llvm/test/CodeGen/X86/avx512fp16vl-fma-intrinsics.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx512vl -mattr=+avx512fp16 --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vl -mattr=+avx512fp16 --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64


define <16 x half> @test_x86_vfnmadd_ph_z_256(<16 x half> %a0, <16 x half> %a1, <16 x half> %a2) {
; CHECK-LABEL: test_x86_vfnmadd_ph_z_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmadd213ph %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x75,0x28,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = call <16 x half> @llvm.fma.v16f16(<16 x half> %a0, <16 x half> %1, <16 x half> %a2)
  ret <16 x half> %2
}

define <16 x half> @test_mask_vfnmadd_ph_256(<16 x half> %a0, <16 x half> %a1, <16 x half> %a2, i16 %mask) {
; X86-LABEL: test_mask_vfnmadd_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmadd_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = call <16 x half> @llvm.fma.v16f16(<16 x half> %a0, <16 x half> %1, <16 x half> %a2)
  %3 = bitcast i16 %mask to <16 x i1>
  %4 = select <16 x i1> %3, <16 x half> %2, <16 x half> %a0
  ret <16 x half> %4
}

define <16 x half> @test_x86_vfnmsubph_z_256(<16 x half> %a0, <16 x half> %a1, <16 x half> %a2) {
; CHECK-LABEL: test_x86_vfnmsubph_z_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmsub213ph %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x75,0x28,0xae,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a2
  %3 = call <16 x half> @llvm.fma.v16f16(<16 x half> %a0, <16 x half> %1, <16 x half> %2)
  ret <16 x half> %3
}

define <16 x half> @test_mask_vfnmsub_ph_256(<16 x half> %a0, <16 x half> %a1, <16 x half> %a2, i16 %mask) {
; X86-LABEL: test_mask_vfnmsub_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmsub_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a2
  %3 = call <16 x half> @llvm.fma.v16f16(<16 x half> %a0, <16 x half> %1, <16 x half> %2)
  %4 = bitcast i16 %mask to <16 x i1>
  %5 = select <16 x i1> %4, <16 x half> %3, <16 x half> %a0
  ret <16 x half> %5
}

define <16 x half>@test_int_x86_avx512_mask3_vfmaddsub_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xb6,0xd1]
; 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_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xb6,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <16 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2)
  %bc = bitcast i16 %x3 to <16 x i1>
  %sel = select <16 x i1> %bc, <16 x half> %res, <16 x half> %x2
  ret <16 x half> %sel
}
declare <16 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.256(<16 x half>, <16 x half>, <16 x half>)

define <16 x half>@test_int_x86_avx512_maskz_vfmaddsub_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub213ph %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xa9,0xa6,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub213ph %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xa9,0xa6,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <16 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2)
  %bc = bitcast i16 %x3 to <16 x i1>
  %sel = select <16 x i1> %bc, <16 x half> %res, <16 x half> zeroinitializer
  ret <16 x half> %sel
}

define <16 x half>@test_int_x86_avx512_mask3_vfmsubadd_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsubadd231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xb7,0xd1]
; 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_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsubadd231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xb7,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %neg = fneg <16 x half> %x2
  %res = call <16 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.256(<16 x half> %x0, <16 x half> %x1, <16 x half> %neg)
  %bc = bitcast i16 %x3 to <16 x i1>
  %sel = select <16 x i1> %bc, <16 x half> %res, <16 x half> %x2
  ret <16 x half> %sel
}

define <16 x half>@test_int_x86_avx512_mask3_vfmsub_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xba,0xd1]
; 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_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xba,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x2
  %2 = call <16 x half> @llvm.fma.v16f16(<16 x half> %x0, <16 x half> %x1, <16 x half> %1)
  %3 = bitcast i16 %x3 to <16 x i1>
  %4 = select <16 x i1> %3, <16 x half> %2, <16 x half> %x2
  ret <16 x half> %4
}

define <16 x half>@test_int_x86_avx512_mask3_vfmadd_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xb8,0xd1]
; 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_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xb8,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <16 x half> @llvm.fma.v16f16(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2)
  %2 = bitcast i16 %x3 to <16 x i1>
  %3 = select <16 x i1> %2, <16 x half> %1, <16 x half> %x2
  ret <16 x half> %3
}

define <16 x half> @test_int_x86_avx512_maskz_vfmadd_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3) {
; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd213ph %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xa9,0xa8,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd213ph %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xa9,0xa8,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <16 x half> @llvm.fma.v16f16(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2)
  %2 = bitcast i16 %x3 to <16 x i1>
  %3 = select <16 x i1> %2, <16 x half> %1, <16 x half> zeroinitializer
  ret <16 x half> %3
}

define <16 x half>@test_int_x86_avx512_mask_vfnmsub_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfnmsub_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmsub_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x1
  %2 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x2
  %3 = call <16 x half> @llvm.fma.v16f16(<16 x half> %x0, <16 x half> %1, <16 x half> %2)
  %4 = bitcast i16 %x3 to <16 x i1>
  %5 = select <16 x i1> %4, <16 x half> %3, <16 x half> %x0
  ret <16 x half> %5
}

