; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=knl | FileCheck %s

declare i32 @llvm.x86.avx512.kortestz.w(i16, i16) nounwind readnone
define i32 @test_kortestz(i16 %a0, i16 %a1) {
; CHECK-LABEL: test_kortestz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k0
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    xorl %eax, %eax
; CHECK-NEXT:    kortestw %k0, %k1
; CHECK-NEXT:    sete %al
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.kortestz.w(i16 %a0, i16 %a1)
  ret i32 %res
}

declare i32 @llvm.x86.avx512.kortestc.w(i16, i16) nounwind readnone
define i32 @test_kortestc(i16 %a0, i16 %a1) {
; CHECK-LABEL: test_kortestc:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k0
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    xorl %eax, %eax
; CHECK-NEXT:    kortestw %k0, %k1
; CHECK-NEXT:    setb %al
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.kortestc.w(i16 %a0, i16 %a1)
  ret i32 %res
}

declare i16 @llvm.x86.avx512.kand.w(i16, i16) nounwind readnone
define i16 @test_kand(i16 %a0, i16 %a1) {
; CHECK-LABEL: test_kand:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k0
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    movw $8, %ax
; CHECK-NEXT:    kmovw %eax, %k2
; CHECK-NEXT:    kandw %k0, %k1, %k0
; CHECK-NEXT:    kandw %k0, %k2, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %t1 = call i16 @llvm.x86.avx512.kand.w(i16 %a0, i16 8)
  %t2 = call i16 @llvm.x86.avx512.kand.w(i16 %t1, i16 %a1)
  ret i16 %t2
}

declare i16 @llvm.x86.avx512.kandn.w(i16, i16) nounwind readnone
define i16 @test_kandn(i16 %a0, i16 %a1) {
; CHECK-LABEL: test_kandn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k0
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    movw $8, %ax
; CHECK-NEXT:    kmovw %eax, %k2
; CHECK-NEXT:    kandnw %k2, %k1, %k1
; CHECK-NEXT:    kandnw %k0, %k1, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %t1 = call i16 @llvm.x86.avx512.kandn.w(i16 %a0, i16 8)
  %t2 = call i16 @llvm.x86.avx512.kandn.w(i16 %t1, i16 %a1)
  ret i16 %t2
}

declare i16 @llvm.x86.avx512.knot.w(i16) nounwind readnone
define i16 @test_knot(i16 %a0) {
; CHECK-LABEL: test_knot:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k0
; CHECK-NEXT:    knotw %k0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %res = call i16 @llvm.x86.avx512.knot.w(i16 %a0)
  ret i16 %res
}

declare i16 @llvm.x86.avx512.kor.w(i16, i16) nounwind readnone
define i16 @test_kor(i16 %a0, i16 %a1) {
; CHECK-LABEL: test_kor:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k0
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    movw $8, %ax
; CHECK-NEXT:    kmovw %eax, %k2
; CHECK-NEXT:    korw %k0, %k1, %k0
; CHECK-NEXT:    korw %k0, %k2, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %t1 = call i16 @llvm.x86.avx512.kor.w(i16 %a0, i16 8)
  %t2 = call i16 @llvm.x86.avx512.kor.w(i16 %t1, i16 %a1)
  ret i16 %t2
}

declare i16 @llvm.x86.avx512.kunpck.bw(i16, i16) nounwind readnone

define i16 @unpckbw_test(i16 %a0, i16 %a1) {
; CHECK-LABEL: unpckbw_test:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k0
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    kunpckbw %k1, %k0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %res = call i16 @llvm.x86.avx512.kunpck.bw(i16 %a0, i16 %a1)
  ret i16 %res
}

declare i16 @llvm.x86.avx512.kxnor.w(i16, i16) nounwind readnone
; TODO: the two kxnor instructions here a no op and should be elimintaed,
; probably by FoldConstantArithmetic in SelectionDAG.
define i16 @test_kxnor(i16 %a0, i16 %a1) {
; CHECK-LABEL: test_kxnor:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k0
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    movw $8, %ax
; CHECK-NEXT:    kmovw %eax, %k2
; CHECK-NEXT:    kxorw %k0, %k1, %k0
; CHECK-NEXT:    kxorw %k0, %k2, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %t1 = call i16 @llvm.x86.avx512.kxnor.w(i16 %a0, i16 8)
  %t2 = call i16 @llvm.x86.avx512.kxnor.w(i16 %t1, i16 %a1)
  ret i16 %t2
}

declare i16 @llvm.x86.avx512.kxor.w(i16, i16) nounwind readnone
define i16 @test_kxor(i16 %a0, i16 %a1) {
; CHECK-LABEL: test_kxor:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k0
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    movw $8, %ax
; CHECK-NEXT:    kmovw %eax, %k2
; CHECK-NEXT:    kxorw %k0, %k1, %k0
; CHECK-NEXT:    kxorw %k0, %k2, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %t1 = call i16 @llvm.x86.avx512.kxor.w(i16 %a0, i16 8)
  %t2 = call i16 @llvm.x86.avx512.kxor.w(i16 %t1, i16 %a1)
  ret i16 %t2
}

define <16 x float> @test_rcp_ps_512(<16 x float> %a0) {
; CHECK-LABEL: test_rcp_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vrcp14ps %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.rcp14.ps.512(<16 x float> %a0, <16 x float> zeroinitializer, i16 -1) ; <<16 x float>> [#uses=1]
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.rcp14.ps.512(<16 x float>, <16 x float>, i16) nounwind readnone

define <8 x double> @test_rcp_pd_512(<8 x double> %a0) {
; CHECK-LABEL: test_rcp_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vrcp14pd %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.rcp14.pd.512(<8 x double> %a0, <8 x double> zeroinitializer, i8 -1) ; <<8 x double>> [#uses=1]
  ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.rcp14.pd.512(<8 x double>, <8 x double>, i8) nounwind readnone

declare <8 x double> @llvm.x86.avx512.mask.rndscale.pd.512(<8 x double>, i32, <8 x double>, i8, i32)

define <8 x double> @test7(<8 x double> %a) {
; CHECK-LABEL: test7:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vrndscalepd $11, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.rndscale.pd.512(<8 x double> %a, i32 11, <8 x double> %a, i8 -1, i32 4)
  ret <8 x double>%res
}

declare <16 x float> @llvm.x86.avx512.mask.rndscale.ps.512(<16 x float>, i32, <16 x float>, i16, i32)

define <16 x float> @test8(<16 x float> %a) {
; CHECK-LABEL: test8:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vrndscaleps $11, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.rndscale.ps.512(<16 x float> %a, i32 11, <16 x float> %a, i16 -1, i32 4)
  ret <16 x float>%res
}

define <16 x float> @test_rsqrt_ps_512(<16 x float> %a0) {
; CHECK-LABEL: test_rsqrt_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vrsqrt14ps %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.rsqrt14.ps.512(<16 x float> %a0, <16 x float> zeroinitializer, i16 -1) ; <<16 x float>> [#uses=1]
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.rsqrt14.ps.512(<16 x float>, <16 x float>, i16) nounwind readnone

define <8 x double> @test_sqrt_pd_512(<8 x double> %a0) {
; CHECK-LABEL: test_sqrt_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsqrtpd %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.sqrt.pd.512(<8 x double> %a0,  <8 x double> zeroinitializer, i8 -1, i32 4)
  ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.mask.sqrt.pd.512(<8 x double>, <8 x double>, i8, i32) nounwind readnone

define <16 x float> @test_sqrt_ps_512(<16 x float> %a0) {
; CHECK-LABEL: test_sqrt_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsqrtps %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sqrt.ps.512(<16 x float> %a0, <16 x float> zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}
define <16 x float> @test_sqrt_round_ps_512(<16 x float> %a0) {
; CHECK-LABEL: test_sqrt_round_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsqrtps {rz-sae}, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sqrt.ps.512(<16 x float> %a0, <16 x float> zeroinitializer, i16 -1, i32 3)
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.sqrt.ps.512(<16 x float>, <16 x float>, i16, i32) nounwind readnone

define <8 x double> @test_getexp_pd_512(<8 x double> %a0) {
; CHECK-LABEL: test_getexp_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vgetexppd %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.getexp.pd.512(<8 x double> %a0,  <8 x double> zeroinitializer, i8 -1, i32 4)
  ret <8 x double> %res
}
define <8 x double> @test_getexp_round_pd_512(<8 x double> %a0) {
; CHECK-LABEL: test_getexp_round_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vgetexppd {sae}, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.getexp.pd.512(<8 x double> %a0,  <8 x double> zeroinitializer, i8 -1, i32 8)
  ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.mask.getexp.pd.512(<8 x double>, <8 x double>, i8, i32) nounwind readnone

define <16 x float> @test_getexp_ps_512(<16 x float> %a0) {
; CHECK-LABEL: test_getexp_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vgetexpps %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.getexp.ps.512(<16 x float> %a0, <16 x float> zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_getexp_round_ps_512(<16 x float> %a0) {
; CHECK-LABEL: test_getexp_round_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vgetexpps {sae}, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.getexp.ps.512(<16 x float> %a0, <16 x float> zeroinitializer, i16 -1, i32 8)
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.getexp.ps.512(<16 x float>, <16 x float>, i16, i32) nounwind readnone

declare <4 x float> @llvm.x86.avx512.mask.sqrt.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) nounwind readnone

define <4 x float> @test_sqrt_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_sqrt_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm2, %xmm3
; CHECK-NEXT:    vsqrtss %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT:    vsqrtss {rd-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vsqrtss {ru-sae}, %xmm1, %xmm0, %xmm4 {%k1} {z}
; CHECK-NEXT:    vsqrtss {rz-sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vaddps %xmm2, %xmm3, %xmm1
; CHECK-NEXT:    vaddps %xmm0, %xmm4, %xmm0
; CHECK-NEXT:    vaddps %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    retq
  %res0 = call <4 x float> @llvm.x86.avx512.mask.sqrt.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask.sqrt.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 1)
  %res2 = call <4 x float> @llvm.x86.avx512.mask.sqrt.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 %mask, i32 2)
  %res3 = call <4 x float> @llvm.x86.avx512.mask.sqrt.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 -1, i32 3)

