llvm/llvm/test/CodeGen/X86/avx512fp16-fma-intrinsics.ll

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


declare <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half>, <32 x half>, <32 x half>, i32)

define <32 x half> @test_x86_vfnmadd_ph_z(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_x86_vfnmadd_ph_z:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmadd213ph %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x48,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %a0, <32 x half> %1, <32 x half> %a2)
  ret <32 x half> %2
}

define <32 x half> @test_mask_vfnmadd_ph(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 %mask) {
; X86-LABEL: test_mask_vfnmadd_ph:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmadd_ph:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %a0, <32 x half> %1, <32 x half> %a2)
  %3 = bitcast i32 %mask to <32 x i1>
  %4 = select <32 x i1> %3, <32 x half> %2, <32 x half> %a0
  ret <32 x half> %4
}

define <32 x half> @test_x86_vfnmsubph_z(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_x86_vfnmsubph_z:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmsub213ph %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x48,0xae,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %a0, <32 x half> %1, <32 x half> %2)
  ret <32 x half> %3
}

define <32 x half> @test_mask_vfnmsub_ph(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 %mask) {
; X86-LABEL: test_mask_vfnmsub_ph:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_vfnmsub_ph:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %a0, <32 x half> %1, <32 x half> %2)
  %4 = bitcast i32 %mask to <32 x i1>
  %5 = select <32 x i1> %4, <32 x half> %3, <32 x half> %a0
  ret <32 x half> %5
}

define <32 x half> @test_x86_vfmaddsubph_z(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_x86_vfmaddsubph_z:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfmaddsub213ph %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x48,0xa6,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 4) #2
  ret <32 x half> %res
}

define <32 x half> @test_mask_fmaddsub_ph(<32 x half> %a, <32 x half> %b, <32 x half> %c, i32 %mask) {
; X86-LABEL: test_mask_fmaddsub_ph:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x96,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_fmaddsub_ph:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x96,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.512(<32 x half> %a, <32 x half> %b, <32 x half> %c, i32 4)
  %bc = bitcast i32 %mask to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %a
  ret <32 x half> %sel
}

declare <32 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.512(<32 x half>, <32 x half>, <32 x half>, i32) nounwind readnone

define <32 x half>@test_int_x86_avx512_mask_vfmaddsub_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfmaddsub_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x96,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfmaddsub_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x96,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 4)
  %bc = bitcast i32 %x3 to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %x0
  ret <32 x half> %sel
}

define <32 x half>@test_int_x86_avx512_mask3_vfmaddsub_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xb6,0xd1]
; X86-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmaddsub_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xb6,0xd1]
; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 4)
  %bc = bitcast i32 %x3 to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %x2
  ret <32 x half> %sel
}

define <32 x half>@test_int_x86_avx512_maskz_vfmaddsub_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmaddsub213ph %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xc9,0xa6,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmaddsub_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmaddsub213ph %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xc9,0xa6,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 4)
  %bc = bitcast i32 %x3 to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> zeroinitializer
  ret <32 x half> %sel
}