define <16 x half>@test_int_x86_avx512_mask3_vfnmsub_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xbe,0xd1]
; 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_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub231ph %ymm1, %ymm0, %ymm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x29,0xbe,0xd1]
; X64-NEXT:    vmovaps %ymm2, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xfc,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x0
  %2 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x2
  %3 = call <16 x half> @llvm.fma.v16f16(<16 x half> %1, <16 x half> %x1, <16 x half> %2)
  %4 = bitcast i16 %x3 to <16 x i1>
  %5 = select <16 x i1> %4, <16 x half> %3, <16 x half> %x2
  ret <16 x half> %5
}

define <16 x half>@test_int_x86_avx512_mask_vfnmadd_ph_256(<16 x half> %x0, <16 x half> %x1, <16 x half> %x2, i16 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfnmadd_ph_256:
; X86:       # %bb.0:
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmadd_ph_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132ph %ymm1, %ymm2, %ymm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x29,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x1
  %2 = call <16 x half> @llvm.fma.v16f16(<16 x half> %x0, <16 x half> %1, <16 x half> %x2)
  %3 = bitcast i16 %x3 to <16 x i1>
  %4 = select <16 x i1> %3, <16 x half> %2, <16 x half> %x0
  ret <16 x half> %4
}

define <16 x half> @test_x86_fma_vfnmadd_ph_256(<16 x half> %a0, <16 x half> %a1, <16 x half> %a2) #0 {
; CHECK-LABEL: test_x86_fma_vfnmadd_ph_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmadd213ph %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x75,0x28,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a0
  %2 = call <16 x half> @llvm.fma.v16f16(<16 x half> %1, <16 x half> %a1, <16 x half> %a2)
  ret <16 x half> %2
}

define <16 x half> @test_x86_fma_vfnmsub_ph_256(<16 x half> %a0, <16 x half> %a1, <16 x half> %a2) #0 {
; CHECK-LABEL: test_x86_fma_vfnmsub_ph_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmsub213ph %ymm0, %ymm1, %ymm0 # encoding: [0x62,0xf6,0x75,0x28,0xae,0xc0]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a0
  %2 = fsub <16 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a0
  %3 = call <16 x half> @llvm.fma.v16f16(<16 x half> %1, <16 x half> %a1, <16 x half> %2)
  ret <16 x half> %3
}

declare <16 x half> @llvm.fma.v16f16(<16 x half>, <16 x half>, <16 x half>)

define <8 x half> @test_x86_vfnmadd_ph_z_128(<8 x half> %a0, <8 x half> %a1, <8 x half> %a2) {
; CHECK-LABEL: test_x86_vfnmadd_ph_z_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmadd213ph %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x75,0x08,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = call <8 x half> @llvm.fma.v8f16(<8 x half> %a0, <8 x half> %1, <8 x half> %a2)
  ret <8 x half> %2
}

define <8 x half> @test_mask_vfnmadd_ph_128(<8 x half> %a0, <8 x half> %a1, <8 x half> %a2, i8 %mask) {
; X86-LABEL: test_mask_vfnmadd_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmadd_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = call <8 x half> @llvm.fma.v8f16(<8 x half> %a0, <8 x half> %1, <8 x half> %a2)
  %3 = bitcast i8 %mask to <8 x i1>
  %4 = select <8 x i1> %3, <8 x half> %2, <8 x half> %a0
  ret <8 x half> %4
}

define <8 x half> @test_x86_vfnmsubph_z_128(<8 x half> %a0, <8 x half> %a1, <8 x half> %a2) {
; CHECK-LABEL: test_x86_vfnmsubph_z_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmsub213ph %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x75,0x08,0xae,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a2
  %3 = call <8 x half> @llvm.fma.v8f16(<8 x half> %a0, <8 x half> %1, <8 x half> %2)
  ret <8 x half> %3
}

define <8 x half> @test_mask_vfnmsub_ph_128(<8 x half> %a0, <8 x half> %a1, <8 x half> %a2, i8 %mask) {
; X86-LABEL: test_mask_vfnmsub_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmsub_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a1
  %2 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a2
  %3 = call <8 x half> @llvm.fma.v8f16(<8 x half> %a0, <8 x half> %1, <8 x half> %2)
  %4 = bitcast i8 %mask to <8 x i1>
  %5 = select <8 x i1> %4, <8 x half> %3, <8 x half> %a0
  ret <8 x half> %5
}

define <8 x half>@test_int_x86_avx512_mask3_vfmaddsub_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb6,0xd1]
; 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_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb6,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <8 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2)
  %bc = bitcast i8 %x3 to <8 x i1>
  %sel = select <8 x i1> %bc, <8 x half> %res, <8 x half> %x2
  ret <8 x half> %sel
}
declare <8 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.128(<8 x half>, <8 x half>, <8 x half>)