  %res.1 = fadd <4 x float> %res0, %res1
  %res.2 = fadd <4 x float> %res2, %res3
  %res   = fadd <4 x float> %res.1, %res.2
  ret <4 x float> %res
}

declare <2 x double> @llvm.x86.avx512.mask.sqrt.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) nounwind readnone

define <2 x double> @test_sqrt_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_sqrt_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm2, %xmm3
; CHECK-NEXT:    vsqrtsd %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT:    vsqrtsd {rd-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vsqrtsd {ru-sae}, %xmm1, %xmm0, %xmm4 {%k1} {z}
; CHECK-NEXT:    vsqrtsd {rz-sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vaddpd %xmm2, %xmm3, %xmm1
; CHECK-NEXT:    vaddpd %xmm0, %xmm4, %xmm0
; CHECK-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    retq
  %res0 = call <2 x double> @llvm.x86.avx512.mask.sqrt.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask.sqrt.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 1)
  %res2 = call <2 x double> @llvm.x86.avx512.mask.sqrt.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 %mask, i32 2)
  %res3 = call <2 x double> @llvm.x86.avx512.mask.sqrt.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 -1, i32 3)

  %res.1 = fadd <2 x double> %res0, %res1
  %res.2 = fadd <2 x double> %res2, %res3
  %res   = fadd <2 x double> %res.1, %res.2
  ret <2 x double> %res
}

define i64 @test_x86_sse2_cvtsd2si64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_sse2_cvtsd2si64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsd2si %xmm0, %rax
; CHECK-NEXT:    retq
  %res = call i64 @llvm.x86.sse2.cvtsd2si64(<2 x double> %a0) ; <i64> [#uses=1]
  ret i64 %res
}
declare i64 @llvm.x86.sse2.cvtsd2si64(<2 x double>) nounwind readnone

define <2 x double> @test_x86_sse2_cvtsi642sd(<2 x double> %a0, i64 %a1) {
; CHECK-LABEL: test_x86_sse2_cvtsi642sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsi2sdq %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.sse2.cvtsi642sd(<2 x double> %a0, i64 %a1) ; <<2 x double>> [#uses=1]
  ret <2 x double> %res
}
declare <2 x double> @llvm.x86.sse2.cvtsi642sd(<2 x double>, i64) nounwind readnone

define i64 @test_x86_avx512_cvttsd2si64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttsd2si64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttsd2si %xmm0, %rcx
; CHECK-NEXT:    vcvttsd2si {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttsd2si64(<2 x double> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttsd2si64(<2 x double> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttsd2si64(<2 x double>, i32) nounwind readnone

define i32 @test_x86_avx512_cvttsd2usi(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttsd2usi:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttsd2usi %xmm0, %ecx
; CHECK-NEXT:    vcvttsd2usi {sae}, %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq
  %res0 = call i32 @llvm.x86.avx512.cvttsd2usi(<2 x double> %a0, i32 4) ;
  %res1 = call i32 @llvm.x86.avx512.cvttsd2usi(<2 x double> %a0, i32 8) ;
  %res2 = add i32 %res0, %res1
  ret i32 %res2
}
declare i32 @llvm.x86.avx512.cvttsd2usi(<2 x double>, i32) nounwind readnone

define i32 @test_x86_avx512_cvttsd2si(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttsd2si:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttsd2si %xmm0, %ecx
; CHECK-NEXT:    vcvttsd2si {sae}, %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq
  %res0 = call i32 @llvm.x86.avx512.cvttsd2si(<2 x double> %a0, i32 4) ;
  %res1 = call i32 @llvm.x86.avx512.cvttsd2si(<2 x double> %a0, i32 8) ;
  %res2 = add i32 %res0, %res1
  ret i32 %res2
}
declare i32 @llvm.x86.avx512.cvttsd2si(<2 x double>, i32) nounwind readnone


define i64 @test_x86_avx512_cvttsd2usi64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttsd2usi64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttsd2usi %xmm0, %rcx
; CHECK-NEXT:    vcvttsd2usi {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttsd2usi64(<2 x double> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttsd2usi64(<2 x double> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttsd2usi64(<2 x double>, i32) nounwind readnone

define i64 @test_x86_sse_cvtss2si64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_sse_cvtss2si64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtss2si %xmm0, %rax
; CHECK-NEXT:    retq
  %res = call i64 @llvm.x86.sse.cvtss2si64(<4 x float> %a0) ; <i64> [#uses=1]
  ret i64 %res
}
declare i64 @llvm.x86.sse.cvtss2si64(<4 x float>) nounwind readnone


define <4 x float> @test_x86_sse_cvtsi642ss(<4 x float> %a0, i64 %a1) {
; CHECK-LABEL: test_x86_sse_cvtsi642ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsi2ssq %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.sse.cvtsi642ss(<4 x float> %a0, i64 %a1) ; <<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.sse.cvtsi642ss(<4 x float>, i64) nounwind readnone


define i32 @test_x86_avx512_cvttss2si(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttss2si:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttss2si {sae}, %xmm0, %ecx
; CHECK-NEXT:    vcvttss2si %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq
  %res0 = call i32 @llvm.x86.avx512.cvttss2si(<4 x float> %a0, i32 8) ;
  %res1 = call i32 @llvm.x86.avx512.cvttss2si(<4 x float> %a0, i32 4) ;
  %res2 = add i32 %res0, %res1
  ret i32 %res2
}
declare i32 @llvm.x86.avx512.cvttss2si(<4 x float>, i32) nounwind readnone

define i64 @test_x86_avx512_cvttss2si64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttss2si64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttss2si %xmm0, %rcx
; CHECK-NEXT:    vcvttss2si {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttss2si64(<4 x float> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttss2si64(<4 x float> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttss2si64(<4 x float>, i32) nounwind readnone

define i32 @test_x86_avx512_cvttss2usi(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttss2usi:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttss2usi {sae}, %xmm0, %ecx
; CHECK-NEXT:    vcvttss2usi %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq
  %res0 = call i32 @llvm.x86.avx512.cvttss2usi(<4 x float> %a0, i32 8) ;
  %res1 = call i32 @llvm.x86.avx512.cvttss2usi(<4 x float> %a0, i32 4) ;
  %res2 = add i32 %res0, %res1
  ret i32 %res2
}
declare i32 @llvm.x86.avx512.cvttss2usi(<4 x float>, i32) nounwind readnone

define i64 @test_x86_avx512_cvttss2usi64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvttss2usi64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvttss2usi %xmm0, %rcx
; CHECK-NEXT:    vcvttss2usi {sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq
  %res0 = call i64 @llvm.x86.avx512.cvttss2usi64(<4 x float> %a0, i32 4) ;
  %res1 = call i64 @llvm.x86.avx512.cvttss2usi64(<4 x float> %a0, i32 8) ;
  %res2 = add i64 %res0, %res1
  ret i64 %res2
}
declare i64 @llvm.x86.avx512.cvttss2usi64(<4 x float>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtsd2usi64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtsd2usi64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsd2usi %xmm0, %rax
; CHECK-NEXT:    vcvtsd2usi {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtsd2usi {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double> %a0, i32 3)
  %res2 = call i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double> %a0, i32 1)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtsd2usi64(<2 x double>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtsd2si64(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtsd2si64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsd2si %xmm0, %rax
; CHECK-NEXT:    vcvtsd2si {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtsd2si {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double> %a0, i32 3)
  %res2 = call i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double> %a0, i32 1)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtsd2si64(<2 x double>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtss2usi64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtss2usi64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtss2usi %xmm0, %rax
; CHECK-NEXT:    vcvtss2usi {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtss2usi {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float> %a0, i32 3)
  %res2 = call i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float> %a0, i32 1)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtss2usi64(<4 x float>, i32) nounwind readnone

define i64 @test_x86_avx512_cvtss2si64(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtss2si64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtss2si %xmm0, %rax
; CHECK-NEXT:    vcvtss2si {rz-sae}, %xmm0, %rcx
; CHECK-NEXT:    addq %rax, %rcx
; CHECK-NEXT:    vcvtss2si {rd-sae}, %xmm0, %rax
; CHECK-NEXT:    addq %rcx, %rax
; CHECK-NEXT:    retq

