; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -disable-peephole -mtriple=i686-unknown-unknown -mattr=+avx512bw --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X86
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512bw --show-mc-encoding | FileCheck %s --check-prefixes=CHECK,X64

define i32 @test_int_x86_avx512_kadd_d(<32 x i16> %A, <32 x i16> %B) nounwind {
; CHECK-LABEL: test_int_x86_avx512_kadd_d:
; CHECK:       # %bb.0: # %entry
; CHECK-NEXT:    vptestmw %zmm0, %zmm0, %k0 # encoding: [0x62,0xf2,0xfd,0x48,0x26,0xc0]
; CHECK-NEXT:    vptestmw %zmm1, %zmm1, %k1 # encoding: [0x62,0xf2,0xf5,0x48,0x26,0xc9]
; CHECK-NEXT:    kaddd %k1, %k0, %k0 # encoding: [0xc4,0xe1,0xfd,0x4a,0xc1]
; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
; CHECK-NEXT:    kortestd %k0, %k0 # encoding: [0xc4,0xe1,0xf9,0x98,0xc0]
; CHECK-NEXT:    sete %al # encoding: [0x0f,0x94,0xc0]
; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
entry:
  %0 = icmp ne <32 x i16> %A, zeroinitializer
  %1 = icmp ne <32 x i16> %B, zeroinitializer
  %2 = call <32 x i1> @llvm.x86.avx512.kadd.d(<32 x i1> %0, <32 x i1> %1)
  %3 = bitcast <32 x i1> %2 to i32
  %4 = icmp eq i32 %3, 0
  %5 = zext i1 %4 to i32
  ret i32 %5
}
declare <32 x i1> @llvm.x86.avx512.kadd.d(<32 x i1>, <32 x i1>)

define i32 @test_int_x86_avx512_kadd_q(<64 x i8> %A, <64 x i8> %B) nounwind {
; X86-LABEL: test_int_x86_avx512_kadd_q:
; X86:       # %bb.0: # %entry
; X86-NEXT:    vptestmb %zmm0, %zmm0, %k0 # encoding: [0x62,0xf2,0x7d,0x48,0x26,0xc0]
; X86-NEXT:    vptestmb %zmm1, %zmm1, %k1 # encoding: [0x62,0xf2,0x75,0x48,0x26,0xc9]
; X86-NEXT:    kaddq %k1, %k0, %k0 # encoding: [0xc4,0xe1,0xfc,0x4a,0xc1]
; X86-NEXT:    kshiftrq $32, %k0, %k1 # encoding: [0xc4,0xe3,0xf9,0x31,0xc8,0x20]
; X86-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
; X86-NEXT:    kortestd %k1, %k0 # encoding: [0xc4,0xe1,0xf9,0x98,0xc1]
; X86-NEXT:    sete %al # encoding: [0x0f,0x94,0xc0]
; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_kadd_q:
; X64:       # %bb.0: # %entry
; X64-NEXT:    vptestmb %zmm0, %zmm0, %k0 # encoding: [0x62,0xf2,0x7d,0x48,0x26,0xc0]
; X64-NEXT:    vptestmb %zmm1, %zmm1, %k1 # encoding: [0x62,0xf2,0x75,0x48,0x26,0xc9]
; X64-NEXT:    kaddq %k1, %k0, %k0 # encoding: [0xc4,0xe1,0xfc,0x4a,0xc1]
; X64-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
; X64-NEXT:    kortestq %k0, %k0 # encoding: [0xc4,0xe1,0xf8,0x98,0xc0]
; X64-NEXT:    sete %al # encoding: [0x0f,0x94,0xc0]
; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT:    retq # encoding: [0xc3]
entry:
  %0 = icmp ne <64 x i8> %A, zeroinitializer
  %1 = icmp ne <64 x i8> %B, zeroinitializer
  %2 = call <64 x i1> @llvm.x86.avx512.kadd.q(<64 x i1> %0, <64 x i1> %1)
  %3 = bitcast <64 x i1> %2 to i64
  %4 = icmp eq i64 %3, 0
  %5 = zext i1 %4 to i32
  ret i32 %5
}
declare <64 x i1> @llvm.x86.avx512.kadd.q(<64 x i1>, <64 x i1>)

define i32 @test_x86_avx512_ktestc_d(<32 x i16> %A, <32 x i16> %B) {
; CHECK-LABEL: test_x86_avx512_ktestc_d:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vptestmw %zmm0, %zmm0, %k0 # encoding: [0x62,0xf2,0xfd,0x48,0x26,0xc0]
; CHECK-NEXT:    vptestmw %zmm1, %zmm1, %k1 # encoding: [0x62,0xf2,0xf5,0x48,0x26,0xc9]
; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
; CHECK-NEXT:    ktestd %k1, %k0 # encoding: [0xc4,0xe1,0xf9,0x99,0xc1]
; CHECK-NEXT:    setb %al # encoding: [0x0f,0x92,0xc0]
; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = icmp ne <32 x i16> %A, zeroinitializer
  %2 = icmp ne <32 x i16> %B, zeroinitializer
  %res = call i32 @llvm.x86.avx512.ktestc.d(<32 x i1> %1, <32 x i1> %2) ; <i32> [#uses=1]
  ret i32 %res
}
declare i32 @llvm.x86.avx512.ktestc.d(<32 x i1>, <32 x i1>) nounwind readnone

define i32 @test_x86_avx512_ktestz_d(<32 x i16> %A, <32 x i16> %B) {
; CHECK-LABEL: test_x86_avx512_ktestz_d:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vptestmw %zmm0, %zmm0, %k0 # encoding: [0x62,0xf2,0xfd,0x48,0x26,0xc0]
; CHECK-NEXT:    vptestmw %zmm1, %zmm1, %k1 # encoding: [0x62,0xf2,0xf5,0x48,0x26,0xc9]
; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
; CHECK-NEXT:    ktestd %k1, %k0 # encoding: [0xc4,0xe1,0xf9,0x99,0xc1]
; CHECK-NEXT:    sete %al # encoding: [0x0f,0x94,0xc0]
; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = icmp ne <32 x i16> %A, zeroinitializer
  %2 = icmp ne <32 x i16> %B, zeroinitializer
  %res = call i32 @llvm.x86.avx512.ktestz.d(<32 x i1> %1, <32 x i1> %2) ; <i32> [#uses=1]
  ret i32 %res
}
declare i32 @llvm.x86.avx512.ktestz.d(<32 x i1>, <32 x i1>) nounwind readnone

define i32 @test_x86_avx512_ktestc_q(<64 x i8> %A, <64 x i8> %B) {
; CHECK-LABEL: test_x86_avx512_ktestc_q:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vptestmb %zmm0, %zmm0, %k0 # encoding: [0x62,0xf2,0x7d,0x48,0x26,0xc0]
; CHECK-NEXT:    vptestmb %zmm1, %zmm1, %k1 # encoding: [0x62,0xf2,0x75,0x48,0x26,0xc9]
; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
; CHECK-NEXT:    ktestq %k1, %k0 # encoding: [0xc4,0xe1,0xf8,0x99,0xc1]
; CHECK-NEXT:    setb %al # encoding: [0x0f,0x92,0xc0]
; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = icmp ne <64 x i8> %A, zeroinitializer
  %2 = icmp ne <64 x i8> %B, zeroinitializer
  %res = call i32 @llvm.x86.avx512.ktestc.q(<64 x i1> %1, <64 x i1> %2) ; <i32> [#uses=1]
  ret i32 %res
}
declare i32 @llvm.x86.avx512.ktestc.q(<64 x i1>, <64 x i1>) nounwind readnone

