; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=i686-unknown -mattr=+avx512cd,+avx512vl | FileCheck %s --check-prefixes=CHECK,X86
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx512cd,+avx512vl | FileCheck %s --check-prefixes=CHECK,X64

declare <4 x i32> @llvm.x86.avx512.mask.lzcnt.d.128(<4 x i32>, <4 x i32>, i8)

define <4 x i32>@test_int_x86_avx512_vplzcnt_d_128(<4 x i32> %x0, <4 x i32> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vplzcnt_d_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vplzcntd %xmm0, %xmm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <4 x i32> @llvm.x86.avx512.mask.lzcnt.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 -1)
  ret <4 x i32> %res
}

define <4 x i32>@test_int_x86_avx512_mask_vplzcnt_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vplzcnt_d_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vplzcntd %xmm0, %xmm1 {%k1}
; X86-NEXT:    vmovdqa %xmm1, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vplzcnt_d_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vplzcntd %xmm0, %xmm1 {%k1}
; X64-NEXT:    vmovdqa %xmm1, %xmm0
; X64-NEXT:    retq
  %res = call <4 x i32> @llvm.x86.avx512.mask.lzcnt.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2)
  ret <4 x i32> %res
}

define <4 x i32>@test_int_x86_avx512_maskz_vplzcnt_d_128(<4 x i32> %x0, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_maskz_vplzcnt_d_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vplzcntd %xmm0, %xmm0 {%k1} {z}
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_maskz_vplzcnt_d_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vplzcntd %xmm0, %xmm0 {%k1} {z}
; X64-NEXT:    retq
  %res = call <4 x i32> @llvm.x86.avx512.mask.lzcnt.d.128(<4 x i32> %x0, <4 x i32> zeroinitializer, i8 %x2)
  ret <4 x i32> %res
}

declare <8 x i32> @llvm.x86.avx512.mask.lzcnt.d.256(<8 x i32>, <8 x i32>, i8)

define <8 x i32>@test_int_x86_avx512_vplzcnt_d_256(<8 x i32> %x0, <8 x i32> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vplzcnt_d_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vplzcntd %ymm0, %ymm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <8 x i32> @llvm.x86.avx512.mask.lzcnt.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 -1)
  ret <8 x i32> %res
}

define <8 x i32>@test_int_x86_avx512_mask_vplzcnt_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vplzcnt_d_256:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vplzcntd %ymm0, %ymm1 {%k1}
; X86-NEXT:    vmovdqa %ymm1, %ymm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vplzcnt_d_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vplzcntd %ymm0, %ymm1 {%k1}
; X64-NEXT:    vmovdqa %ymm1, %ymm0
; X64-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.mask.lzcnt.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2)
  ret <8 x i32> %res
}

declare <2 x i64> @llvm.x86.avx512.mask.lzcnt.q.128(<2 x i64>, <2 x i64>, i8)

define <2 x i64>@test_int_x86_avx512_vplzcnt_q_128(<2 x i64> %x0, <2 x i64> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vplzcnt_q_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vplzcntq %xmm0, %xmm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <2 x i64> @llvm.x86.avx512.mask.lzcnt.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 -1)
  ret <2 x i64> %res
}

define <2 x i64>@test_int_x86_avx512_mask_vplzcnt_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vplzcnt_q_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vplzcntq %xmm0, %xmm1 {%k1}
; X86-NEXT:    vmovdqa %xmm1, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vplzcnt_q_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vplzcntq %xmm0, %xmm1 {%k1}
; X64-NEXT:    vmovdqa %xmm1, %xmm0
; X64-NEXT:    retq
  %res = call <2 x i64> @llvm.x86.avx512.mask.lzcnt.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2)
  ret <2 x i64> %res
}

declare <4 x i64> @llvm.x86.avx512.mask.lzcnt.q.256(<4 x i64>, <4 x i64>, i8)

define <4 x i64>@test_int_x86_avx512_vplzcnt_q_256(<4 x i64> %x0, <4 x i64> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vplzcnt_q_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vplzcntq %ymm0, %ymm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <4 x i64> @llvm.x86.avx512.mask.lzcnt.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 -1)
  ret <4 x i64> %res
}

define <4 x i64>@test_int_x86_avx512_mask_vplzcnt_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vplzcnt_q_256:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vplzcntq %ymm0, %ymm1 {%k1}
; X86-NEXT:    vmovdqa %ymm1, %ymm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vplzcnt_q_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vplzcntq %ymm0, %ymm1 {%k1}
; X64-NEXT:    vmovdqa %ymm1, %ymm0
; X64-NEXT:    retq
  %res = call <4 x i64> @llvm.x86.avx512.mask.lzcnt.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2)
  ret <4 x i64> %res
}