  %res = call i64 @llvm.x86.avx512.vcvtss2si64(<4 x float> %a0, i32 4)
  %res1 = call i64 @llvm.x86.avx512.vcvtss2si64(<4 x float> %a0, i32 3)
  %res2 = call i64 @llvm.x86.avx512.vcvtss2si64(<4 x float> %a0, i32 1)
  %res3 = add i64 %res, %res1
  %res4 = add i64 %res3, %res2
  ret i64 %res4
}
declare i64 @llvm.x86.avx512.vcvtss2si64(<4 x float>, i32) nounwind readnone

define i32 @test_x86_avx512_cvtsd2usi32(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtsd2usi32:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsd2usi %xmm0, %eax
; CHECK-NEXT:    vcvtsd2usi {rz-sae}, %xmm0, %ecx
; CHECK-NEXT:    addl %eax, %ecx
; CHECK-NEXT:    vcvtsd2usi {rd-sae}, %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq

  %res = call i32 @llvm.x86.avx512.vcvtsd2usi32(<2 x double> %a0, i32 4)
  %res1 = call i32 @llvm.x86.avx512.vcvtsd2usi32(<2 x double> %a0, i32 3)
  %res2 = call i32 @llvm.x86.avx512.vcvtsd2usi32(<2 x double> %a0, i32 1)
  %res3 = add i32 %res, %res1
  %res4 = add i32 %res3, %res2
  ret i32 %res4
}
declare i32 @llvm.x86.avx512.vcvtsd2usi32(<2 x double>, i32) nounwind readnone

define i32 @test_x86_avx512_cvtsd2si32(<2 x double> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtsd2si32:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsd2si %xmm0, %eax
; CHECK-NEXT:    vcvtsd2si {rz-sae}, %xmm0, %ecx
; CHECK-NEXT:    addl %eax, %ecx
; CHECK-NEXT:    vcvtsd2si {rd-sae}, %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq

  %res = call i32 @llvm.x86.avx512.vcvtsd2si32(<2 x double> %a0, i32 4)
  %res1 = call i32 @llvm.x86.avx512.vcvtsd2si32(<2 x double> %a0, i32 3)
  %res2 = call i32 @llvm.x86.avx512.vcvtsd2si32(<2 x double> %a0, i32 1)
  %res3 = add i32 %res, %res1
  %res4 = add i32 %res3, %res2
  ret i32 %res4
}
declare i32 @llvm.x86.avx512.vcvtsd2si32(<2 x double>, i32) nounwind readnone

define i32 @test_x86_avx512_cvtss2usi32(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtss2usi32:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtss2usi %xmm0, %eax
; CHECK-NEXT:    vcvtss2usi {rz-sae}, %xmm0, %ecx
; CHECK-NEXT:    addl %eax, %ecx
; CHECK-NEXT:    vcvtss2usi {rd-sae}, %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq

  %res = call i32 @llvm.x86.avx512.vcvtss2usi32(<4 x float> %a0, i32 4)
  %res1 = call i32 @llvm.x86.avx512.vcvtss2usi32(<4 x float> %a0, i32 3)
  %res2 = call i32 @llvm.x86.avx512.vcvtss2usi32(<4 x float> %a0, i32 1)
  %res3 = add i32 %res, %res1
  %res4 = add i32 %res3, %res2
  ret i32 %res4
}
declare i32 @llvm.x86.avx512.vcvtss2usi32(<4 x float>, i32) nounwind readnone

define i32 @test_x86_avx512_cvtss2si32(<4 x float> %a0) {
; CHECK-LABEL: test_x86_avx512_cvtss2si32:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtss2si %xmm0, %eax
; CHECK-NEXT:    vcvtss2si {rz-sae}, %xmm0, %ecx
; CHECK-NEXT:    addl %eax, %ecx
; CHECK-NEXT:    vcvtss2si {rd-sae}, %xmm0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    retq

  %res = call i32 @llvm.x86.avx512.vcvtss2si32(<4 x float> %a0, i32 4)
  %res1 = call i32 @llvm.x86.avx512.vcvtss2si32(<4 x float> %a0, i32 3)
  %res2 = call i32 @llvm.x86.avx512.vcvtss2si32(<4 x float> %a0, i32 1)
  %res3 = add i32 %res, %res1
  %res4 = add i32 %res3, %res2
  ret i32 %res4
}
declare i32 @llvm.x86.avx512.vcvtss2si32(<4 x float>, i32) nounwind readnone

define <16 x float> @test_x86_vcvtph2ps_512(<16 x i16> %a0) {
; CHECK-LABEL: test_x86_vcvtph2ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtph2ps %ymm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.vcvtph2ps.512(<16 x i16> %a0, <16 x float> zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_x86_vcvtph2ps_512_sae(<16 x i16> %a0) {
; CHECK-LABEL: test_x86_vcvtph2ps_512_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtph2ps {sae}, %ymm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.vcvtph2ps.512(<16 x i16> %a0, <16 x float> zeroinitializer, i16 -1, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_x86_vcvtph2ps_512_rrk(<16 x i16> %a0,<16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_vcvtph2ps_512_rrk:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtph2ps %ymm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.vcvtph2ps.512(<16 x i16> %a0, <16 x float> %a1, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_x86_vcvtph2ps_512_sae_rrkz(<16 x i16> %a0, i16 %mask) {
; CHECK-LABEL: test_x86_vcvtph2ps_512_sae_rrkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtph2ps {sae}, %ymm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.vcvtph2ps.512(<16 x i16> %a0, <16 x float> zeroinitializer, i16 %mask, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_x86_vcvtph2ps_512_rrkz(<16 x i16> %a0, i16 %mask) {
; CHECK-LABEL: test_x86_vcvtph2ps_512_rrkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtph2ps %ymm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.vcvtph2ps.512(<16 x i16> %a0, <16 x float> zeroinitializer, i16 %mask, i32 4)
  ret <16 x float> %res
}

declare <16 x float> @llvm.x86.avx512.mask.vcvtph2ps.512(<16 x i16>, <16 x float>, i16, i32) nounwind readonly

define <16 x i16> @test_x86_vcvtps2ph_256(<16 x float> %a0, <16 x i16> %src, i16 %mask, <16 x i16> * %dst) {
; CHECK-LABEL: test_x86_vcvtps2ph_256:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtps2ph $2, %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vcvtps2ph $2, %zmm0, %ymm2 {%k1} {z}
; CHECK-NEXT:    vpaddw %ymm1, %ymm2, %ymm1
; CHECK-NEXT:    vcvtps2ph $2, %zmm0, (%rsi)
; CHECK-NEXT:    vmovdqa %ymm1, %ymm0
; CHECK-NEXT:    retq
  %res1 = call <16 x i16> @llvm.x86.avx512.mask.vcvtps2ph.512(<16 x float> %a0, i32 2, <16 x i16> zeroinitializer, i16 -1)
  %res2 = call <16 x i16> @llvm.x86.avx512.mask.vcvtps2ph.512(<16 x float> %a0, i32 2, <16 x i16> zeroinitializer, i16 %mask)
  %res3 = call <16 x i16> @llvm.x86.avx512.mask.vcvtps2ph.512(<16 x float> %a0, i32 2, <16 x i16> %src, i16 %mask)
  store <16 x i16> %res1, <16 x i16> * %dst
  %res  = add <16 x i16> %res2, %res3
  ret <16 x i16> %res
}

declare <16 x i16> @llvm.x86.avx512.mask.vcvtps2ph.512(<16 x float>, i32, <16 x i16>, i16) nounwind readonly

define <16 x float> @test_x86_vbroadcast_ss_512(i8* %a0) {
; CHECK-LABEL: test_x86_vbroadcast_ss_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.vbroadcast.ss.512(i8* %a0) ; <<16 x float>> [#uses=1]
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.vbroadcast.ss.512(i8*) nounwind readonly

define <8 x double> @test_x86_vbroadcast_sd_512(i8* %a0) {
; CHECK-LABEL: test_x86_vbroadcast_sd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.vbroadcast.sd.512(i8* %a0) ; <<8 x double>> [#uses=1]
  ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.vbroadcast.sd.512(i8*) nounwind readonly

 define i16 @test_cmpps(<16 x float> %a, <16 x float> %b) {
; CHECK-LABEL: test_cmpps:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpleps {sae}, %zmm1, %zmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
   %res = call i16 @llvm.x86.avx512.mask.cmp.ps.512(<16 x float> %a, <16 x float> %b, i32 2, i16 -1, i32 8)
   ret i16 %res
 }
 declare i16 @llvm.x86.avx512.mask.cmp.ps.512(<16 x float> , <16 x float> , i32, i16, i32)

 define i8 @test_cmppd(<8 x double> %a, <8 x double> %b) {
; CHECK-LABEL: test_cmppd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpneqpd %zmm1, %zmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
; CHECK-NEXT:    retq
   %res = call i8 @llvm.x86.avx512.mask.cmp.pd.512(<8 x double> %a, <8 x double> %b, i32 4, i8 -1, i32 4)
   ret i8 %res
 }
 declare i8 @llvm.x86.avx512.mask.cmp.pd.512(<8 x double> , <8 x double> , i32, i8, i32)