define i32 @test_x86_avx512_ktestz_q(<64 x i8> %A, <64 x i8> %B) {
; CHECK-LABEL: test_x86_avx512_ktestz_q:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vptestmb %zmm0, %zmm0, %k0 # encoding: [0x62,0xf2,0x7d,0x48,0x26,0xc0]
; CHECK-NEXT:    vptestmb %zmm1, %zmm1, %k1 # encoding: [0x62,0xf2,0x75,0x48,0x26,0xc9]
; CHECK-NEXT:    xorl %eax, %eax # encoding: [0x31,0xc0]
; CHECK-NEXT:    ktestq %k1, %k0 # encoding: [0xc4,0xe1,0xf8,0x99,0xc1]
; CHECK-NEXT:    sete %al # encoding: [0x0f,0x94,0xc0]
; CHECK-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = icmp ne <64 x i8> %A, zeroinitializer
  %2 = icmp ne <64 x i8> %B, zeroinitializer
  %res = call i32 @llvm.x86.avx512.ktestz.q(<64 x i1> %1, <64 x i1> %2) ; <i32> [#uses=1]
  ret i32 %res
}
declare i32 @llvm.x86.avx512.ktestz.q(<64 x i1>, <64 x i1>) nounwind readnone

define <32 x i16> @test_mask_packs_epi32_rr_512(<16 x i32> %a, <16 x i32> %b) {
; CHECK-LABEL: test_mask_packs_epi32_rr_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpackssdw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x6b,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  ret <32 x i16> %1
}

define <32 x i16> @test_mask_packs_epi32_rrk_512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask) {
; X86-LABEL: test_mask_packs_epi32_rrk_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpackssdw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x6b,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rrk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpackssdw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x6b,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packs_epi32_rrkz_512(<16 x i32> %a, <16 x i32> %b, i32 %mask) {
; X86-LABEL: test_mask_packs_epi32_rrkz_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpackssdw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x6b,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rrkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpackssdw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x6b,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packs_epi32_rm_512(<16 x i32> %a, <16 x i32>* %ptr_b) {
; X86-LABEL: test_mask_packs_epi32_rm_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpackssdw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x6b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rm_512:
; X64:       # %bb.0:
; X64-NEXT:    vpackssdw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x6b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  ret <32 x i16> %1
}

define <32 x i16> @test_mask_packs_epi32_rmk_512(<16 x i32> %a, <16 x i32>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
; X86-LABEL: test_mask_packs_epi32_rmk_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackssdw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x6b,0x08]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rmk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackssdw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x6b,0x0f]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packs_epi32_rmkz_512(<16 x i32> %a, <16 x i32>* %ptr_b, i32 %mask) {
; X86-LABEL: test_mask_packs_epi32_rmkz_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackssdw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x6b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rmkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackssdw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x6b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packs_epi32_rmb_512(<16 x i32> %a, i32* %ptr_b) {
; X86-LABEL: test_mask_packs_epi32_rmb_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpackssdw (%eax){1to16}, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x58,0x6b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rmb_512:
; X64:       # %bb.0:
; X64-NEXT:    vpackssdw (%rdi){1to16}, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x58,0x6b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %q = load i32, i32* %ptr_b
  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  ret <32 x i16> %1
}

define <32 x i16> @test_mask_packs_epi32_rmbk_512(<16 x i32> %a, i32* %ptr_b, <32 x i16> %passThru, i32 %mask) {
; X86-LABEL: test_mask_packs_epi32_rmbk_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackssdw (%eax){1to16}, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x59,0x6b,0x08]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rmbk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackssdw (%rdi){1to16}, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x59,0x6b,0x0f]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %q = load i32, i32* %ptr_b
  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packs_epi32_rmbkz_512(<16 x i32> %a, i32* %ptr_b, i32 %mask) {
; X86-LABEL: test_mask_packs_epi32_rmbkz_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackssdw (%eax){1to16}, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xd9,0x6b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi32_rmbkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackssdw (%rdi){1to16}, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xd9,0x6b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %q = load i32, i32* %ptr_b
  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
  %1 = call <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
  ret <32 x i16> %3
}

declare <32 x i16> @llvm.x86.avx512.packssdw.512(<16 x i32>, <16 x i32>)

define <64 x i8> @test_mask_packs_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
; CHECK-LABEL: test_mask_packs_epi16_rr_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpacksswb %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x63,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
  ret <64 x i8> %1
}

define <64 x i8> @test_mask_packs_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask) {
; X86-LABEL: test_mask_packs_epi16_rrk_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpacksswb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x63,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi16_rrk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
; X64-NEXT:    vpacksswb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x63,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
  ret <64 x i8> %3
}

define <64 x i8> @test_mask_packs_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i64 %mask) {
; X86-LABEL: test_mask_packs_epi16_rrkz_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpacksswb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x63,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi16_rrkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
; X64-NEXT:    vpacksswb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x63,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
  ret <64 x i8> %3
}

define <64 x i8> @test_mask_packs_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
; X86-LABEL: test_mask_packs_epi16_rm_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpacksswb (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x63,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi16_rm_512:
; X64:       # %bb.0:
; X64-NEXT:    vpacksswb (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x63,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <32 x i16>, <32 x i16>* %ptr_b
  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
  ret <64 x i8> %1
}

define <64 x i8> @test_mask_packs_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <64 x i8> %passThru, i64 %mask) {
; X86-LABEL: test_mask_packs_epi16_rmk_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpacksswb (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x63,0x08]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi16_rmk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
; X64-NEXT:    vpacksswb (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x63,0x0f]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <32 x i16>, <32 x i16>* %ptr_b
  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
  ret <64 x i8> %3
}

define <64 x i8> @test_mask_packs_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i64 %mask) {
; X86-LABEL: test_mask_packs_epi16_rmkz_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpacksswb (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x63,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packs_epi16_rmkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
; X64-NEXT:    vpacksswb (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x63,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <32 x i16>, <32 x i16>* %ptr_b
  %1 = call <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
  ret <64 x i8> %3
}

declare <64 x i8> @llvm.x86.avx512.packsswb.512(<32 x i16>, <32 x i16>)


define <32 x i16> @test_mask_packus_epi32_rr_512(<16 x i32> %a, <16 x i32> %b) {
; CHECK-LABEL: test_mask_packus_epi32_rr_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpackusdw %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7d,0x48,0x2b,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  ret <32 x i16> %1
}

define <32 x i16> @test_mask_packus_epi32_rrk_512(<16 x i32> %a, <16 x i32> %b, <32 x i16> %passThru, i32 %mask) {
; X86-LABEL: test_mask_packus_epi32_rrk_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpackusdw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x2b,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rrk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpackusdw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x2b,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packus_epi32_rrkz_512(<16 x i32> %a, <16 x i32> %b, i32 %mask) {
; X86-LABEL: test_mask_packus_epi32_rrkz_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpackusdw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x2b,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rrkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpackusdw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x2b,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packus_epi32_rm_512(<16 x i32> %a, <16 x i32>* %ptr_b) {
; X86-LABEL: test_mask_packus_epi32_rm_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpackusdw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7d,0x48,0x2b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rm_512:
; X64:       # %bb.0:
; X64-NEXT:    vpackusdw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7d,0x48,0x2b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  ret <32 x i16> %1
}