define <32 x half>@test_int_x86_avx512_mask3_vfmsubadd_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsubadd231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xb7,0xd1]
; X86-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmsubadd_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsubadd231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xb7,0xd1]
; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %neg = fneg <32 x half> %x2
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmaddsub.ph.512(<32 x half> %x0, <32 x half> %x1, <32 x half> %neg, i32 4)
  %bc = bitcast i32 %x3 to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %x2
  ret <32 x half> %sel
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrb_rne(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 %mask) {
; X86-LABEL: test_mask_round_vfmadd512_ph_rrb_rne:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132ph {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x19,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ph_rrb_rne:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132ph {rn-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x19,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 8) nounwind
  %bc = bitcast i32 %mask to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %a0
  ret <32 x half> %sel
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrb_rtn(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 %mask) {
; X86-LABEL: test_mask_round_vfmadd512_ph_rrb_rtn:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132ph {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x39,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ph_rrb_rtn:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132ph {rd-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x39,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 9) nounwind
  %bc = bitcast i32 %mask to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %a0
  ret <32 x half> %sel
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrb_rtp(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 %mask) {
; X86-LABEL: test_mask_round_vfmadd512_ph_rrb_rtp:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132ph {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x59,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ph_rrb_rtp:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132ph {ru-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x59,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 10) nounwind
  %bc = bitcast i32 %mask to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %a0
  ret <32 x half> %sel
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrb_rtz(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 %mask) {
; X86-LABEL: test_mask_round_vfmadd512_ph_rrb_rtz:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132ph {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x79,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ph_rrb_rtz:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132ph {rz-sae}, %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x79,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 11) nounwind
  %bc = bitcast i32 %mask to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %a0
  ret <32 x half> %sel
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrb_current(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 %mask) {
; X86-LABEL: test_mask_round_vfmadd512_ph_rrb_current:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_round_vfmadd512_ph_rrb_current:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 4) nounwind
  %bc = bitcast i32 %mask to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %a0
  ret <32 x half> %sel
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrbz_rne(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ph_rrbz_rne:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfmadd213ph {rn-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x18,0xa8,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 8) nounwind
  ret <32 x half> %res
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrbz_rtn(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ph_rrbz_rtn:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfmadd213ph {rd-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x38,0xa8,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 9) nounwind
  ret <32 x half> %res
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrbz_rtp(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ph_rrbz_rtp:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfmadd213ph {ru-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x58,0xa8,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 10) nounwind
  ret <32 x half> %res
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrbz_rtz(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ph_rrbz_rtz:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfmadd213ph {rz-sae}, %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x78,0xa8,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 11) nounwind
  ret <32 x half> %res
}

define <32 x half> @test_mask_round_vfmadd512_ph_rrbz_current(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) {
; CHECK-LABEL: test_mask_round_vfmadd512_ph_rrbz_current:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfmadd213ph %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x48,0xa8,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2, i32 4) nounwind
  ret <32 x half> %res
}

define <32 x half>@test_int_x86_avx512_mask3_vfmsub_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmsub_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmsub231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xba,0xd1]
; X86-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmsub_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmsub231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xba,0xd1]
; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %x0, <32 x half> %x1, <32 x half> %1)
  %3 = bitcast i32 %x3 to <32 x i1>
  %4 = select <32 x i1> %3, <32 x half> %2, <32 x half> %x2
  ret <32 x half> %4
}

define <32 x half>@test_int_x86_avx512_mask_vfmadd_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfmadd_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x98,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfmadd_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x98,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x half> @llvm.x86.avx512fp16.vfmadd.ph.512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 4)
  %bc = bitcast i32 %x3 to <32 x i1>
  %sel = select <32 x i1> %bc, <32 x half> %res, <32 x half> %x0
  ret <32 x half> %sel
}

define <32 x half>@test_int_x86_avx512_mask3_vfmadd_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xb8,0xd1]
; X86-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xb8,0xd1]
; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x half> @llvm.fma.v32f16(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x half> %1, <32 x half> %x2
  ret <32 x half> %3
}

define <32 x half> @test_int_x86_avx512_maskz_vfmadd_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd213ph %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xc9,0xa8,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd213ph %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0xc9,0xa8,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x half> @llvm.fma.v32f16(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x half> %1, <32 x half> zeroinitializer
  ret <32 x half> %3
}

define <32 x half>@test_int_x86_avx512_mask_vfnmsub_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfnmsub_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9e,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmsub_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9e,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %x0, <32 x half> %1, <32 x half> %2)
  %4 = bitcast i32 %x3 to <32 x i1>
  %5 = select <32 x i1> %4, <32 x half> %3, <32 x half> %x0
  ret <32 x half> %5
}

define <32 x half>@test_int_x86_avx512_mask3_vfnmsub_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask3_vfnmsub_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmsub231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xbe,0xd1]
; X86-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfnmsub_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmsub231ph %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf6,0x7d,0x49,0xbe,0xd1]
; X64-NEXT:    vmovaps %zmm2, %zmm0 # encoding: [0x62,0xf1,0x7c,0x48,0x28,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %1, <32 x half> %x1, <32 x half> %2)
  %4 = bitcast i32 %x3 to <32 x i1>
  %5 = select <32 x i1> %4, <32 x half> %3, <32 x half> %x2
  ret <32 x half> %5
}