 ; fp min - max
define <8 x double> @test_vmaxpd(<8 x double> %a0, <8 x double> %a1) {
; CHECK-LABEL: test_vmaxpd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmaxpd %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.max.pd.512(<8 x double> %a0, <8 x double> %a1,
                    <8 x double>zeroinitializer, i8 -1, i32 4)
  ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.mask.max.pd.512(<8 x double>, <8 x double>,
                    <8 x double>, i8, i32)

define <8 x double> @test_vminpd(<8 x double> %a0, <8 x double> %a1) {
; CHECK-LABEL: test_vminpd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vminpd %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.min.pd.512(<8 x double> %a0, <8 x double> %a1,
                    <8 x double>zeroinitializer, i8 -1, i32 4)
  ret <8 x double> %res
}
declare <8 x double> @llvm.x86.avx512.mask.min.pd.512(<8 x double>, <8 x double>,
                    <8 x double>, i8, i32)

define i8 @test_vptestmq(<8 x i64> %a0, <8 x i64> %a1, i8 %m) {
; CHECK-LABEL: test_vptestmq:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vptestmq %zmm1, %zmm0, %k0
; CHECK-NEXT:    kmovw %k0, %ecx
; CHECK-NEXT:    vptestmq %zmm1, %zmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    addb %cl, %al
; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %res = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 -1)
  %res1 = call i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64> %a0, <8 x i64> %a1, i8 %m)
  %res2 = add i8 %res1, %res
  ret i8 %res2
}
declare i8 @llvm.x86.avx512.ptestm.q.512(<8 x i64>, <8 x i64>, i8)

define i16 @test_vptestmd(<16 x i32> %a0, <16 x i32> %a1, i16 %m) {
; CHECK-LABEL: test_vptestmd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vptestmd %zmm1, %zmm0, %k0
; CHECK-NEXT:    kmovw %k0, %ecx
; CHECK-NEXT:    vptestmd %zmm1, %zmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %res = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 -1)
  %res1 = call i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32> %a0, <16 x i32> %a1, i16 %m)
  %res2 = add i16 %res1, %res
  ret i16 %res2
}
declare i16 @llvm.x86.avx512.ptestm.d.512(<16 x i32>, <16 x i32>, i16)

define void @test_mask_store_ss(i8* %ptr, <4 x float> %data, i8 %mask) {
; CHECK-LABEL: test_mask_store_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmovss %xmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
 call void @llvm.x86.avx512.mask.store.ss(i8* %ptr, <4 x float> %data, i8 %mask)
 ret void
}

declare void @llvm.x86.avx512.mask.store.ss(i8*, <4 x float>, i8 )

declare <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
declare <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
declare <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)

define <16 x float> @test_vsubps_rn(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vsubps_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {rn-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 0)
  ret <16 x float> %res
}

define <16 x float> @test_vsubps_rd(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vsubps_rd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {rd-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 1)
  ret <16 x float> %res
}

define <16 x float> @test_vsubps_ru(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vsubps_ru:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {ru-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_vsubps_rz(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vsubps_rz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {rz-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 3)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_rn(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vmulps_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmulps {rn-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 0)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_rd(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vmulps_rd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmulps {rd-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 1)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_ru(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vmulps_ru:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmulps {ru-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_rz(<16 x float> %a0, <16 x float> %a1) {
; CHECK-LABEL: test_vmulps_rz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmulps {rz-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 -1, i32 3)
  ret <16 x float> %res
}

;; mask float
define <16 x float> @test_vmulps_mask_rn(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {rn-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 %mask, i32 0)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_mask_rd(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_rd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {rd-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 %mask, i32 1)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_mask_ru(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_ru:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {ru-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 %mask, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_mask_rz(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_rz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {rz-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> zeroinitializer, i16 %mask, i32 3)
  ret <16 x float> %res
}

;; With Passthru value
define <16 x float> @test_vmulps_mask_passthru_rn(<16 x float> %a0, <16 x float> %a1, <16 x float> %passthru, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_passthru_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> %passthru, i16 %mask, i32 0)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_mask_passthru_rd(<16 x float> %a0, <16 x float> %a1, <16 x float> %passthru, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_passthru_rd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {rd-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> %passthru, i16 %mask, i32 1)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_mask_passthru_ru(<16 x float> %a0, <16 x float> %a1, <16 x float> %passthru, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_passthru_ru:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {ru-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> %passthru, i16 %mask, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_vmulps_mask_passthru_rz(<16 x float> %a0, <16 x float> %a1, <16 x float> %passthru, i16 %mask) {
; CHECK-LABEL: test_vmulps_mask_passthru_rz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulps {rz-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.mul.ps.512(<16 x float> %a0, <16 x float> %a1,
                    <16 x float> %passthru, i16 %mask, i32 3)
  ret <16 x float> %res
}

;; mask double
define <8 x double> @test_vmulpd_mask_rn(<8 x double> %a0, <8 x double> %a1, i8 %mask) {
; CHECK-LABEL: test_vmulpd_mask_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulpd {rn-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a0, <8 x double> %a1,
                    <8 x double> zeroinitializer, i8 %mask, i32 0)
  ret <8 x double> %res
}

define <8 x double> @test_vmulpd_mask_rd(<8 x double> %a0, <8 x double> %a1, i8 %mask) {
; CHECK-LABEL: test_vmulpd_mask_rd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulpd {rd-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a0, <8 x double> %a1,
                    <8 x double> zeroinitializer, i8 %mask, i32 1)
  ret <8 x double> %res
}

define <8 x double> @test_vmulpd_mask_ru(<8 x double> %a0, <8 x double> %a1, i8 %mask) {
; CHECK-LABEL: test_vmulpd_mask_ru:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulpd {ru-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a0, <8 x double> %a1,
                    <8 x double> zeroinitializer, i8 %mask, i32 2)
  ret <8 x double> %res
}

define <8 x double> @test_vmulpd_mask_rz(<8 x double> %a0, <8 x double> %a1, i8 %mask) {
; CHECK-LABEL: test_vmulpd_mask_rz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmulpd {rz-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.mul.pd.512(<8 x double> %a0, <8 x double> %a1,
                    <8 x double> zeroinitializer, i8 %mask, i32 3)
  ret <8 x double> %res
}

define <8 x i64> @test_mul_epi32_rr(<16 x i32> %a, <16 x i32> %b) {
; CHECK-LABEL: test_mul_epi32_rr:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpmuldq %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rrk(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask) {
; CHECK-LABEL: test_mul_epi32_rrk:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmuldq %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %mul = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> %passThru
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rrkz(<16 x i32> %a, <16 x i32> %b, i8 %mask) {
; CHECK-LABEL: test_mul_epi32_rrkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmuldq %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %mul = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> zeroinitializer
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rm(<16 x i32> %a, <16 x i32>* %ptr_b) {
; CHECK-LABEL: test_mul_epi32_rm:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpmuldq (%rdi), %zmm0, %zmm0
; CHECK-NEXT:    retq
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %res = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rmk(<16 x i32> %a, <16 x i32>* %ptr_b, <8 x i64> %passThru, i8 %mask) {
; CHECK-LABEL: test_mul_epi32_rmk:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuldq (%rdi), %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %mul = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> %passThru
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rmkz(<16 x i32> %a, <16 x i32>* %ptr_b, i8 %mask) {
; CHECK-LABEL: test_mul_epi32_rmkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuldq (%rdi), %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %mul = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> zeroinitializer
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rmb(<16 x i32> %a, i64* %ptr_b) {
; CHECK-LABEL: test_mul_epi32_rmb:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpmuldq (%rdi){1to8}, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %q = load i64, i64* %ptr_b
  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
  %b = bitcast <8 x i64> %b64 to <16 x i32>
  %res = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rmbk(<16 x i32> %a, i64* %ptr_b, <8 x i64> %passThru, i8 %mask) {
; CHECK-LABEL: test_mul_epi32_rmbk:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuldq (%rdi){1to8}, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %q = load i64, i64* %ptr_b
  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
  %b = bitcast <8 x i64> %b64 to <16 x i32>
  %mul = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> %passThru
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epi32_rmbkz(<16 x i32> %a, i64* %ptr_b, i8 %mask) {
; CHECK-LABEL: test_mul_epi32_rmbkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuldq (%rdi){1to8}, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %q = load i64, i64* %ptr_b
  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
  %b = bitcast <8 x i64> %b64 to <16 x i32>
  %mul = call <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> zeroinitializer
  ret < 8 x i64> %res
}