define <32 x i16> @test_mask_packus_epi32_rmk_512(<16 x i32> %a, <16 x i32>* %ptr_b, <32 x i16> %passThru, i32 %mask) {
; X86-LABEL: test_mask_packus_epi32_rmk_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackusdw (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x2b,0x08]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rmk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackusdw (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x2b,0x0f]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packus_epi32_rmkz_512(<16 x i32> %a, <16 x i32>* %ptr_b, i32 %mask) {
; X86-LABEL: test_mask_packus_epi32_rmkz_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackusdw (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x2b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rmkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackusdw (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x2b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packus_epi32_rmb_512(<16 x i32> %a, i32* %ptr_b) {
; X86-LABEL: test_mask_packus_epi32_rmb_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpackusdw (%eax){1to16}, %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7d,0x58,0x2b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rmb_512:
; X64:       # %bb.0:
; X64-NEXT:    vpackusdw (%rdi){1to16}, %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7d,0x58,0x2b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %q = load i32, i32* %ptr_b
  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  ret <32 x i16> %1
}

define <32 x i16> @test_mask_packus_epi32_rmbk_512(<16 x i32> %a, i32* %ptr_b, <32 x i16> %passThru, i32 %mask) {
; X86-LABEL: test_mask_packus_epi32_rmbk_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackusdw (%eax){1to16}, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x59,0x2b,0x08]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rmbk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackusdw (%rdi){1to16}, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0x7d,0x59,0x2b,0x0f]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %q = load i32, i32* %ptr_b
  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %passThru
  ret <32 x i16> %3
}

define <32 x i16> @test_mask_packus_epi32_rmbkz_512(<16 x i32> %a, i32* %ptr_b, i32 %mask) {
; X86-LABEL: test_mask_packus_epi32_rmbkz_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackusdw (%eax){1to16}, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xd9,0x2b,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi32_rmbkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpackusdw (%rdi){1to16}, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xd9,0x2b,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %q = load i32, i32* %ptr_b
  %vecinit.i = insertelement <16 x i32> undef, i32 %q, i32 0
  %b = shufflevector <16 x i32> %vecinit.i, <16 x i32> undef, <16 x i32> zeroinitializer
  %1 = call <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32> %a, <16 x i32> %b)
  %2 = bitcast i32 %mask to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
  ret <32 x i16> %3
}

declare <32 x i16> @llvm.x86.avx512.packusdw.512(<16 x i32>, <16 x i32>)

define <64 x i8> @test_mask_packus_epi16_rr_512(<32 x i16> %a, <32 x i16> %b) {
; CHECK-LABEL: test_mask_packus_epi16_rr_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x67,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
  ret <64 x i8> %1
}

define <64 x i8> @test_mask_packus_epi16_rrk_512(<32 x i16> %a, <32 x i16> %b, <64 x i8> %passThru, i64 %mask) {
; X86-LABEL: test_mask_packus_epi16_rrk_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpackuswb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x67,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi16_rrk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
; X64-NEXT:    vpackuswb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x67,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
  ret <64 x i8> %3
}

define <64 x i8> @test_mask_packus_epi16_rrkz_512(<32 x i16> %a, <32 x i16> %b, i64 %mask) {
; X86-LABEL: test_mask_packus_epi16_rrkz_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x67,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi16_rrkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
; X64-NEXT:    vpackuswb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x67,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
  ret <64 x i8> %3
}

define <64 x i8> @test_mask_packus_epi16_rm_512(<32 x i16> %a, <32 x i16>* %ptr_b) {
; X86-LABEL: test_mask_packus_epi16_rm_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpackuswb (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x67,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi16_rm_512:
; X64:       # %bb.0:
; X64-NEXT:    vpackuswb (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x67,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <32 x i16>, <32 x i16>* %ptr_b
  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
  ret <64 x i8> %1
}

define <64 x i8> @test_mask_packus_epi16_rmk_512(<32 x i16> %a, <32 x i16>* %ptr_b, <64 x i8> %passThru, i64 %mask) {
; X86-LABEL: test_mask_packus_epi16_rmk_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackuswb (%eax), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x67,0x08]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi16_rmk_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
; X64-NEXT:    vpackuswb (%rdi), %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0x67,0x0f]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <32 x i16>, <32 x i16>* %ptr_b
  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %passThru
  ret <64 x i8> %3
}

define <64 x i8> @test_mask_packus_epi16_rmkz_512(<32 x i16> %a, <32 x i16>* %ptr_b, i64 %mask) {
; X86-LABEL: test_mask_packus_epi16_rmkz_512:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x08]
; X86-NEXT:    vpackuswb (%eax), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x67,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_mask_packus_epi16_rmkz_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rsi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xce]
; X64-NEXT:    vpackuswb (%rdi), %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x67,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %b = load <32 x i16>, <32 x i16>* %ptr_b
  %1 = call <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16> %a, <32 x i16> %b)
  %2 = bitcast i64 %mask to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> zeroinitializer
  ret <64 x i8> %3
}

declare <64 x i8> @llvm.x86.avx512.packuswb.512(<32 x i16>, <32 x i16>)

define <32 x i16>@test_int_x86_avx512_mask_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_vpermt2var_hi_512:
; X86:       # %bb.0:
; X86-NEXT:    vmovdqa64 %zmm1, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd9]
; X86-NEXT:    vpermt2w %zmm2, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x7d,0xda]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpermt2w %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x7d,0xca]
; X86-NEXT:    vpaddw %zmm3, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpermt2var_hi_512:
; X64:       # %bb.0:
; X64-NEXT:    vmovdqa64 %zmm1, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd9]
; X64-NEXT:    vpermt2w %zmm2, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x7d,0xda]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpermt2w %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x7d,0xca]
; X64-NEXT:    vpaddw %zmm3, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x1, <32 x i16> %x0, <32 x i16> %x2)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x1
  %4 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x1, <32 x i16> %x0, <32 x i16> %x2)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

define <32 x i16>@test_int_x86_avx512_maskz_vpermt2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_maskz_vpermt2var_hi_512:
; X86:       # %bb.0:
; X86-NEXT:    vmovdqa64 %zmm1, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd9]
; X86-NEXT:    vpermt2w %zmm2, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x7d,0xda]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpermt2w %zmm2, %zmm0, %zmm1 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x7d,0xca]
; X86-NEXT:    vpaddw %zmm3, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_maskz_vpermt2var_hi_512:
; X64:       # %bb.0:
; X64-NEXT:    vmovdqa64 %zmm1, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd9]
; X64-NEXT:    vpermt2w %zmm2, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x7d,0xda]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpermt2w %zmm2, %zmm0, %zmm1 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x7d,0xca]
; X64-NEXT:    vpaddw %zmm3, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x1, <32 x i16> %x0, <32 x i16> %x2)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> zeroinitializer
  %4 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x1, <32 x i16> %x0, <32 x i16> %x2)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

declare <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16>, <32 x i16>, <32 x i16>)