define <8 x i32> @test_x86_vbroadcastmw_256(i16 %a0) {
; X86-LABEL: test_x86_vbroadcastmw_256:
; X86:       # %bb.0:
; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    vpbroadcastd %eax, %ymm0
; X86-NEXT:    retl
;
; X64-LABEL: test_x86_vbroadcastmw_256:
; X64:       # %bb.0:
; X64-NEXT:    movzwl %di, %eax
; X64-NEXT:    vpbroadcastd %eax, %ymm0
; X64-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.broadcastmw.256(i16 %a0) ;
  ret <8 x i32> %res
}
declare <8 x i32> @llvm.x86.avx512.broadcastmw.256(i16)

define <4 x i32> @test_x86_vbroadcastmw_128(i16 %a0) {
; X86-LABEL: test_x86_vbroadcastmw_128:
; X86:       # %bb.0:
; X86-NEXT:    movzwl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    vpbroadcastd %eax, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: test_x86_vbroadcastmw_128:
; X64:       # %bb.0:
; X64-NEXT:    movzwl %di, %eax
; X64-NEXT:    vpbroadcastd %eax, %xmm0
; X64-NEXT:    retq
  %res = call <4 x i32> @llvm.x86.avx512.broadcastmw.128(i16 %a0) ;
  ret <4 x i32> %res
}
declare <4 x i32> @llvm.x86.avx512.broadcastmw.128(i16)

define <4 x i64> @test_x86_broadcastmb_256(i8 %a0) {
; X86-LABEL: test_x86_broadcastmb_256:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    vpbroadcastq %xmm0, %ymm0
; X86-NEXT:    retl
;
; X64-LABEL: test_x86_broadcastmb_256:
; X64:       # %bb.0:
; X64-NEXT:    # kill: def $edi killed $edi def $rdi
; X64-NEXT:    movzbl %dil, %eax
; X64-NEXT:    vpbroadcastq %rax, %ymm0
; X64-NEXT:    retq
  %res = call <4 x i64> @llvm.x86.avx512.broadcastmb.256(i8 %a0) ;
  ret <4 x i64> %res
}
declare <4 x i64> @llvm.x86.avx512.broadcastmb.256(i8)

define <2 x i64> @test_x86_broadcastmb_128(i8 %a0) {
; X86-LABEL: test_x86_broadcastmb_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    vmovd %eax, %xmm0
; X86-NEXT:    vpbroadcastq %xmm0, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: test_x86_broadcastmb_128:
; X64:       # %bb.0:
; X64-NEXT:    # kill: def $edi killed $edi def $rdi
; X64-NEXT:    movzbl %dil, %eax
; X64-NEXT:    vpbroadcastq %rax, %xmm0
; X64-NEXT:    retq
  %res = call <2 x i64> @llvm.x86.avx512.broadcastmb.128(i8 %a0) ;
  ret <2 x i64> %res
}
declare <2 x i64> @llvm.x86.avx512.broadcastmb.128(i8)

declare <4 x i32> @llvm.x86.avx512.mask.conflict.d.128(<4 x i32>, <4 x i32>, i8)

define <4 x i32> @test_int_x86_avx512_vpconflict_d_128(<4 x i32> %x0, <4 x i32> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vpconflict_d_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpconflictd %xmm0, %xmm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <4 x i32> @llvm.x86.avx512.mask.conflict.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 -1)
  ret <4 x i32> %res
}

define <4 x i32> @test_int_x86_avx512_mask_vpconflict_d_128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vpconflict_d_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictd %xmm0, %xmm1 {%k1}
; X86-NEXT:    vmovdqa %xmm1, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vpconflict_d_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictd %xmm0, %xmm1 {%k1}
; X64-NEXT:    vmovdqa %xmm1, %xmm0
; X64-NEXT:    retq
  %res = call <4 x i32> @llvm.x86.avx512.mask.conflict.d.128(<4 x i32> %x0, <4 x i32> %x1, i8 %x2)
  ret <4 x i32> %res
}

define <4 x i32> @test_int_x86_avx512_maskz_vpconflict_d_128(<4 x i32> %x0, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_maskz_vpconflict_d_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictd %xmm0, %xmm0 {%k1} {z}
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_maskz_vpconflict_d_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictd %xmm0, %xmm0 {%k1} {z}
; X64-NEXT:    retq
  %res = call <4 x i32> @llvm.x86.avx512.mask.conflict.d.128(<4 x i32> %x0, <4 x i32> zeroinitializer, i8 %x2)
  ret <4 x i32> %res
}

declare <8 x i32> @llvm.x86.avx512.mask.conflict.d.256(<8 x i32>, <8 x i32>, i8)

define <8 x i32> @test_int_x86_avx512_vpconflict_d_256(<8 x i32> %x0, <8 x i32> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vpconflict_d_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpconflictd %ymm0, %ymm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <8 x i32> @llvm.x86.avx512.mask.conflict.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 -1)
  ret <8 x i32> %res
}