declare <8 x i64> @llvm.x86.avx512.pmul.dq.512(<16 x i32>, <16 x i32>)

define <8 x i64> @test_mul_epu32_rr(<16 x i32> %a, <16 x i32> %b) {
; CHECK-LABEL: test_mul_epu32_rr:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpmuludq %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rrk(<16 x i32> %a, <16 x i32> %b, <8 x i64> %passThru, i8 %mask) {
; CHECK-LABEL: test_mul_epu32_rrk:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmuludq %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %mul = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> %passThru
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rrkz(<16 x i32> %a, <16 x i32> %b, i8 %mask) {
; CHECK-LABEL: test_mul_epu32_rrkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmuludq %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %mul = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> zeroinitializer
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rm(<16 x i32> %a, <16 x i32>* %ptr_b) {
; CHECK-LABEL: test_mul_epu32_rm:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpmuludq (%rdi), %zmm0, %zmm0
; CHECK-NEXT:    retq
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %res = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rmk(<16 x i32> %a, <16 x i32>* %ptr_b, <8 x i64> %passThru, i8 %mask) {
; CHECK-LABEL: test_mul_epu32_rmk:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuludq (%rdi), %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %mul = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> %passThru
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rmkz(<16 x i32> %a, <16 x i32>* %ptr_b, i8 %mask) {
; CHECK-LABEL: test_mul_epu32_rmkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuludq (%rdi), %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %b = load <16 x i32>, <16 x i32>* %ptr_b
  %mul = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> zeroinitializer
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rmb(<16 x i32> %a, i64* %ptr_b) {
; CHECK-LABEL: test_mul_epu32_rmb:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpmuludq (%rdi){1to8}, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %q = load i64, i64* %ptr_b
  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
  %b = bitcast <8 x i64> %b64 to <16 x i32>
  %res = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rmbk(<16 x i32> %a, i64* %ptr_b, <8 x i64> %passThru, i8 %mask) {
; CHECK-LABEL: test_mul_epu32_rmbk:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuludq (%rdi){1to8}, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %q = load i64, i64* %ptr_b
  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
  %b = bitcast <8 x i64> %b64 to <16 x i32>
  %mul = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> %passThru
  ret < 8 x i64> %res
}

define <8 x i64> @test_mul_epu32_rmbkz(<16 x i32> %a, i64* %ptr_b, i8 %mask) {
; CHECK-LABEL: test_mul_epu32_rmbkz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmuludq (%rdi){1to8}, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %q = load i64, i64* %ptr_b
  %vecinit.i = insertelement <8 x i64> undef, i64 %q, i32 0
  %b64 = shufflevector <8 x i64> %vecinit.i, <8 x i64> undef, <8 x i32> zeroinitializer
  %b = bitcast <8 x i64> %b64 to <16 x i32>
  %mul = call <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32> %a, <16 x i32> %b)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res = select <8 x i1> %mask.cast, <8 x i64> %mul, <8 x i64> zeroinitializer
  ret < 8 x i64> %res
}

declare <8 x i64> @llvm.x86.avx512.pmulu.dq.512(<16 x i32>, <16 x i32>)

define <16 x float> @test_mm512_maskz_add_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_add_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {rn-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_maskz_add_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_add_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {rd-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_maskz_add_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_add_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {ru-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_maskz_add_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_add_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {rz-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 3)
  ret <16 x float> %res
}


define <16 x float> @test_mm512_maskz_add_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_add_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_add_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_add_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_mask_add_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_add_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {rd-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_mask_add_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_add_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {ru-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_add_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_add_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps {rz-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 3)
  ret <16 x float> %res
}


define <16 x float> @test_mm512_mask_add_round_ps_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_add_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddps %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 4)
  ret <16 x float> %res
}


define <16 x float> @test_mm512_add_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_add_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vaddps {rn-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_add_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_add_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vaddps {rd-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_add_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_add_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vaddps {ru-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_add_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_add_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vaddps {rz-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 3)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_add_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_add_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vaddps %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.add.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)

define <16 x float> @test_mm512_mask_sub_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_sub_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vsubps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_mask_sub_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_sub_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vsubps {rd-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_mask_sub_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_sub_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vsubps {ru-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_sub_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_sub_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vsubps {rz-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 3)
  ret <16 x float> %res
}


define <16 x float> @test_mm512_mask_sub_round_ps_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_sub_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vsubps %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_sub_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_sub_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {rn-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_sub_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_sub_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {rd-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_sub_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_sub_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {ru-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_sub_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_sub_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps {rz-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 3)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_sub_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_sub_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vsubps %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.sub.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_maskz_div_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_div_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {rn-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_maskz_div_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_div_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {rd-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_maskz_div_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_div_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {ru-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_maskz_div_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_div_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {rz-sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 3)
  ret <16 x float> %res
}


define <16 x float> @test_mm512_maskz_div_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_div_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_div_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_div_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {rn-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_mask_div_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_div_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {rd-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_mask_div_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_div_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {ru-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_div_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_div_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps {rz-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 3)
  ret <16 x float> %res
}


define <16 x float> @test_mm512_mask_div_round_ps_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_div_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vdivps %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 4)
  ret <16 x float> %res
}


define <16 x float> @test_mm512_div_round_ps_rn_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_div_round_ps_rn_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vdivps {rn-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 0)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_div_round_ps_rd_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_div_round_ps_rd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vdivps {rd-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 1)
  ret <16 x float> %res
}
define <16 x float> @test_mm512_div_round_ps_ru_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_div_round_ps_ru_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vdivps {ru-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 2)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_div_round_ps_rz_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_div_round_ps_rz_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vdivps {rz-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 3)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_div_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_div_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vdivps %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.div.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)

define <16 x float> @test_mm512_maskz_min_round_ps_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_min_round_ps_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vminps {sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.min.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_maskz_min_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_min_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vminps %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.min.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_min_round_ps_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_min_round_ps_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vminps {sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.min.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_min_round_ps_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_min_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vminps %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.min.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_min_round_ps_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_min_round_ps_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vminps {sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.min.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_min_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_min_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vminps %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.min.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.min.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)

define <16 x float> @test_mm512_maskz_max_round_ps_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_max_round_ps_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxps {sae}, %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.max.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_maskz_max_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_maskz_max_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxps %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.max.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_max_round_ps_sae(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_max_round_ps_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxps {sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.max.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_mask_max_round_ps_current(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask) {
; CHECK-LABEL: test_mm512_mask_max_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxps %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.max.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float> %src, i16 %mask, i32 4)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_max_round_ps_sae(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_max_round_ps_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmaxps {sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.max.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 8)
  ret <16 x float> %res
}

define <16 x float> @test_mm512_max_round_ps_current(<16 x float> %a0, <16 x float> %a1, i16 %mask) {
; CHECK-LABEL: test_mm512_max_round_ps_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmaxps %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.max.ps.512(<16 x float> %a0, <16 x float> %a1, <16 x float>zeroinitializer, i16 -1, i32 4)
  ret <16 x float> %res
}
declare <16 x float> @llvm.x86.avx512.mask.max.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)

declare <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>, <4 x float>, <4 x float>, i8, i32) nounwind readnone

define <4 x float> @test_mask_add_ss_rn(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_ss_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddss {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 0)
  ret <4 x float> %res
}

define <4 x float> @test_mask_add_ss_rd(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_ss_rd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddss {rd-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 1)
  ret <4 x float> %res
}

define <4 x float> @test_mask_add_ss_ru(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_ss_ru:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddss {ru-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 2)
  ret <4 x float> %res
}

define <4 x float> @test_mask_add_ss_rz(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_ss_rz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddss {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 3)
  ret <4 x float> %res
}

define <4 x float> @test_mask_add_ss_current(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_ss_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddss %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 4)
  ret <4 x float> %res
}

define <4 x float> @test_maskz_add_ss_rn(<4 x float> %a0, <4 x float> %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_add_ss_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddss {rn-sae}, %xmm1, %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 %mask, i32 0)
  ret <4 x float> %res
}

define <4 x float> @test_add_ss_rn(<4 x float> %a0, <4 x float> %a1) {
; CHECK-LABEL: test_add_ss_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vaddss {rn-sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 -1, i32 0)
  ret <4 x float> %res
}