define <32 x i16>@test_int_x86_avx512_mask_vpermi2var_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_vpermi2var_hi_512:
; X86:       # %bb.0:
; X86-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
; X86-NEXT:    vpermt2w %zmm2, %zmm1, %zmm3 # encoding: [0x62,0xf2,0xf5,0x48,0x7d,0xda]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpermi2w %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x75,0xca]
; X86-NEXT:    vpaddw %zmm3, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_vpermi2var_hi_512:
; X64:       # %bb.0:
; X64-NEXT:    vmovdqa64 %zmm0, %zmm3 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xd8]
; X64-NEXT:    vpermt2w %zmm2, %zmm1, %zmm3 # encoding: [0x62,0xf2,0xf5,0x48,0x7d,0xda]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpermi2w %zmm2, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x75,0xca]
; X64-NEXT:    vpaddw %zmm3, %zmm1, %zmm0 # encoding: [0x62,0xf1,0x75,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x1
  %4 = call <32 x i16> @llvm.x86.avx512.vpermi2var.hi.512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

declare <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8>, <64 x i8>)

define <64 x i8> @test_int_x86_avx512_mask_pavg_b_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_pavg_b_512:
; X86:       # %bb.0:
; X86-NEXT:    vpavgb %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe0,0xd9]
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpavgb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe0,0xd1]
; X86-NEXT:    vpaddb %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfc,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pavg_b_512:
; X64:       # %bb.0:
; X64-NEXT:    vpavgb %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe0,0xd9]
; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
; X64-NEXT:    vpavgb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe0,0xd1]
; X64-NEXT:    vpaddb %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfc,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> %x0, <64 x i8> %x1)
  %2 = bitcast i64 %x3 to <64 x i1>
  %3 = select <64 x i1> %2, <64 x i8> %1, <64 x i8> %x2
  %4 = call <64 x i8> @llvm.x86.avx512.pavg.b.512(<64 x i8> %x0, <64 x i8> %x1)
  %res2 = add <64 x i8> %3, %4
  ret <64 x i8> %res2
}

declare <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16>, <32 x i16>)

define <32 x i16> @test_int_x86_avx512_mask_pavg_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_pavg_w_512:
; X86:       # %bb.0:
; X86-NEXT:    vpavgw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe3,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpavgw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe3,0xd1]
; X86-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pavg_w_512:
; X64:       # %bb.0:
; X64-NEXT:    vpavgw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe3,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpavgw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe3,0xd1]
; X64-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.pavg.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

declare <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8>, <64 x i8>)

define <64 x i8>@test_int_x86_avx512_pshuf_b_512(<64 x i8> %x0, <64 x i8> %x1) {
; CHECK-LABEL: test_int_x86_avx512_pshuf_b_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpshufb %zmm1, %zmm0, %zmm0 # encoding: [0x62,0xf2,0x7d,0x48,0x00,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1)
  ret <64 x i8> %res
}

define <64 x i8>@test_int_x86_avx512_pshuf_b_512_mask(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2, i64 %mask) {
; X86-LABEL: test_int_x86_avx512_pshuf_b_512_mask:
; X86:       # %bb.0:
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpshufb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x00,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_pshuf_b_512_mask:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
; X64-NEXT:    vpshufb %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x00,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1)
  %mask.cast = bitcast i64 %mask to <64 x i1>
  %res2 = select <64 x i1> %mask.cast, <64 x i8> %res, <64 x i8> %x2
  ret <64 x i8> %res2
}

define <64 x i8>@test_int_x86_avx512_pshuf_b_512_maskz(<64 x i8> %x0, <64 x i8> %x1, i64 %mask) {
; X86-LABEL: test_int_x86_avx512_pshuf_b_512_maskz:
; X86:       # %bb.0:
; X86-NEXT:    kmovq {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpshufb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x00,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_pshuf_b_512_maskz:
; X64:       # %bb.0:
; X64-NEXT:    kmovq %rdi, %k1 # encoding: [0xc4,0xe1,0xfb,0x92,0xcf]
; X64-NEXT:    vpshufb %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x7d,0xc9,0x00,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <64 x i8> @llvm.x86.avx512.pshuf.b.512(<64 x i8> %x0, <64 x i8> %x1)
  %mask.cast = bitcast i64 %mask to <64 x i1>
  %res2 = select <64 x i1> %mask.cast, <64 x i8> %res, <64 x i8> zeroinitializer
  ret <64 x i8> %res2
}

declare <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16>, <32 x i16>)

define <32 x i16> @test_int_x86_avx512_mask_pmulhu_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_pmulhu_w_512:
; X86:       # %bb.0:
; X86-NEXT:    vpmulhuw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe4,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmulhuw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe4,0xd1]
; X86-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmulhu_w_512:
; X64:       # %bb.0:
; X64-NEXT:    vpmulhuw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe4,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmulhuw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe4,0xd1]
; X64-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.pmulhu.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

declare <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16>, <32 x i16>)

define <32 x i16> @test_int_x86_avx512_mask_pmulh_w_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_pmulh_w_512:
; X86:       # %bb.0:
; X86-NEXT:    vpmulhw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe5,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmulhw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe5,0xd1]
; X86-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmulh_w_512:
; X64:       # %bb.0:
; X64-NEXT:    vpmulhw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xe5,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmulhw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe5,0xd1]
; X64-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.pmulh.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

declare <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16>, <32 x i16>)

define <32 x i16> @test_int_x86_avx512_mask_pmulhr_sw_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_pmulhr_sw_512:
; X86:       # %bb.0:
; X86-NEXT:    vpmulhrsw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0x7d,0x48,0x0b,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmulhrsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x0b,0xd1]
; X86-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmulhr_sw_512:
; X64:       # %bb.0:
; X64-NEXT:    vpmulhrsw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0x7d,0x48,0x0b,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmulhrsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x0b,0xd1]
; X64-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.pmul.hr.sw.512(<32 x i16> %x0, <32 x i16> %x1)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

define <32 x i8>@test_int_x86_avx512_mask_pmov_wb_512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_pmov_wb_512:
; X86:       # %bb.0:
; X86-NEXT:    vpmovwb %zmm0, %ymm2 # encoding: [0x62,0xf2,0x7e,0x48,0x30,0xc2]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmovwb %zmm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x30,0xc1]
; X86-NEXT:    vpmovwb %zmm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0xc9,0x30,0xc0]
; X86-NEXT:    vpaddb %ymm0, %ymm1, %ymm0 # encoding: [0xc5,0xf5,0xfc,0xc0]
; X86-NEXT:    vpaddb %ymm0, %ymm2, %ymm0 # encoding: [0xc5,0xed,0xfc,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmov_wb_512:
; X64:       # %bb.0:
; X64-NEXT:    vpmovwb %zmm0, %ymm2 # encoding: [0x62,0xf2,0x7e,0x48,0x30,0xc2]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmovwb %zmm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x30,0xc1]
; X64-NEXT:    vpmovwb %zmm0, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0xc9,0x30,0xc0]
; X64-NEXT:    vpaddb %ymm0, %ymm1, %ymm0 # encoding: [0xc5,0xf5,0xfc,0xc0]
; X64-NEXT:    vpaddb %ymm0, %ymm2, %ymm0 # encoding: [0xc5,0xed,0xfc,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = trunc <32 x i16> %x0 to <32 x i8>
  %2 = trunc <32 x i16> %x0 to <32 x i8>
  %3 = bitcast i32 %x2 to <32 x i1>
  %4 = select <32 x i1> %3, <32 x i8> %2, <32 x i8> %x1
  %5 = trunc <32 x i16> %x0 to <32 x i8>
  %6 = bitcast i32 %x2 to <32 x i1>
  %7 = select <32 x i1> %6, <32 x i8> %5, <32 x i8> zeroinitializer
  %res3 = add <32 x i8> %1, %4
  %res4 = add <32 x i8> %res3, %7
  ret <32 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmov.wb.mem.512(i8* %ptr, <32 x i16>, i32)