define <8 x half>@test_int_x86_avx512_maskz_vfmaddsub_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub213ph %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0x89,0xa6,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub213ph %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0x89,0xa6,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <8 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2)
  %bc = bitcast i8 %x3 to <8 x i1>
  %sel = select <8 x i1> %bc, <8 x half> %res, <8 x half> zeroinitializer
  ret <8 x half> %sel
}

define <8 x half>@test_int_x86_avx512_mask3_vfmsubadd_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsubadd231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb7,0xd1]
; 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_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsubadd231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb7,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %neg = fneg <8 x half> %x2
  %res = call <8 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.128(<8 x half> %x0, <8 x half> %x1, <8 x half> %neg)
  %bc = bitcast i8 %x3 to <8 x i1>
  %sel = select <8 x i1> %bc, <8 x half> %res, <8 x half> %x2
  ret <8 x half> %sel
}

define <8 x half>@test_int_x86_avx512_mask3_vfmsub_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xba,0xd1]
; 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_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xba,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x2
  %2 = call <8 x half> @llvm.fma.v8f16(<8 x half> %x0, <8 x half> %x1, <8 x half> %1)
  %3 = bitcast i8 %x3 to <8 x i1>
  %4 = select <8 x i1> %3, <8 x half> %2, <8 x half> %x2
  ret <8 x half> %4
}

define <8 x half>@test_int_x86_avx512_mask3_vfmadd_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb8,0xd1]
; 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_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb8,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <8 x half> @llvm.fma.v8f16(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2)
  %2 = bitcast i8 %x3 to <8 x i1>
  %3 = select <8 x i1> %2, <8 x half> %1, <8 x half> %x2
  ret <8 x half> %3
}

define <8 x half> @test_int_x86_avx512_maskz_vfmadd_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3) {
; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd213ph %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0x89,0xa8,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd213ph %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0x89,0xa8,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <8 x half> @llvm.fma.v8f16(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2)
  %2 = bitcast i8 %x3 to <8 x i1>
  %3 = select <8 x i1> %2, <8 x half> %1, <8 x half> zeroinitializer
  ret <8 x half> %3
}

define <8 x half>@test_int_x86_avx512_mask_vfnmsub_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfnmsub_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmsub_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x1
  %2 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x2
  %3 = call <8 x half> @llvm.fma.v8f16(<8 x half> %x0, <8 x half> %1, <8 x half> %2)
  %4 = bitcast i8 %x3 to <8 x i1>
  %5 = select <8 x i1> %4, <8 x half> %3, <8 x half> %x0
  ret <8 x half> %5
}

define <8 x half>@test_int_x86_avx512_mask3_vfnmsub_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xbe,0xd1]
; 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_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub231ph %xmm1, %xmm0, %xmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xbe,0xd1]
; X64-NEXT:    vmovaps %xmm2, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x0
  %2 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x2
  %3 = call <8 x half> @llvm.fma.v8f16(<8 x half> %1, <8 x half> %x1, <8 x half> %2)
  %4 = bitcast i8 %x3 to <8 x i1>
  %5 = select <8 x i1> %4, <8 x half> %3, <8 x half> %x2
  ret <8 x half> %5
}

define <8 x half>@test_int_x86_avx512_mask_vfnmadd_ph_128(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfnmadd_ph_128:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmadd_ph_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132ph %xmm1, %xmm2, %xmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x09,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %x1
  %2 = call <8 x half> @llvm.fma.v8f16(<8 x half> %x0, <8 x half> %1, <8 x half> %x2)
  %3 = bitcast i8 %x3 to <8 x i1>
  %4 = select <8 x i1> %3, <8 x half> %2, <8 x half> %x0
  ret <8 x half> %4
}

define <8 x half> @test_x86_fma_vfnmadd_ph_128(<8 x half> %a0, <8 x half> %a1, <8 x half> %a2) #0 {
; CHECK-LABEL: test_x86_fma_vfnmadd_ph_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmadd213ph %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x75,0x08,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a0
  %2 = call <8 x half> @llvm.fma.v8f16(<8 x half> %1, <8 x half> %a1, <8 x half> %a2)
  ret <8 x half> %2
}

define <8 x half> @test_x86_fma_vfnmsub_ph_128(<8 x half> %a0, <8 x half> %a1, <8 x half> %a2) #0 {
; CHECK-LABEL: test_x86_fma_vfnmsub_ph_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmsub213ph %xmm0, %xmm1, %xmm0 # encoding: [0x62,0xf6,0x75,0x08,0xae,0xc0]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a0
  %2 = fsub <8 x half> <half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00, half -0.000000e+00>, %a0
  %3 = call <8 x half> @llvm.fma.v8f16(<8 x half> %1, <8 x half> %a1, <8 x half> %2)
  ret <8 x half> %3
}

declare <8 x half> @llvm.fma.v8f16(<8 x half>, <8 x half>, <8 x half>)