define <32 x half>@test_int_x86_avx512_mask_vfnmadd_ph_512(<32 x half> %x0, <32 x half> %x1, <32 x half> %x2, i32 %x3){
; X86-LABEL: test_int_x86_avx512_mask_vfnmadd_ph_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfnmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9c,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vfnmadd_ph_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfnmadd132ph %zmm1, %zmm2, %zmm0 {%k1} # encoding: [0x62,0xf6,0x6d,0x49,0x9c,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %x0, <32 x half> %1, <32 x half> %x2)
  %3 = bitcast i32 %x3 to <32 x i1>
  %4 = select <32 x i1> %3, <32 x half> %2, <32 x half> %x0
  ret <32 x half> %4
}

define <32 x half> @test_x86_fma_vfnmadd_ph_512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) #0 {
; CHECK-LABEL: test_x86_fma_vfnmadd_ph_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmadd213ph %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x48,0xac,0xc2]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %1, <32 x half> %a1, <32 x half> %a2)
  ret <32 x half> %2
}

define <32 x half> @test_x86_fma_vfnmsub_ph_512(<32 x half> %a0, <32 x half> %a1, <32 x half> %a2) #0 {
; CHECK-LABEL: test_x86_fma_vfnmsub_ph_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vfnmsub213ph %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf6,0x75,0x48,0xae,0xc0]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = fsub <32 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, 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 <32 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, 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 <32 x half> @llvm.fma.v32f16(<32 x half> %1, <32 x half> %a1, <32 x half> %2)
  ret <32 x half> %3
}

define <8 x half>@test_int_x86_avx512_mask3_vfmadd_sh(<8 x half> %x0, <8 x half> %x1, ptr%ptr_b, i8 %x3, i32 %x4) {
; X86-LABEL: test_int_x86_avx512_mask3_vfmadd_sh:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vfmadd231sh (%eax), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb9,0x08]
; X86-NEXT:    vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask3_vfmadd_sh:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vfmadd231sh (%rdi), %xmm0, %xmm1 {%k1} # encoding: [0x62,0xf6,0x7d,0x09,0xb9,0x0f]
; X64-NEXT:    vmovaps %xmm1, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xf8,0x28,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %q = load half, ptr %ptr_b
  %vecinit.i = insertelement <8 x half> undef, half %q, i32 0
  %1 = extractelement <8 x half> %x0, i64 0
  %2 = extractelement <8 x half> %vecinit.i, i64 0
  %3 = extractelement <8 x half> %x1, i64 0
  %4 = call half @llvm.fma.f16(half %1, half %2, half %3)
  %5 = bitcast i8 %x3 to <8 x i1>
  %6 = extractelement <8 x i1> %5, i64 0
  %7 = select i1 %6, half %4, half %3
  %8 = insertelement <8 x half> %x1, half %7, i64 0
  ret <8 x half> %8
}

define <8 x half>@test_int_x86_avx512_maskz_vfmadd_sh(<8 x half> %x0, <8 x half> %x1, <8 x half> %x2, i8 %x3, i32 %x4 ){
; X86-LABEL: test_int_x86_avx512_maskz_vfmadd_sh:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vfmadd213sh %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0x89,0xa9,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vfmadd_sh:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vfmadd213sh %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf6,0x75,0x89,0xa9,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = extractelement <8 x half> %x0, i64 0
  %2 = extractelement <8 x half> %x1, i64 0
  %3 = extractelement <8 x half> %x2, i64 0
  %4 = call half @llvm.fma.f16(half %1, half %2, half %3)
  %5 = bitcast i8 %x3 to <8 x i1>
  %6 = extractelement <8 x i1> %5, i64 0
  %7 = select i1 %6, half %4, half 0.000000e+00
  %8 = insertelement <8 x half> %x0, half %7, i64 0
  %9 = extractelement <8 x half> %x0, i64 0
  %10 = extractelement <8 x half> %x1, i64 0
  %11 = extractelement <8 x half> %x2, i64 0
  %12 = call half @llvm.x86.avx512fp16.vfmadd.f16(half %9, half %10, half %11, i32 3)
  %13 = bitcast i8 %x3 to <8 x i1>
  %14 = extractelement <8 x i1> %13, i64 0
  %15 = select i1 %14, half %12, half 0.000000e+00
  %16 = insertelement <8 x half> %x0, half %15, i64 0
  %res2 = fadd <8 x half> %8, %16
  ret <8 x half> %8
}

