnmsub213pd %zmm
@@ -72,6 +114,13 @@ define <8 x double> @test_x86_vfnmsubpd_z(<8 x double> %a0, <8 x double> %a1, <8
 }
 declare <8 x double> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfnmsub_pd
+  ; CHECK: vfnmsub213pd %zmm
+  %res = call <8 x double> @llvm.x86.fma.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
+}
+
 define <16 x float> @test_x86_vfmaddsubps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
   ; CHECK-LABEL: test_x86_vfmaddsubps_z
   ; CHECK: vfmaddsub213ps %zmm
@@ -96,6 +145,13 @@ define <8 x double> @test_x86_vfmaddsubpd_z(<8 x double> %a0, <8 x double> %a1,
 }
 declare <8 x double> @llvm.x86.fma.mask.vfmaddsub.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
 
+define <8 x double> @test_mask_vfmaddsub_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmaddsub_pd
+  ; CHECK: vfmaddsub213pd %zmm
+  %res = call <8 x double> @llvm.x86.fma.mask.vfmaddsub.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
+  ret <8 x double> %res
+}
+
 define <16 x float> @test_x86_vfmsubaddps_z(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2) {
   ; CHECK-LABEL: test_x86_vfmsubaddps_z
   ; CHECK: vfmsubadd213ps %zmm
@@ -104,6 +160,13 @@ define <16 x float> @test_x86_vfmsubaddps_z(<16 x float> %a0, <16 x float> %a1,
 }
 declare <16 x float> @llvm.x86.fma.mask.vfmsubadd.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32) nounwind readnone
 
+define <16 x float> @test_mask_vfmsubadd_ps(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubadd_ps
+  ; CHECK: vfmsubadd213ps %zmm
+  %res = call <16 x float> @llvm.x86.fma.mask.vfmsubadd.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %a2, i16 %mask, i32 4) nounwind
+  ret <16 x float> %res
+}
+
 define <8 x double> @test_x86_vfmsubaddpd_z(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2) {
   ; CHECK-LABEL: test_x86_vfmsubaddpd_z
   ; CHECK: vfmsubadd213pd %zmm
@@ -111,3 +174,11 @@ define <8 x double> @test_x86_vfmsubaddpd_z(<8 x double> %a0, <8 x double> %a1,
   ret <8 x double> %res
 }
 declare <8 x double> @llvm.x86.fma.mask.vfmsubadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
+
+define <8 x double> @test_mask_vfmsubadd_pd(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubadd_pd
+  ; CHECK: vfmsubadd213pd %zmm
+  %res = call <8 x double> @llvm.x86.fma.mask.vfmsubadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
+  ret <8 x double> %res
+}
+
diff --git a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
index 45f8d6de970..678f252dea4 100644
--- a/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
+++ b/llvm/test/CodeGen/X86/avx512bwvl-intrinsics.ll
@@ -611,3 +611,388 @@ define <8 x i8> @test_mask_ucmp_w_128(<8 x i16> %a0, <8 x i16> %a1, i8 %mask) {
 }
 
 declare i8 @llvm.x86.avx512.mask.ucmp.w.128(<8 x i16>, <8 x i16>, i32, i8) nounwind readnone
+
+declare <8 x float> @llvm.x86.fma.mask.vfmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_mask_vfmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd256_ps
+  ; CHECK: vfmadd213ps %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0xa8,0xc2]
+  %res = call <8 x float> @llvm.x86.fma.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.fma.mask.vfmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_mask_vfmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps
+  ; CHECK: vfmadd213ps %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa8,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.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.fma.mask.vfmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8)
+
+define <4 x double> @test_mask_fmadd256_pd(<4 x double> %a, <4 x double> %b, <4 x double> %c, i8 %mask) {
+; CHECK-LABEL: test_mask_fmadd256_pd:
+; CHECK: vfmadd213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa8,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.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.fma.mask.vfmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8)
+
+define <2 x double> @test_mask_fmadd128_pd(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask) {
+; CHECK-LABEL: test_mask_fmadd128_pd:
+; CHECK: vfmadd213pd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa8,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.mask.vfmadd.pd.128(<2 x double> %a, <2 x double> %b, <2 x double> %c, i8 %mask)
+  ret <2 x double> %res
+}
+
+declare <8 x float> @llvm.x86.fma.mask.vfmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_mask_vfmsub256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsub256_ps
+  ; CHECK: vfmsub213ps %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0xaa,0xc2]
+  %res = call <8 x float> @llvm.x86.fma.mask.vfmsub.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.fma.mask.vfmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_mask_vfmsub128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsub128_ps
+  ; CHECK: vfmsub213ps %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xaa,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.mask.vfmsub.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.fma.mask.vfmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_mask_vfmsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsub256_pd
+  ; CHECK: vfmsub213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xaa,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.mask.vfmsub.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.fma.mask.vfmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_mask_vfmsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsub128_pd
+  ; CHECK: vfmsub213pd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xaa,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.mask.vfmsub.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.fma.mask.vfnmadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_mask_vfnmadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmadd256_ps