define <4 x float> @test_mask_add_ss_current_memfold(<4 x float> %a0, float* %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_ss_current_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vaddss (%rdi), %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %a1.val = load float, float* %a1
  %a1v0 = insertelement <4 x float> undef, float %a1.val, i32 0
  %a1v1 = insertelement <4 x float> %a1v0, float 0.000000e+00, i32 1
  %a1v2 = insertelement <4 x float> %a1v1, float 0.000000e+00, i32 2
  %a1v  = insertelement <4 x float> %a1v2, float 0.000000e+00, i32 3
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1v, <4 x float> %a2, i8 %mask, i32 4)
  ret <4 x float> %res
}

define <4 x float> @test_maskz_add_ss_current_memfold(<4 x float> %a0, float* %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_add_ss_current_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vaddss (%rdi), %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %a1.val = load float, float* %a1
  %a1v0 = insertelement <4 x float> undef, float %a1.val, i32 0
  %a1v1 = insertelement <4 x float> %a1v0, float 0.000000e+00, i32 1
  %a1v2 = insertelement <4 x float> %a1v1, float 0.000000e+00, i32 2
  %a1v  = insertelement <4 x float> %a1v2, float 0.000000e+00, i32 3
  %res = call <4 x float> @llvm.x86.avx512.mask.add.ss.round(<4 x float>%a0, <4 x float> %a1v, <4 x float> zeroinitializer, i8 %mask, i32 4)
  ret <4 x float> %res
}

declare <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>, <2 x double>, <2 x double>, i8, i32) nounwind readnone

define <2 x double> @test_mask_add_sd_rn(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_sd_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddsd {rn-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 0)
  ret <2 x double> %res
}

define <2 x double> @test_mask_add_sd_rd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_sd_rd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddsd {rd-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 1)
  ret <2 x double> %res
}

define <2 x double> @test_mask_add_sd_ru(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_sd_ru:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddsd {ru-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 2)
  ret <2 x double> %res
}

define <2 x double> @test_mask_add_sd_rz(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_sd_rz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddsd {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 3)
  ret <2 x double> %res
}

define <2 x double> @test_mask_add_sd_current(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_sd_current:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddsd %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 4)
  ret <2 x double> %res
}

define <2 x double> @test_maskz_add_sd_rn(<2 x double> %a0, <2 x double> %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_add_sd_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vaddsd {rn-sae}, %xmm1, %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 %mask, i32 0)
  ret <2 x double> %res
}

define <2 x double> @test_add_sd_rn(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_add_sd_rn:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vaddsd {rn-sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 -1, i32 0)
  ret <2 x double> %res
}

define <2 x double> @test_mask_add_sd_current_memfold(<2 x double> %a0, double* %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_add_sd_current_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vaddsd (%rdi), %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %a1.val = load double, double* %a1
  %a1v0 = insertelement <2 x double> undef, double %a1.val, i32 0
  %a1v = insertelement <2 x double> %a1v0, double 0.000000e+00, i32 1
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1v, <2 x double> %a2, i8 %mask, i32 4)
  ret <2 x double> %res
}

define <2 x double> @test_maskz_add_sd_current_memfold(<2 x double> %a0, double* %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_add_sd_current_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vaddsd (%rdi), %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %a1.val = load double, double* %a1
  %a1v0 = insertelement <2 x double> undef, double %a1.val, i32 0
  %a1v = insertelement <2 x double> %a1v0, double 0.000000e+00, i32 1
  %res = call <2 x double> @llvm.x86.avx512.mask.add.sd.round(<2 x double>%a0, <2 x double> %a1v, <2 x double> zeroinitializer, i8 %mask, i32 4)
  ret <2 x double> %res
}

declare <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>, <4 x float>, <4 x float>, i8, i32) nounwind readnone

define <4 x float> @test_mask_max_ss_sae(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_max_ss_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxss {sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 8)
  ret <4 x float> %res
}

define <4 x float> @test_maskz_max_ss_sae(<4 x float> %a0, <4 x float> %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_max_ss_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxss {sae}, %xmm1, %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 %mask, i32 8)
  ret <4 x float> %res
}

define <4 x float> @test_max_ss_sae(<4 x float> %a0, <4 x float> %a1) {
; CHECK-LABEL: test_max_ss_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmaxss {sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 -1, i32 8)
  ret <4 x float> %res
}

define <4 x float> @test_mask_max_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_max_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxss %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 4)
  ret <4 x float> %res
}

define <4 x float> @test_maskz_max_ss(<4 x float> %a0, <4 x float> %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_max_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxss %xmm1, %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 %mask, i32 4)
  ret <4 x float> %res
}

define <4 x float> @test_max_ss(<4 x float> %a0, <4 x float> %a1) {
; CHECK-LABEL: test_max_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmaxss %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 -1, i32 4)
  ret <4 x float> %res
}

define <4 x float> @test_mask_max_ss_memfold(<4 x float> %a0, float* %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_max_ss_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmaxss (%rdi), %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %a1.val = load float, float* %a1
  %a1v0 = insertelement <4 x float> undef, float %a1.val, i32 0
  %a1v1 = insertelement <4 x float> %a1v0, float 0.000000e+00, i32 1
  %a1v2 = insertelement <4 x float> %a1v1, float 0.000000e+00, i32 2
  %a1v  = insertelement <4 x float> %a1v2, float 0.000000e+00, i32 3
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1v, <4 x float> %a2, i8 %mask, i32 4)
  ret <4 x float> %res
}

define <4 x float> @test_maskz_max_ss_memfold(<4 x float> %a0, float* %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_max_ss_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmaxss (%rdi), %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %a1.val = load float, float* %a1
  %a1v0 = insertelement <4 x float> undef, float %a1.val, i32 0
  %a1v1 = insertelement <4 x float> %a1v0, float 0.000000e+00, i32 1
  %a1v2 = insertelement <4 x float> %a1v1, float 0.000000e+00, i32 2
  %a1v  = insertelement <4 x float> %a1v2, float 0.000000e+00, i32 3
  %res = call <4 x float> @llvm.x86.avx512.mask.max.ss.round(<4 x float>%a0, <4 x float> %a1v, <4 x float> zeroinitializer, i8 %mask, i32 4)
  ret <4 x float> %res
}
declare <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>, <2 x double>, <2 x double>, i8, i32) nounwind readnone

define <2 x double> @test_mask_max_sd_sae(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_max_sd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxsd {sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 8)
  ret <2 x double> %res
}

define <2 x double> @test_maskz_max_sd_sae(<2 x double> %a0, <2 x double> %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_max_sd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxsd {sae}, %xmm1, %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 %mask, i32 8)
  ret <2 x double> %res
}

define <2 x double> @test_max_sd_sae(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_max_sd_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmaxsd {sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 -1, i32 8)
  ret <2 x double> %res
}

define <2 x double> @test_mask_max_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_max_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxsd %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 4)
  ret <2 x double> %res
}

define <2 x double> @test_maskz_max_sd(<2 x double> %a0, <2 x double> %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_max_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmaxsd %xmm1, %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 %mask, i32 4)
  ret <2 x double> %res
}

define <2 x double> @test_max_sd(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_max_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmaxsd %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 -1, i32 4)
  ret <2 x double> %res
}

define <2 x double> @test_mask_max_sd_memfold(<2 x double> %a0, double* %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_mask_max_sd_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmaxsd (%rdi), %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %a1.val = load double, double* %a1
  %a1v0 = insertelement <2 x double> undef, double %a1.val, i32 0
  %a1v = insertelement <2 x double> %a1v0, double 0.000000e+00, i32 1
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1v, <2 x double> %a2, i8 %mask, i32 4)
  ret <2 x double> %res
}

define <2 x double> @test_maskz_max_sd_memfold(<2 x double> %a0, double* %a1, i8 %mask) {
; CHECK-LABEL: test_maskz_max_sd_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmaxsd (%rdi), %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %a1.val = load double, double* %a1
  %a1v0 = insertelement <2 x double> undef, double %a1.val, i32 0
  %a1v = insertelement <2 x double> %a1v0, double 0.000000e+00, i32 1
  %res = call <2 x double> @llvm.x86.avx512.mask.max.sd.round(<2 x double>%a0, <2 x double> %a1v, <2 x double> zeroinitializer, i8 %mask, i32 4)
  ret <2 x double> %res
}

define <2 x double> @test_x86_avx512_cvtsi2sd64(<2 x double> %a, i64 %b) {
; CHECK-LABEL: test_x86_avx512_cvtsi2sd64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsi2sdq %rdi, {rz-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.cvtsi2sd64(<2 x double> %a, i64 %b, i32 3) ; <<<2 x double>> [#uses=1]
  ret <2 x double> %res
}
declare <2 x double> @llvm.x86.avx512.cvtsi2sd64(<2 x double>, i64, i32) nounwind readnone

define <4 x float> @test_x86_avx512_cvtsi2ss32(<4 x float> %a, i32 %b) {
; CHECK-LABEL: test_x86_avx512_cvtsi2ss32:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsi2ssl %edi, {rz-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.cvtsi2ss32(<4 x float> %a, i32 %b, i32 3) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.avx512.cvtsi2ss32(<4 x float>, i32, i32) nounwind readnone