define void @test_int_x86_avx512_mask_pmov_wb_mem_512(i8* %ptr, <32 x i16> %x1, i32 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_pmov_wb_mem_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpmovwb %zmm0, (%eax) # encoding: [0x62,0xf2,0x7e,0x48,0x30,0x00]
; X86-NEXT:    vpmovwb %zmm0, (%eax) {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x30,0x00]
; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmov_wb_mem_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpmovwb %zmm0, (%rdi) # encoding: [0x62,0xf2,0x7e,0x48,0x30,0x07]
; X64-NEXT:    vpmovwb %zmm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x30,0x07]
; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT:    retq # encoding: [0xc3]
    call void @llvm.x86.avx512.mask.pmov.wb.mem.512(i8* %ptr, <32 x i16> %x1, i32 -1)
    call void @llvm.x86.avx512.mask.pmov.wb.mem.512(i8* %ptr, <32 x i16> %x1, i32 %x2)
    ret void
}

declare <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16>, <32 x i8>, i32)

define <32 x i8>@test_int_x86_avx512_mask_pmovs_wb_512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_pmovs_wb_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmovswb %zmm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x20,0xc1]
; X86-NEXT:    vpmovswb %zmm0, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0xc9,0x20,0xc2]
; X86-NEXT:    vpaddb %ymm2, %ymm1, %ymm1 # encoding: [0xc5,0xf5,0xfc,0xca]
; X86-NEXT:    vpmovswb %zmm0, %ymm0 # encoding: [0x62,0xf2,0x7e,0x48,0x20,0xc0]
; X86-NEXT:    vpaddb %ymm1, %ymm0, %ymm0 # encoding: [0xc5,0xfd,0xfc,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmovs_wb_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmovswb %zmm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x20,0xc1]
; X64-NEXT:    vpmovswb %zmm0, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0xc9,0x20,0xc2]
; X64-NEXT:    vpaddb %ymm2, %ymm1, %ymm1 # encoding: [0xc5,0xf5,0xfc,0xca]
; X64-NEXT:    vpmovswb %zmm0, %ymm0 # encoding: [0x62,0xf2,0x7e,0x48,0x20,0xc0]
; X64-NEXT:    vpaddb %ymm1, %ymm0, %ymm0 # encoding: [0xc5,0xfd,0xfc,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
    %res0 = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 -1)
    %res1 = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2)
    %res2 = call <32 x i8> @llvm.x86.avx512.mask.pmovs.wb.512(<32 x i16> %x0, <32 x i8> zeroinitializer, i32 %x2)
    %res3 = add <32 x i8> %res0, %res1
    %res4 = add <32 x i8> %res3, %res2
    ret <32 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmovs.wb.mem.512(i8* %ptr, <32 x i16>, i32)

define void @test_int_x86_avx512_mask_pmovs_wb_mem_512(i8* %ptr, <32 x i16> %x1, i32 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_pmovs_wb_mem_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpmovswb %zmm0, (%eax) # encoding: [0x62,0xf2,0x7e,0x48,0x20,0x00]
; X86-NEXT:    vpmovswb %zmm0, (%eax) {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x20,0x00]
; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmovs_wb_mem_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpmovswb %zmm0, (%rdi) # encoding: [0x62,0xf2,0x7e,0x48,0x20,0x07]
; X64-NEXT:    vpmovswb %zmm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x20,0x07]
; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT:    retq # encoding: [0xc3]
    call void @llvm.x86.avx512.mask.pmovs.wb.mem.512(i8* %ptr, <32 x i16> %x1, i32 -1)
    call void @llvm.x86.avx512.mask.pmovs.wb.mem.512(i8* %ptr, <32 x i16> %x1, i32 %x2)
    ret void
}

declare <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16>, <32 x i8>, i32)

define <32 x i8>@test_int_x86_avx512_mask_pmovus_wb_512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_pmovus_wb_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmovuswb %zmm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x10,0xc1]
; X86-NEXT:    vpmovuswb %zmm0, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0xc9,0x10,0xc2]
; X86-NEXT:    vpaddb %ymm2, %ymm1, %ymm1 # encoding: [0xc5,0xf5,0xfc,0xca]
; X86-NEXT:    vpmovuswb %zmm0, %ymm0 # encoding: [0x62,0xf2,0x7e,0x48,0x10,0xc0]
; X86-NEXT:    vpaddb %ymm1, %ymm0, %ymm0 # encoding: [0xc5,0xfd,0xfc,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmovus_wb_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmovuswb %zmm0, %ymm1 {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x10,0xc1]
; X64-NEXT:    vpmovuswb %zmm0, %ymm2 {%k1} {z} # encoding: [0x62,0xf2,0x7e,0xc9,0x10,0xc2]
; X64-NEXT:    vpaddb %ymm2, %ymm1, %ymm1 # encoding: [0xc5,0xf5,0xfc,0xca]
; X64-NEXT:    vpmovuswb %zmm0, %ymm0 # encoding: [0x62,0xf2,0x7e,0x48,0x10,0xc0]
; X64-NEXT:    vpaddb %ymm1, %ymm0, %ymm0 # encoding: [0xc5,0xfd,0xfc,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
    %res0 = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 -1)
    %res1 = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> %x0, <32 x i8> %x1, i32 %x2)
    %res2 = call <32 x i8> @llvm.x86.avx512.mask.pmovus.wb.512(<32 x i16> %x0, <32 x i8> zeroinitializer, i32 %x2)
    %res3 = add <32 x i8> %res0, %res1
    %res4 = add <32 x i8> %res3, %res2
    ret <32 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmovus.wb.mem.512(i8* %ptr, <32 x i16>, i32)

define void @test_int_x86_avx512_mask_pmovus_wb_mem_512(i8* %ptr, <32 x i16> %x1, i32 %x2) {
; X86-LABEL: test_int_x86_avx512_mask_pmovus_wb_mem_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x08]
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpmovuswb %zmm0, (%eax) # encoding: [0x62,0xf2,0x7e,0x48,0x10,0x00]
; X86-NEXT:    vpmovuswb %zmm0, (%eax) {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x10,0x00]
; X86-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmovus_wb_mem_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %esi, %k1 # encoding: [0xc5,0xfb,0x92,0xce]
; X64-NEXT:    vpmovuswb %zmm0, (%rdi) # encoding: [0x62,0xf2,0x7e,0x48,0x10,0x07]
; X64-NEXT:    vpmovuswb %zmm0, (%rdi) {%k1} # encoding: [0x62,0xf2,0x7e,0x49,0x10,0x07]
; X64-NEXT:    vzeroupper # encoding: [0xc5,0xf8,0x77]
; X64-NEXT:    retq # encoding: [0xc3]
    call void @llvm.x86.avx512.mask.pmovus.wb.mem.512(i8* %ptr, <32 x i16> %x1, i32 -1)
    call void @llvm.x86.avx512.mask.pmovus.wb.mem.512(i8* %ptr, <32 x i16> %x1, i32 %x2)
    ret void
}