+  ; CHECK: vfnmadd213ps %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0xac,0xc2]
+  %res = call <8 x float> @llvm.x86.fma.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.fma.mask.vfnmadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_mask_vfnmadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmadd128_ps
+  ; CHECK: vfnmadd213ps %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xac,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.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.fma.mask.vfnmadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_mask_vfnmadd256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmadd256_pd
+  ; CHECK: vfnmadd213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xac,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.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.fma.mask.vfnmadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_mask_vfnmadd128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmadd128_pd
+  ; CHECK: vfnmadd213pd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xac,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.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.fma.mask.vfnmsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_mask_vfnmsub256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmsub256_ps
+  ; CHECK: vfnmsub213ps %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0xae,0xc2]
+  %res = call <8 x float> @llvm.x86.fma.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.fma.mask.vfnmsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_mask_vfnmsub128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmsub128_ps
+  ; CHECK: vfnmsub213ps %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xae,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.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.fma.mask.vfnmsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_mask_vfnmsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmsub256_pd
+  ; CHECK: vfnmsub213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xae,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.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.fma.mask.vfnmsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_mask_vfnmsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfnmsub128_pd
+  ; CHECK: vfnmsub213pd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xae,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.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 <8 x float> @llvm.x86.fma.mask.vfmaddsub.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_mask_fmaddsub256_ps(<8 x float> %a, <8 x float> %b, <8 x float> %c, i8 %mask) {
+; CHECK-LABEL: test_mask_fmaddsub256_ps:
+; CHECK: vfmaddsub213ps %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0xa6,0xc2]
+  %res = call <8 x float> @llvm.x86.fma.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.fma.mask.vfmaddsub.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_mask_fmaddsub128_ps(<4 x float> %a, <4 x float> %b, <4 x float> %c, i8 %mask) {
+; CHECK-LABEL: test_mask_fmaddsub128_ps:
+; CHECK: vfmaddsub213ps %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa6,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.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.fma.mask.vfmaddsub.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_mask_vfmaddsub256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmaddsub256_pd
+  ; CHECK: vfmaddsub213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa6,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.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.fma.mask.vfmaddsub.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_mask_vfmaddsub128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmaddsub128_pd
+  ; CHECK: vfmaddsub213pd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa6,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.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 <8 x float> @llvm.x86.fma.mask.vfmsubadd.ps.256(<8 x float>, <8 x float>, <8 x float>, i8) nounwind readnone
+
+define <8 x float> @test_mask_vfmsubadd256_ps(<8 x float> %a0, <8 x float> %a1, <8 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubadd256_ps
+  ; CHECK: vfmsubadd213ps %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x29,0xa7,0xc2]
+  %res = call <8 x float> @llvm.x86.fma.mask.vfmsubadd.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.fma.mask.vfmsubadd.ps.128(<4 x float>, <4 x float>, <4 x float>, i8) nounwind readnone
+
+define <4 x float> @test_mask_vfmsubadd128_ps(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubadd128_ps
+  ; CHECK: vfmsubadd213ps %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa7,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.mask.vfmsubadd.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.fma.mask.vfmsubadd.pd.256(<4 x double>, <4 x double>, <4 x double>, i8) nounwind readnone
+
+define <4 x double> @test_mask_vfmsubadd256_pd(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubadd256_pd
+  ; CHECK: vfmsubadd213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa7,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.mask.vfmsubadd.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.fma.mask.vfmsubadd.pd.128(<2 x double>, <2 x double>, <2 x double>, i8) nounwind readnone
+
+define <2 x double> @test_mask_vfmsubadd128_pd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubadd128_pd
+  ; CHECK: vfmsubadd213pd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa7,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.mask.vfmsubadd.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_vfmsubadd128rm_pd(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubadd128rm_pd