define <4 x float> @test_x86_avx512_cvtsi2ss64(<4 x float> %a, i64 %b) {
; CHECK-LABEL: test_x86_avx512_cvtsi2ss64:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtsi2ssq %rdi, {rz-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.cvtsi2ss64(<4 x float> %a, i64 %b, i32 3) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.avx512.cvtsi2ss64(<4 x float>, i64, i32) nounwind readnone

define <4 x float> @test_x86_avx512__mm_cvt_roundu32_ss (<4 x float> %a, i32 %b)
; CHECK-LABEL: test_x86_avx512__mm_cvt_roundu32_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2ssl %edi, {rd-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %res = call <4 x float> @llvm.x86.avx512.cvtusi2ss(<4 x float> %a, i32 %b, i32 1) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}

define <4 x float> @test_x86_avx512__mm_cvt_roundu32_ss_mem(<4 x float> %a, i32* %ptr)
; CHECK-LABEL: test_x86_avx512__mm_cvt_roundu32_ss_mem:
; CHECK:       ## BB#0:
; CHECK-NEXT:    movl (%rdi), %eax
; CHECK-NEXT:    vcvtusi2ssl %eax, {rd-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %b = load i32, i32* %ptr
  %res = call <4 x float> @llvm.x86.avx512.cvtusi2ss(<4 x float> %a, i32 %b, i32 1) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}

define <4 x float> @test_x86_avx512__mm_cvtu32_ss(<4 x float> %a, i32 %b)
; CHECK-LABEL: test_x86_avx512__mm_cvtu32_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2ssl %edi, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %res = call <4 x float> @llvm.x86.avx512.cvtusi2ss(<4 x float> %a, i32 %b, i32 4) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}

define <4 x float> @test_x86_avx512__mm_cvtu32_ss_mem(<4 x float> %a, i32* %ptr)
; CHECK-LABEL: test_x86_avx512__mm_cvtu32_ss_mem:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2ssl (%rdi), %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %b = load i32, i32* %ptr
  %res = call <4 x float> @llvm.x86.avx512.cvtusi2ss(<4 x float> %a, i32 %b, i32 4) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.avx512.cvtusi2ss(<4 x float>, i32, i32) nounwind readnone

define <4 x float> @_mm_cvt_roundu64_ss (<4 x float> %a, i64 %b)
; CHECK-LABEL: _mm_cvt_roundu64_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2ssq %rdi, {rd-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %res = call <4 x float> @llvm.x86.avx512.cvtusi642ss(<4 x float> %a, i64 %b, i32 1) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}

define <4 x float> @_mm_cvtu64_ss(<4 x float> %a, i64 %b)
; CHECK-LABEL: _mm_cvtu64_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2ssq %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %res = call <4 x float> @llvm.x86.avx512.cvtusi642ss(<4 x float> %a, i64 %b, i32 4) ; <<<4 x float>> [#uses=1]
  ret <4 x float> %res
}
declare <4 x float> @llvm.x86.avx512.cvtusi642ss(<4 x float>, i64, i32) nounwind readnone

define <2 x double> @test_x86_avx512_mm_cvtu32_sd(<2 x double> %a, i32 %b)
; CHECK-LABEL: test_x86_avx512_mm_cvtu32_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2sdl %edi, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %res = call <2 x double> @llvm.x86.avx512.cvtusi2sd(<2 x double> %a, i32 %b) ; <<<2 x double>> [#uses=1]
  ret <2 x double> %res
}
declare <2 x double> @llvm.x86.avx512.cvtusi2sd(<2 x double>, i32) nounwind readnone

define <2 x double> @test_x86_avx512_mm_cvtu64_sd(<2 x double> %a, i64 %b)
; CHECK-LABEL: test_x86_avx512_mm_cvtu64_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2sdq %rdi, {rd-sae}, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %res = call <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double> %a, i64 %b, i32 1) ; <<<2 x double>> [#uses=1]
  ret <2 x double> %res
}

define <2 x double> @test_x86_avx512__mm_cvt_roundu64_sd(<2 x double> %a, i64 %b)
; CHECK-LABEL: test_x86_avx512__mm_cvt_roundu64_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcvtusi2sdq %rdi, %xmm0, %xmm0
; CHECK-NEXT:    retq
{
  %res = call <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double> %a, i64 %b, i32 4) ; <<<2 x double>> [#uses=1]
  ret <2 x double> %res
}
declare <2 x double> @llvm.x86.avx512.cvtusi642sd(<2 x double>, i64, i32) nounwind readnone

declare <16 x i32> @llvm.x86.avx512.mask.vpermi2var.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)

define <16 x i32>@test_int_x86_avx512_mask_vpermi2var_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, <16 x i32> %x4, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm3
; CHECK-NEXT:    vpermi2d (%rdi), %zmm0, %zmm3 {%k1}
; CHECK-NEXT:    vpermi2d %zmm2, %zmm0, %zmm1
; CHECK-NEXT:    vpaddd %zmm1, %zmm3, %zmm0
; CHECK-NEXT:    retq
  %x2 = load <16 x i32>, <16 x i32>* %x2p
  %res = call <16 x i32> @llvm.x86.avx512.mask.vpermi2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.vpermi2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x4, i16 -1)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <8 x double> @llvm.x86.avx512.mask.vpermi2var.pd.512(<8 x double>, <8 x i64>, <8 x double>, i8)

define <8 x double>@test_int_x86_avx512_mask_vpermi2var_pd_512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %zmm1, %zmm3
; CHECK-NEXT:    vpermi2pd %zmm2, %zmm0, %zmm3
; CHECK-NEXT:    vpermi2pd %zmm2, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vaddpd %zmm3, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.vpermi2var.pd.512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 %x3)
  %res1 = call <8 x double> @llvm.x86.avx512.mask.vpermi2var.pd.512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 -1)
  %res2 = fadd <8 x double> %res, %res1
  ret <8 x double> %res2
}

declare <16 x float> @llvm.x86.avx512.mask.vpermi2var.ps.512(<16 x float>, <16 x i32>, <16 x float>, i16)

define <16 x float>@test_int_x86_avx512_mask_vpermi2var_ps_512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %zmm1, %zmm3
; CHECK-NEXT:    vpermi2ps %zmm2, %zmm0, %zmm3
; CHECK-NEXT:    vpermi2ps %zmm2, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vaddps %zmm3, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.vpermi2var.ps.512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 %x3)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.vpermi2var.ps.512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 -1)
  %res2 = fadd <16 x float> %res, %res1
  ret <16 x float> %res2
}

declare <8 x i64> @llvm.x86.avx512.mask.vpermi2var.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)

define <8 x i64>@test_int_x86_avx512_mask_vpermi2var_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermi2var_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm3
; CHECK-NEXT:    vpermi2q %zmm2, %zmm0, %zmm3
; CHECK-NEXT:    vpermi2q %zmm2, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vpaddq %zmm3, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.mask.vpermi2var.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
  %res1 = call <8 x i64> @llvm.x86.avx512.mask.vpermi2var.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
  %res2 = add <8 x i64> %res, %res1
  ret <8 x i64> %res2
}

declare <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)

define <16 x i32>@test_int_x86_avx512_maskz_vpermt2var_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32>* %x2p, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm2
; CHECK-NEXT:    vpermt2d (%rdi), %zmm0, %zmm2 {%k1} {z}
; CHECK-NEXT:    vpermt2d %zmm1, %zmm0, %zmm1
; CHECK-NEXT:    vpaddd %zmm1, %zmm2, %zmm0
; CHECK-NEXT:    retq
  %x2 = load <16 x i32>, <16 x i32>* %x2p
  %res = call <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
  %res1 = call <16 x i32> @llvm.x86.avx512.maskz.vpermt2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x1, i16 -1)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <8 x double> @llvm.x86.avx512.maskz.vpermt2var.pd.512(<8 x i64>, <8 x double>, <8 x double>, i8)

define <8 x double>@test_int_x86_avx512_maskz_vpermt2var_pd_512(<8 x i64> %x0, <8 x double> %x1, double* %x2ptr, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vmovapd %zmm1, %zmm2
; CHECK-NEXT:    vpermt2pd (%rdi){1to8}, %zmm0, %zmm2 {%k1} {z}
; CHECK-NEXT:    vpermt2pd %zmm1, %zmm0, %zmm1
; CHECK-NEXT:    vaddpd %zmm1, %zmm2, %zmm0
; CHECK-NEXT:    retq
  %x2s = load double, double* %x2ptr
  %x2ins = insertelement <8 x double> undef, double %x2s, i32 0
  %x2 = shufflevector <8 x double> %x2ins, <8 x double> undef, <8 x i32> zeroinitializer
  %res = call <8 x double> @llvm.x86.avx512.maskz.vpermt2var.pd.512(<8 x i64> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3)
  %res1 = call <8 x double> @llvm.x86.avx512.maskz.vpermt2var.pd.512(<8 x i64> %x0, <8 x double> %x1, <8 x double> %x1, i8 -1)
  %res2 = fadd <8 x double> %res, %res1
  ret <8 x double> %res2
}