declare <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8>, <64 x i8>)

define <32 x i16> @test_int_x86_avx512_mask_pmaddubs_w_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_pmaddubs_w_512:
; X86:       # %bb.0:
; X86-NEXT:    vpmaddubsw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0x7d,0x48,0x04,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmaddubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x04,0xd1]
; X86-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmaddubs_w_512:
; X64:       # %bb.0:
; X64-NEXT:    vpmaddubsw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0x7d,0x48,0x04,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmaddubsw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0x7d,0x49,0x04,0xd1]
; X64-NEXT:    vpaddw %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> %x0, <64 x i8> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.pmaddubs.w.512(<64 x i8> %x0, <64 x i8> %x1)
  %res2 = add <32 x i16> %3, %4
  ret <32 x i16> %res2
}

declare <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16>, <32 x i16>)

define <16 x i32> @test_int_x86_avx512_mask_pmaddw_d_512(<32 x i16> %x0, <32 x i16> %x1, <16 x i32> %x2, i16 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_pmaddw_d_512:
; X86:       # %bb.0:
; X86-NEXT:    vpmaddwd %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xf5,0xd9]
; X86-NEXT:    kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpmaddwd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xf5,0xd1]
; X86-NEXT:    vpaddd %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfe,0xc3]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_pmaddw_d_512:
; X64:       # %bb.0:
; X64-NEXT:    vpmaddwd %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf1,0x7d,0x48,0xf5,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpmaddwd %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xf5,0xd1]
; X64-NEXT:    vpaddd %zmm3, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfe,0xc3]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i16 %x3 to <16 x i1>
  %3 = select <16 x i1> %2, <16 x i32> %1, <16 x i32> %x2
  %4 = call <16 x i32> @llvm.x86.avx512.pmaddw.d.512(<32 x i16> %x0, <32 x i16> %x1)
  %res2 = add <16 x i32> %3, %4
  ret <16 x i32> %res2
}

declare <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8>, <64 x i8>, i32)

define <32 x i16>@test_int_x86_avx512_mask_dbpsadbw_512(<64 x i8> %x0, <64 x i8> %x1, <32 x i16> %x3, i32 %x4) {
; X86-LABEL: test_int_x86_avx512_mask_dbpsadbw_512:
; X86:       # %bb.0:
; X86-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf3,0x7d,0x48,0x42,0xd9,0x02]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x49,0x42,0xd1,0x02]
; X86-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xc9,0x42,0xc1,0x02]
; X86-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X86-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_dbpsadbw_512:
; X64:       # %bb.0:
; X64-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf3,0x7d,0x48,0x42,0xd9,0x02]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf3,0x7d,0x49,0x42,0xd1,0x02]
; X64-NEXT:    vdbpsadbw $2, %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf3,0x7d,0xc9,0x42,0xc1,0x02]
; X64-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X64-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2)
  %2 = bitcast i32 %x4 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x3
  %4 = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2)
  %5 = bitcast i32 %x4 to <32 x i1>
  %6 = select <32 x i1> %5, <32 x i16> %4, <32 x i16> zeroinitializer
  %7 = call <32 x i16> @llvm.x86.avx512.dbpsadbw.512(<64 x i8> %x0, <64 x i8> %x1, i32 2)
  %res3 = add <32 x i16> %3, %6
  %res4 = add <32 x i16> %res3, %7
  ret <32 x i16> %res4
}

declare  <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8>, <64 x i8>)

define  <8 x i64>@test_int_x86_avx512_mask_psadb_w_512(<64 x i8> %x0, <64 x i8> %x1, <64 x i8> %x2){
; CHECK-LABEL: test_int_x86_avx512_mask_psadb_w_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpsadbw %zmm1, %zmm0, %zmm1 # encoding: [0x62,0xf1,0x7d,0x48,0xf6,0xc9]
; CHECK-NEXT:    vpsadbw %zmm2, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xf6,0xc2]
; CHECK-NEXT:    vpaddq %zmm0, %zmm1, %zmm0 # encoding: [0x62,0xf1,0xf5,0x48,0xd4,0xc0]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call  <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8> %x0, <64 x i8> %x1)
  %res1 = call  <8 x i64> @llvm.x86.avx512.psad.bw.512(<64 x i8> %x0, <64 x i8> %x2)
  %res2 = add  <8 x i64> %res, %res1
  ret  <8 x i64> %res2
}

declare <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16>, <32 x i16>) nounwind readnone

define <32 x i16> @test_x86_avx512_psrlv_w_512_const() optsize {
; X86-LABEL: test_x86_avx512_psrlv_w_512_const:
; X86:       # %bb.0:
; X86-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,65535]
; X86-NEXT:    # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0x05,A,A,A,A]
; X86-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}, kind: FK_Data_4
; X86-NEXT:    vpsrlvw {{\.LCPI.*}}, %zmm0, %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x10,0x05,A,A,A,A]
; X86-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}, kind: FK_Data_4
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_psrlv_w_512_const:
; X64:       # %bb.0:
; X64-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,65535]
; X64-NEXT:    # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0x05,A,A,A,A]
; X64-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
; X64-NEXT:    vpsrlvw {{.*}}(%rip), %zmm0, %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x10,0x05,A,A,A,A]
; X64-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
; X64-NEXT:    retq # encoding: [0xc3]
  %res1 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> <i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 -1>, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,  i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 -1>)
  ret <32 x i16> %res1
}

define <32 x i16>@test_int_x86_avx512_mask_psrlv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_psrlv32hi:
; X86:       # %bb.0:
; X86-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x10,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x10,0xd1]
; X86-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x10,0xc1]
; X86-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X86-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_psrlv32hi:
; X64:       # %bb.0:
; X64-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x10,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x10,0xd1]
; X64-NEXT:    vpsrlvw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x10,0xc1]
; X64-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X64-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %5 = bitcast i32 %x3 to <32 x i1>
  %6 = select <32 x i1> %5, <32 x i16> %4, <32 x i16> zeroinitializer
  %7 = call <32 x i16> @llvm.x86.avx512.psrlv.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %res3 = add <32 x i16> %3, %6
  %res4 = add <32 x i16> %res3, %7
  ret <32 x i16> %res4
}

declare <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16>, <32 x i16>)

define <32 x i16>@test_int_x86_avx512_mask_psrav32_hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_psrav32_hi:
; X86:       # %bb.0:
; X86-NEXT:    vpsravw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x11,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsravw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x11,0xd1]
; X86-NEXT:    vpsravw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x11,0xc1]
; X86-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X86-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_psrav32_hi:
; X64:       # %bb.0:
; X64-NEXT:    vpsravw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x11,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsravw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x11,0xd1]
; X64-NEXT:    vpsravw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x11,0xc1]
; X64-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X64-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %5 = bitcast i32 %x3 to <32 x i1>
  %6 = select <32 x i1> %5, <32 x i16> %4, <32 x i16> zeroinitializer
  %7 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %res3 = add <32 x i16> %3, %6
  %res4 = add <32 x i16> %res3, %7
  ret <32 x i16> %res4
}