define void @fmadd_sh_mask_memfold(ptr %a, ptr %b, i8 %c) {
; X86-LABEL: fmadd_sh_mask_memfold:
; X86:       # %bb.0:
; X86-NEXT:    kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x0c]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x08]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %ecx # encoding: [0x8b,0x4c,0x24,0x04]
; X86-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    # encoding: [0x62,0xf5,0x7e,0x08,0x10,0x01]
; X86-NEXT:    vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X86-NEXT:    # encoding: [0x62,0xf5,0x7e,0x08,0x10,0x08]
; X86-NEXT:    vfmadd213sh %xmm0, %xmm0, %xmm1 # encoding: [0x62,0xf6,0x7d,0x08,0xa9,0xc8]
; X86-NEXT:    vmovsh %xmm1, %xmm0, %xmm0 {%k1} # encoding: [0x62,0xf5,0x7e,0x09,0x10,0xc1]
; X86-NEXT:    vmovsh %xmm0, (%ecx) # encoding: [0x62,0xf5,0x7e,0x08,0x11,0x01]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: fmadd_sh_mask_memfold:
; X64:       # %bb.0:
; X64-NEXT:    vmovsh {{.*#+}} xmm0 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT:    # encoding: [0x62,0xf5,0x7e,0x08,0x10,0x07]
; X64-NEXT:    vmovsh {{.*#+}} xmm1 = mem[0],zero,zero,zero,zero,zero,zero,zero
; X64-NEXT:    # encoding: [0x62,0xf5,0x7e,0x08,0x10,0x0e]
; X64-NEXT:    vfmadd213sh %xmm0, %xmm0, %xmm1 # encoding: [0x62,0xf6,0x7d,0x08,0xa9,0xc8]
; X64-NEXT:    kmovd %edx, %k1 # encoding: [0xc5,0xfb,0x92,0xca]
; X64-NEXT:    vmovsh %xmm1, %xmm0, %xmm0 {%k1} # encoding: [0x62,0xf5,0x7e,0x09,0x10,0xc1]
; X64-NEXT:    vmovsh %xmm0, (%rdi) # encoding: [0x62,0xf5,0x7e,0x08,0x11,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %a.val = load half, ptr %a
  %av0 = insertelement <8 x half> undef, half %a.val, i32 0
  %av1 = insertelement <8 x half> %av0, half 0.000000e+00, i32 1
  %av2 = insertelement <8 x half> %av1, half 0.000000e+00, i32 2
  %av3 = insertelement <8 x half> %av2, half 0.000000e+00, i32 3
  %av4 = insertelement <8 x half> %av3, half 0.000000e+00, i32 4
  %av5 = insertelement <8 x half> %av4, half 0.000000e+00, i32 5
  %av6 = insertelement <8 x half> %av5, half 0.000000e+00, i32 6
  %av  = insertelement <8 x half> %av6, half 0.000000e+00, i32 7

  %b.val = load half, ptr %b
  %bv0 = insertelement <8 x half> undef, half %b.val, i32 0
  %bv1 = insertelement <8 x half> %bv0, half 0.000000e+00, i32 1
  %bv2 = insertelement <8 x half> %bv1, half 0.000000e+00, i32 2
  %bv3 = insertelement <8 x half> %bv2, half 0.000000e+00, i32 3
  %bv4 = insertelement <8 x half> %bv3, half 0.000000e+00, i32 4
  %bv5 = insertelement <8 x half> %bv4, half 0.000000e+00, i32 5
  %bv6 = insertelement <8 x half> %bv5, half 0.000000e+00, i32 6
  %bv =  insertelement <8 x half> %bv6, half 0.000000e+00, i32 7
  %1 = extractelement <8 x half> %av, i64 0
  %2 = extractelement <8 x half> %bv, i64 0
  %3 = extractelement <8 x half> %av, i64 0
  %4 = call half @llvm.fma.f16(half %1, half %2, half %3)
  %5 = bitcast i8 %c to <8 x i1>
  %6 = extractelement <8 x i1> %5, i64 0
  %7 = select i1 %6, half %4, half %1
  %8 = insertelement <8 x half> %av, half %7, i64 0
  %sr = extractelement <8 x half> %8, i32 0
  store half %sr, ptr %a
  ret void
}

declare half @llvm.fma.f16(half, half, half)
declare half @llvm.x86.avx512fp16.vfmadd.f16(half, half, half, i32)

declare <32 x half> @llvm.fma.v32f16(<32 x half>, <32 x half>, <32 x half>)