+  ; CHECK: vfmsubadd213pd (%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa7,0x07]
+  %a2 = load <2 x double>* %ptr_a2
+  %res = call <2 x double> @llvm.x86.fma.mask.vfmsubadd.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 double> @llvm.x86.fma.mask.vfmsubadd.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32) nounwind readnone
+define <8 x double> @test_mask_vfmsubaddrm_pd(<8 x double> %a0, <8 x double> %a1, <8 x double>* %ptr_a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmsubaddrm_pd
+  ; CHECK: vfmsubadd213pd  (%rdi), %zmm1, %zmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x49,0xa7,0x07]
+  %a2 = load <8 x double>* %ptr_a2, align 8
+  %res = call <8 x double> @llvm.x86.fma.mask.vfmsubadd.pd.512(<8 x double> %a0, <8 x double> %a1, <8 x double> %a2, i8 %mask, i32 4) nounwind
+  ret <8 x double> %res
+}
+
+define <4 x float> @test_mask_vfmadd128_ps_r(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_r
+  ; CHECK: vfmadd213ps %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa8,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.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_rz(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rz
+  ; CHECK: vfmadd213ps %xmm2, %xmm1, %xmm0 ## encoding: [0x62,0xf2,0x75,0x08,0xa8,0xc2]
+  %res = call <4 x float> @llvm.x86.fma.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_rmk(<4 x float> %a0, <4 x float> %a1, <4 x float>* %ptr_a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmk
+  ; CHECK: vfmadd213ps	(%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa8,0x07]
+  %a2 = load <4 x float>* %ptr_a2
+  %res = call <4 x float> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmka
+  ; CHECK: vfmadd213ps     (%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x09,0xa8,0x07]
+  %a2 = load <4 x float>* %ptr_a2, align 8
+  %res = call <4 x float> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmkz
+  ; CHECK: vfmadd213ps	(%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0x07]
+  %a2 = load <4 x float>* %ptr_a2
+  %res = call <4 x float> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmkza
+  ; CHECK: vfmadd213ps	(%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0x71,0xa8,0x07]
+  %a2 = load <4 x float>* %ptr_a2, align 4
+  %res = call <4 x float> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmb
+  ; CHECK: vfmadd213ps	(%rdi){1to4}, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
+  %q = load 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.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmba
+  ; CHECK: vfmadd213ps	(%rdi){1to4}, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0x75,0x19,0xa8,0x07]
+  %q = load 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.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmbz
+  ; CHECK: vfmadd213ps	(%rdi){1to4}, %xmm1, %xmm0  ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
+  %q = load 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.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_ps_rmbza
+  ; CHECK: vfmadd213ps	(%rdi){1to4}, %xmm1, %xmm0  ## encoding: [0x62,0xf2,0x75,0x18,0xa8,0x07]
+  %q = load 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.fma.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_r(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd128_pd_r
+  ; CHECK: vfmadd213pd %xmm2, %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa8,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.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_rz(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2) {
+  ; CHECK-LABEL: test_mask_vfmadd128_pd_rz
+  ; CHECK: vfmadd213pd %xmm2, %xmm1, %xmm0 ## encoding: [0x62,0xf2,0xf5,0x08,0xa8,0xc2]
+  %res = call <2 x double> @llvm.x86.fma.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 <2 x double> @test_mask_vfmadd128_pd_rmk(<2 x double> %a0, <2 x double> %a1, <2 x double>* %ptr_a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd128_pd_rmk
+  ; CHECK: vfmadd213pd	(%rdi), %xmm1, %xmm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x09,0xa8,0x07]
+  %a2 = load <2 x double>* %ptr_a2
+  %res = call <2 x double> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd128_pd_rmkz
+  ; CHECK: vfmadd213pd	(%rdi), %xmm1, %xmm0 ## encoding: [0xc4,0xe2,0xf1,0xa8,0x07]
+  %a2 = load <2 x double>* %ptr_a2
+  %res = call <2 x double> @llvm.x86.fma.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_r(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd256_pd_r
+  ; CHECK: vfmadd213pd %ymm2, %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa8,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.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_rz(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2) {
+  ; CHECK-LABEL: test_mask_vfmadd256_pd_rz
+  ; CHECK: vfmadd213pd %ymm2, %ymm1, %ymm0 ## encoding: [0x62,0xf2,0xf5,0x28,0xa8,0xc2]
+  %res = call <4 x double> @llvm.x86.fma.mask.vfmadd.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_vfmadd256_pd_rmk(<4 x double> %a0, <4 x double> %a1, <4 x double>* %ptr_a2, i8 %mask) {
+  ; CHECK-LABEL: test_mask_vfmadd256_pd_rmk
+  ; CHECK: vfmadd213pd	(%rdi), %ymm1, %ymm0 {%k1} ## encoding: [0x62,0xf2,0xf5,0x29,0xa8,0x07]
+  %a2 = load <4 x double>* %ptr_a2
+  %res = call <4 x double> @llvm.x86.fma.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) {
+  ; CHECK-LABEL: test_mask_vfmadd256_pd_rmkz
+  ; CHECK: vfmadd213pd	(%rdi), %ymm1, %ymm0 ## encoding: [0xc4,0xe2,0xf5,0xa8,0x07]
+  %a2 = load <4 x double>* %ptr_a2
+  %res = call <4 x double> @llvm.x86.fma.mask.vfmadd.pd.256(<4 x double> %a0, <4 x double> %a1, <4 x double> %a2, i8 -1) nounwind
+  ret <4 x double> %res
+}