define <32 x i16>@test_int_x86_avx512_mask_psrav32_hi_const(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_psrav32_hi_const:
; X86:       # %bb.0:
; X86-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [2,9,65524,23,65510,37,65496,51,2,9,65524,23,65510,37,65496,51,2,9,65524,23,65510,37,65496,51,2,9,65524,23,65510,37,65496,51]
; X86-NEXT:    # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0x05,A,A,A,A]
; X86-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}, kind: FK_Data_4
; X86-NEXT:    vpsravw {{\.LCPI.*}}, %zmm0, %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x11,0x05,A,A,A,A]
; X86-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}, kind: FK_Data_4
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_psrav32_hi_const:
; X64:       # %bb.0:
; X64-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [2,9,65524,23,65510,37,65496,51,2,9,65524,23,65510,37,65496,51,2,9,65524,23,65510,37,65496,51,2,9,65524,23,65510,37,65496,51]
; X64-NEXT:    # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0x05,A,A,A,A]
; X64-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
; X64-NEXT:    vpsravw {{.*}}(%rip), %zmm0, %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x11,0x05,A,A,A,A]
; X64-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.psrav.w.512(<32 x i16> <i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51, i16 2, i16 9, i16 -12, i16 23, i16 -26, i16 37, i16 -40, i16 51>, <32 x i16> <i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49, i16 1, i16 10, i16 35, i16 52, i16 69, i16 9, i16 16, i16 49>)
  ret <32 x i16> %1
}

define <32 x i16>@test_int_x86_avx512_mask_psllv32hi(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_psllv32hi:
; X86:       # %bb.0:
; X86-NEXT:    vpsllvw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x12,0xd9]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsllvw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x12,0xd1]
; X86-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x12,0xc1]
; X86-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X86-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_psllv32hi:
; X64:       # %bb.0:
; X64-NEXT:    vpsllvw %zmm1, %zmm0, %zmm3 # encoding: [0x62,0xf2,0xfd,0x48,0x12,0xd9]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsllvw %zmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf2,0xfd,0x49,0x12,0xd1]
; X64-NEXT:    vpsllvw %zmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xfd,0xc9,0x12,0xc1]
; X64-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X64-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %5 = bitcast i32 %x3 to <32 x i1>
  %6 = select <32 x i1> %5, <32 x i16> %4, <32 x i16> zeroinitializer
  %7 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> %x0, <32 x i16> %x1)
  %res3 = add <32 x i16> %3, %6
  %res4 = add <32 x i16> %res3, %7
  ret <32 x i16> %res4
}

declare <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16>, <32 x i16>)

define <32 x i16>@test_int_x86_avx512_mask_permvar_hi_512(<32 x i16> %x0, <32 x i16> %x1, <32 x i16> %x2, i32 %x3) {
; X86-LABEL: test_int_x86_avx512_mask_permvar_hi_512:
; X86:       # %bb.0:
; X86-NEXT:    vpermw %zmm0, %zmm1, %zmm3 # encoding: [0x62,0xf2,0xf5,0x48,0x8d,0xd8]
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpermw %zmm0, %zmm1, %zmm2 {%k1} # encoding: [0x62,0xf2,0xf5,0x49,0x8d,0xd0]
; X86-NEXT:    vpermw %zmm0, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0x8d,0xc0]
; X86-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X86-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_int_x86_avx512_mask_permvar_hi_512:
; X64:       # %bb.0:
; X64-NEXT:    vpermw %zmm0, %zmm1, %zmm3 # encoding: [0x62,0xf2,0xf5,0x48,0x8d,0xd8]
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpermw %zmm0, %zmm1, %zmm2 {%k1} # encoding: [0x62,0xf2,0xf5,0x49,0x8d,0xd0]
; X64-NEXT:    vpermw %zmm0, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0xf5,0xc9,0x8d,0xc0]
; X64-NEXT:    vpaddw %zmm3, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xfd,0xc3]
; X64-NEXT:    vpaddw %zmm0, %zmm2, %zmm0 # encoding: [0x62,0xf1,0x6d,0x48,0xfd,0xc0]
; X64-NEXT:    retq # encoding: [0xc3]
  %1 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1)
  %2 = bitcast i32 %x3 to <32 x i1>
  %3 = select <32 x i1> %2, <32 x i16> %1, <32 x i16> %x2
  %4 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1)
  %5 = bitcast i32 %x3 to <32 x i1>
  %6 = select <32 x i1> %5, <32 x i16> %4, <32 x i16> zeroinitializer
  %7 = call <32 x i16> @llvm.x86.avx512.permvar.hi.512(<32 x i16> %x0, <32 x i16> %x1)
  %res3 = add <32 x i16> %3, %6
  %res4 = add <32 x i16> %res3, %7
  ret <32 x i16> %res4
}

define <32 x i16> @test_x86_avx512_psll_w_512(<32 x i16> %a0, <8 x i16> %a1) {
; CHECK-LABEL: test_x86_avx512_psll_w_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpsllw %xmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xf1,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  ret <32 x i16> %res
}
define <32 x i16> @test_x86_avx512_mask_psll_w_512(<32 x i16> %a0, <8 x i16> %a1, <32 x i16> %passthru, i32 %mask) {
; X86-LABEL: test_x86_avx512_mask_psll_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsllw %xmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xf1,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_mask_psll_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsllw %xmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xf1,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
  ret <32 x i16> %res2
}
define <32 x i16> @test_x86_avx512_maskz_psll_w_512(<32 x i16> %a0, <8 x i16> %a1, i32 %mask) {
; X86-LABEL: test_x86_avx512_maskz_psll_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsllw %xmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xf1,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_maskz_psll_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsllw %xmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xf1,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
  ret <32 x i16> %res2
}
declare <32 x i16> @llvm.x86.avx512.psll.w.512(<32 x i16>, <8 x i16>) nounwind readnone


define <32 x i16> @test_x86_avx512_psllv_w_512_const() optsize {
; X86-LABEL: test_x86_avx512_psllv_w_512_const:
; X86:       # %bb.0:
; X86-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,65535]
; X86-NEXT:    # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0x05,A,A,A,A]
; X86-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}, kind: FK_Data_4
; X86-NEXT:    vpsllvw {{\.LCPI.*}}, %zmm0, %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x12,0x05,A,A,A,A]
; X86-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}, kind: FK_Data_4
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_psllv_w_512_const:
; X64:       # %bb.0:
; X64-NEXT:    vmovdqa64 {{.*#+}} zmm0 = [4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,4,65535]
; X64-NEXT:    # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0x05,A,A,A,A]
; X64-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
; X64-NEXT:    vpsllvw {{.*}}(%rip), %zmm0, %zmm0 # encoding: [0x62,0xf2,0xfd,0x48,0x12,0x05,A,A,A,A]
; X64-NEXT:    # fixup A - offset: 6, value: {{\.LCPI.*}}-4, kind: reloc_riprel_4byte
; X64-NEXT:    retq # encoding: [0xc3]
  %res1 = call <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16> <i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4, i16 4,  i16 4, i16 4, i16 4, i16 4, i16 4, i16 -1>, <32 x i16> <i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1, i16 1,  i16 1, i16 1, i16 -1>)
  ret <32 x i16> %res1
}
declare <32 x i16> @llvm.x86.avx512.psllv.w.512(<32 x i16>, <32 x i16>) nounwind readnone