define <8 x i32> @test_int_x86_avx512_mask_vpconflict_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vpconflict_d_256:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictd %ymm0, %ymm1 {%k1}
; X86-NEXT:    vmovdqa %ymm1, %ymm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vpconflict_d_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictd %ymm0, %ymm1 {%k1}
; X64-NEXT:    vmovdqa %ymm1, %ymm0
; X64-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.mask.conflict.d.256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2)
  ret <8 x i32> %res
}

define <8 x i32> @test_int_x86_avx512_maskz_vpconflict_d_256(<8 x i32> %x0, <8 x i32> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_maskz_vpconflict_d_256:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictd %ymm0, %ymm0 {%k1} {z}
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_maskz_vpconflict_d_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictd %ymm0, %ymm0 {%k1} {z}
; X64-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.mask.conflict.d.256(<8 x i32> %x0, <8 x i32> zeroinitializer, i8 %x2)
  ret <8 x i32> %res
}

declare <2 x i64> @llvm.x86.avx512.mask.conflict.q.128(<2 x i64>, <2 x i64>, i8)

define <2 x i64> @test_int_x86_avx512_vpconflict_q_128(<2 x i64> %x0, <2 x i64> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vpconflict_q_128:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpconflictq %xmm0, %xmm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <2 x i64> @llvm.x86.avx512.mask.conflict.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 -1)
  ret <2 x i64> %res
}

define <2 x i64> @test_int_x86_avx512_mask_vpconflict_q_128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vpconflict_q_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictq %xmm0, %xmm1 {%k1}
; X86-NEXT:    vmovdqa %xmm1, %xmm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vpconflict_q_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictq %xmm0, %xmm1 {%k1}
; X64-NEXT:    vmovdqa %xmm1, %xmm0
; X64-NEXT:    retq
  %res = call <2 x i64> @llvm.x86.avx512.mask.conflict.q.128(<2 x i64> %x0, <2 x i64> %x1, i8 %x2)
  ret <2 x i64> %res
}

define <2 x i64> @test_int_x86_avx512_maskz_vpconflict_q_128(<2 x i64> %x0, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_maskz_vpconflict_q_128:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictq %xmm0, %xmm0 {%k1} {z}
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_maskz_vpconflict_q_128:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictq %xmm0, %xmm0 {%k1} {z}
; X64-NEXT:    retq
  %res = call <2 x i64> @llvm.x86.avx512.mask.conflict.q.128(<2 x i64> %x0, <2 x i64> zeroinitializer, i8 %x2)
  ret <2 x i64> %res
}

declare <4 x i64> @llvm.x86.avx512.mask.conflict.q.256(<4 x i64>, <4 x i64>, i8)

define <4 x i64> @test_int_x86_avx512_vpconflict_q_256(<4 x i64> %x0, <4 x i64> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vpconflict_q_256:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpconflictq %ymm0, %ymm0
; CHECK-NEXT:    ret{{[l|q]}}
  %res = call <4 x i64> @llvm.x86.avx512.mask.conflict.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 -1)
  ret <4 x i64> %res
}

define <4 x i64> @test_int_x86_avx512_mask_vpconflict_q_256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_vpconflict_q_256:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictq %ymm0, %ymm1 {%k1}
; X86-NEXT:    vmovdqa %ymm1, %ymm0
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_mask_vpconflict_q_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictq %ymm0, %ymm1 {%k1}
; X64-NEXT:    vmovdqa %ymm1, %ymm0
; X64-NEXT:    retq
  %res = call <4 x i64> @llvm.x86.avx512.mask.conflict.q.256(<4 x i64> %x0, <4 x i64> %x1, i8 %x2)
  ret <4 x i64> %res
}

define <4 x i64> @test_int_x86_avx512_maskz_vpconflict_q_256(<4 x i64> %x0, i8 %x2) {
; X86-LABEL: test_int_x86_avx512_maskz_vpconflict_q_256:
; X86:       # %bb.0:
; X86-NEXT:    movzbl {{[0-9]+}}(%esp), %eax
; X86-NEXT:    kmovw %eax, %k1
; X86-NEXT:    vpconflictq %ymm0, %ymm0 {%k1} {z}
; X86-NEXT:    retl
;
; X64-LABEL: test_int_x86_avx512_maskz_vpconflict_q_256:
; X64:       # %bb.0:
; X64-NEXT:    kmovw %edi, %k1
; X64-NEXT:    vpconflictq %ymm0, %ymm0 {%k1} {z}
; X64-NEXT:    retq
  %res = call <4 x i64> @llvm.x86.avx512.mask.conflict.q.256(<4 x i64> %x0, <4 x i64> zeroinitializer, i8 %x2)
  ret <4 x i64> %res
}