define <32 x i16> @test_x86_avx512_pslli_w_512(<32 x i16> %a0) {
; CHECK-LABEL: test_x86_avx512_pslli_w_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpsllw $7, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x71,0xf0,0x07]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  ret <32 x i16> %res
}
define <32 x i16> @test_x86_avx512_mask_pslli_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) {
; X86-LABEL: test_x86_avx512_mask_pslli_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsllw $7, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x75,0x49,0x71,0xf0,0x07]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_mask_pslli_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsllw $7, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x75,0x49,0x71,0xf0,0x07]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
  ret <32 x i16> %res2
}
define <32 x i16> @test_x86_avx512_maskz_pslli_w_512(<32 x i16> %a0, i32 %mask) {
; X86-LABEL: test_x86_avx512_maskz_pslli_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsllw $7, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x71,0xf0,0x07]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_maskz_pslli_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsllw $7, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x71,0xf0,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
  ret <32 x i16> %res2
}
declare <32 x i16> @llvm.x86.avx512.pslli.w.512(<32 x i16>, i32) nounwind readnone


define <32 x i16> @test_x86_avx512_psra_w_512(<32 x i16> %a0, <8 x i16> %a1) {
; CHECK-LABEL: test_x86_avx512_psra_w_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpsraw %xmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xe1,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  ret <32 x i16> %res
}
define <32 x i16> @test_x86_avx512_mask_psra_w_512(<32 x i16> %a0, <8 x i16> %a1, <32 x i16> %passthru, i32 %mask) {
; X86-LABEL: test_x86_avx512_mask_psra_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsraw %xmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe1,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_mask_psra_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsraw %xmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xe1,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
  ret <32 x i16> %res2
}
define <32 x i16> @test_x86_avx512_maskz_psra_w_512(<32 x i16> %a0, <8 x i16> %a1, i32 %mask) {
; X86-LABEL: test_x86_avx512_maskz_psra_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsraw %xmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe1,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_maskz_psra_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsraw %xmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xe1,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
  ret <32 x i16> %res2
}
declare <32 x i16> @llvm.x86.avx512.psra.w.512(<32 x i16>, <8 x i16>) nounwind readnone


define <32 x i16> @test_x86_avx512_psrai_w_512(<32 x i16> %a0) {
; CHECK-LABEL: test_x86_avx512_psrai_w_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpsraw $7, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x71,0xe0,0x07]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  ret <32 x i16> %res
}
define <32 x i16> @test_x86_avx512_mask_psrai_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) {
; X86-LABEL: test_x86_avx512_mask_psrai_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsraw $7, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x75,0x49,0x71,0xe0,0x07]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_mask_psrai_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsraw $7, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x75,0x49,0x71,0xe0,0x07]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
  ret <32 x i16> %res2
}
define <32 x i16> @test_x86_avx512_maskz_psrai_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) {
; X86-LABEL: test_x86_avx512_maskz_psrai_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsraw $7, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x71,0xe0,0x07]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_maskz_psrai_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsraw $7, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x71,0xe0,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
  ret <32 x i16> %res2
}
declare <32 x i16> @llvm.x86.avx512.psrai.w.512(<32 x i16>, i32) nounwind readnone


define <32 x i16> @test_x86_avx512_psrl_w_512(<32 x i16> %a0, <8 x i16> %a1) {
; CHECK-LABEL: test_x86_avx512_psrl_w_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xd1,0xc1]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  ret <32 x i16> %res
}
define <32 x i16> @test_x86_avx512_mask_psrl_w_512(<32 x i16> %a0, <8 x i16> %a1, <32 x i16> %passthru, i32 %mask) {
; X86-LABEL: test_x86_avx512_mask_psrl_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsrlw %xmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xd1,0xd1]
; X86-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_mask_psrl_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsrlw %xmm1, %zmm0, %zmm2 {%k1} # encoding: [0x62,0xf1,0x7d,0x49,0xd1,0xd1]
; X64-NEXT:    vmovdqa64 %zmm2, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc2]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
  ret <32 x i16> %res2
}
define <32 x i16> @test_x86_avx512_maskz_psrl_w_512(<32 x i16> %a0, <8 x i16> %a1, i32 %mask) {
; X86-LABEL: test_x86_avx512_maskz_psrl_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xd1,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_maskz_psrl_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsrlw %xmm1, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0xd1,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
  ret <32 x i16> %res2
}
declare <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16>, <8 x i16>) nounwind readnone

define <32 x i16> @test_x86_avx512_psrl_w_512_load(<32 x i16> %a0, <8 x i16>* %p) {
; X86-LABEL: test_x86_avx512_psrl_w_512_load:
; X86:       # %bb.0:
; X86-NEXT:    movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04]
; X86-NEXT:    vpsrlw (%eax), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xd1,0x00]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_psrl_w_512_load:
; X64:       # %bb.0:
; X64-NEXT:    vpsrlw (%rdi), %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0xd1,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %a1 = load <8 x i16>, <8 x i16>* %p
  %res = call <32 x i16> @llvm.x86.avx512.psrl.w.512(<32 x i16> %a0, <8 x i16> %a1) ; <<32 x i16>> [#uses=1]
  ret <32 x i16> %res
}

define <32 x i16> @test_x86_avx512_psrli_w_512(<32 x i16> %a0) {
; CHECK-LABEL: test_x86_avx512_psrli_w_512:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpsrlw $7, %zmm0, %zmm0 # encoding: [0x62,0xf1,0x7d,0x48,0x71,0xd0,0x07]
; CHECK-NEXT:    ret{{[l|q]}} # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  ret <32 x i16> %res
}
define <32 x i16> @test_x86_avx512_mask_psrli_w_512(<32 x i16> %a0, <32 x i16> %passthru, i32 %mask) {
; X86-LABEL: test_x86_avx512_mask_psrli_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsrlw $7, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x75,0x49,0x71,0xd0,0x07]
; X86-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_mask_psrli_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsrlw $7, %zmm0, %zmm1 {%k1} # encoding: [0x62,0xf1,0x75,0x49,0x71,0xd0,0x07]
; X64-NEXT:    vmovdqa64 %zmm1, %zmm0 # encoding: [0x62,0xf1,0xfd,0x48,0x6f,0xc1]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> %passthru
  ret <32 x i16> %res2
}
define <32 x i16> @test_x86_avx512_maskz_psrli_w_512(<32 x i16> %a0, i32 %mask) {
; X86-LABEL: test_x86_avx512_maskz_psrli_w_512:
; X86:       # %bb.0:
; X86-NEXT:    kmovd {{[0-9]+}}(%esp), %k1 # encoding: [0xc4,0xe1,0xf9,0x90,0x4c,0x24,0x04]
; X86-NEXT:    vpsrlw $7, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x71,0xd0,0x07]
; X86-NEXT:    retl # encoding: [0xc3]
;
; X64-LABEL: test_x86_avx512_maskz_psrli_w_512:
; X64:       # %bb.0:
; X64-NEXT:    kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf]
; X64-NEXT:    vpsrlw $7, %zmm0, %zmm0 {%k1} {z} # encoding: [0x62,0xf1,0x7d,0xc9,0x71,0xd0,0x07]
; X64-NEXT:    retq # encoding: [0xc3]
  %res = call <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16> %a0, i32 7) ; <<32 x i16>> [#uses=1]
  %mask.cast = bitcast i32 %mask to <32 x i1>
  %res2 = select <32 x i1> %mask.cast, <32 x i16> %res, <32 x i16> zeroinitializer
  ret <32 x i16> %res2
}
declare <32 x i16> @llvm.x86.avx512.psrli.w.512(<32 x i16>, i32) nounwind readnone