declare <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32>, <16 x float>, <16 x float>, i16)

define <16 x float>@test_int_x86_avx512_maskz_vpermt2var_ps_512(<16 x i32> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %zmm1, %zmm3
; CHECK-NEXT:    vpermt2ps %zmm2, %zmm0, %zmm3
; CHECK-NEXT:    vpermt2ps %zmm2, %zmm0, %zmm1 {%k1} {z}
; CHECK-NEXT:    vaddps %zmm3, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3)
  %res1 = call <16 x float> @llvm.x86.avx512.maskz.vpermt2var.ps.512(<16 x i32> %x0, <16 x float> %x1, <16 x float> %x2, i16 -1)
  %res2 = fadd <16 x float> %res, %res1
  ret <16 x float> %res2
}


declare <8 x i64> @llvm.x86.avx512.maskz.vpermt2var.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)

define <8 x i64>@test_int_x86_avx512_maskz_vpermt2var_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vpermt2var_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm3
; CHECK-NEXT:    vpermt2q %zmm2, %zmm0, %zmm3
; CHECK-NEXT:    vpermt2q %zmm2, %zmm0, %zmm1 {%k1} {z}
; CHECK-NEXT:    vpaddq %zmm3, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.maskz.vpermt2var.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
  %res1 = call <8 x i64> @llvm.x86.avx512.maskz.vpermt2var.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
  %res2 = add <8 x i64> %res, %res1
  ret <8 x i64> %res2
}

declare <16 x i32> @llvm.x86.avx512.mask.vpermt2var.d.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)

define <16 x i32>@test_int_x86_avx512_mask_vpermt2var_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_vpermt2var_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm3
; CHECK-NEXT:    vpermt2d %zmm2, %zmm0, %zmm3
; CHECK-NEXT:    vpermt2d %zmm2, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vpaddd %zmm3, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.vpermt2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.vpermt2var.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <8 x double> @llvm.x86.avx512.mask.scalef.pd.512(<8 x double>, <8 x double>, <8 x double>, i8, i32)
define <8 x double>@test_int_x86_avx512_mask_scalef_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_scalef_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vscalefpd {rz-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vscalefpd {rn-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    vaddpd %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.scalef.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 %x3, i32 3)
  %res1 = call <8 x double> @llvm.x86.avx512.mask.scalef.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x double> %x2, i8 -1, i32 0)
  %res2 = fadd <8 x double> %res, %res1
  ret <8 x double> %res2
}

declare <16 x float> @llvm.x86.avx512.mask.scalef.ps.512(<16 x float>, <16 x float>, <16 x float>, i16, i32)
define <16 x float>@test_int_x86_avx512_mask_scalef_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_scalef_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vscalefps {ru-sae}, %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vscalefps {rn-sae}, %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    vaddps %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.scalef.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 %x3, i32 2)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.scalef.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x float> %x2, i16 -1, i32 0)
  %res2 = fadd <16 x float> %res, %res1
  ret <16 x float> %res2
}

declare <16 x i8> @llvm.x86.avx512.mask.pmov.qb.512(<8 x i64>, <16 x i8>, i8)

define <16 x i8>@test_int_x86_avx512_mask_pmov_qb_512(<8 x i64> %x0, <16 x i8> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_qb_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovqb %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovqb %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovqb %zmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i8> @llvm.x86.avx512.mask.pmov.qb.512(<8 x i64> %x0, <16 x i8> %x1, i8 -1)
    %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmov.qb.512(<8 x i64> %x0, <16 x i8> %x1, i8 %x2)
    %res2 = call <16 x i8> @llvm.x86.avx512.mask.pmov.qb.512(<8 x i64> %x0, <16 x i8> zeroinitializer, i8 %x2)
    %res3 = add <16 x i8> %res0, %res1
    %res4 = add <16 x i8> %res3, %res2
    ret <16 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmov.qb.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmov_qb_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_qb_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovqb %zmm0, (%rdi)
; CHECK-NEXT:    vpmovqb %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmov.qb.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmov.qb.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <16 x i8> @llvm.x86.avx512.mask.pmovs.qb.512(<8 x i64>, <16 x i8>, i8)

define <16 x i8>@test_int_x86_avx512_mask_pmovs_qb_512(<8 x i64> %x0, <16 x i8> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_qb_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovsqb %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovsqb %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovsqb %zmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i8> @llvm.x86.avx512.mask.pmovs.qb.512(<8 x i64> %x0, <16 x i8> %x1, i8 -1)
    %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmovs.qb.512(<8 x i64> %x0, <16 x i8> %x1, i8 %x2)
    %res2 = call <16 x i8> @llvm.x86.avx512.mask.pmovs.qb.512(<8 x i64> %x0, <16 x i8> zeroinitializer, i8 %x2)
    %res3 = add <16 x i8> %res0, %res1
    %res4 = add <16 x i8> %res3, %res2
    ret <16 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmovs.qb.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmovs_qb_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_qb_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovsqb %zmm0, (%rdi)
; CHECK-NEXT:    vpmovsqb %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovs.qb.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmovs.qb.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <16 x i8> @llvm.x86.avx512.mask.pmovus.qb.512(<8 x i64>, <16 x i8>, i8)

define <16 x i8>@test_int_x86_avx512_mask_pmovus_qb_512(<8 x i64> %x0, <16 x i8> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_qb_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovusqb %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovusqb %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovusqb %zmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i8> @llvm.x86.avx512.mask.pmovus.qb.512(<8 x i64> %x0, <16 x i8> %x1, i8 -1)
    %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmovus.qb.512(<8 x i64> %x0, <16 x i8> %x1, i8 %x2)
    %res2 = call <16 x i8> @llvm.x86.avx512.mask.pmovus.qb.512(<8 x i64> %x0, <16 x i8> zeroinitializer, i8 %x2)
    %res3 = add <16 x i8> %res0, %res1
    %res4 = add <16 x i8> %res3, %res2
    ret <16 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmovus.qb.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmovus_qb_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_qb_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovusqb %zmm0, (%rdi)
; CHECK-NEXT:    vpmovusqb %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovus.qb.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmovus.qb.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <8 x i16> @llvm.x86.avx512.mask.pmov.qw.512(<8 x i64>, <8 x i16>, i8)

define <8 x i16>@test_int_x86_avx512_mask_pmov_qw_512(<8 x i64> %x0, <8 x i16> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_qw_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovqw %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovqw %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovqw %zmm0, %xmm0
; CHECK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddw %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <8 x i16> @llvm.x86.avx512.mask.pmov.qw.512(<8 x i64> %x0, <8 x i16> %x1, i8 -1)
    %res1 = call <8 x i16> @llvm.x86.avx512.mask.pmov.qw.512(<8 x i64> %x0, <8 x i16> %x1, i8 %x2)
    %res2 = call <8 x i16> @llvm.x86.avx512.mask.pmov.qw.512(<8 x i64> %x0, <8 x i16> zeroinitializer, i8 %x2)
    %res3 = add <8 x i16> %res0, %res1
    %res4 = add <8 x i16> %res3, %res2
    ret <8 x i16> %res4
}

declare void @llvm.x86.avx512.mask.pmov.qw.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmov_qw_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_qw_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovqw %zmm0, (%rdi)
; CHECK-NEXT:    vpmovqw %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmov.qw.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmov.qw.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <8 x i16> @llvm.x86.avx512.mask.pmovs.qw.512(<8 x i64>, <8 x i16>, i8)

define <8 x i16>@test_int_x86_avx512_mask_pmovs_qw_512(<8 x i64> %x0, <8 x i16> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_qw_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovsqw %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovsqw %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovsqw %zmm0, %xmm0
; CHECK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddw %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <8 x i16> @llvm.x86.avx512.mask.pmovs.qw.512(<8 x i64> %x0, <8 x i16> %x1, i8 -1)
    %res1 = call <8 x i16> @llvm.x86.avx512.mask.pmovs.qw.512(<8 x i64> %x0, <8 x i16> %x1, i8 %x2)
    %res2 = call <8 x i16> @llvm.x86.avx512.mask.pmovs.qw.512(<8 x i64> %x0, <8 x i16> zeroinitializer, i8 %x2)
    %res3 = add <8 x i16> %res0, %res1
    %res4 = add <8 x i16> %res3, %res2
    ret <8 x i16> %res4
}

declare void @llvm.x86.avx512.mask.pmovs.qw.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmovs_qw_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_qw_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovsqw %zmm0, (%rdi)
; CHECK-NEXT:    vpmovsqw %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovs.qw.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmovs.qw.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <8 x i16> @llvm.x86.avx512.mask.pmovus.qw.512(<8 x i64>, <8 x i16>, i8)

define <8 x i16>@test_int_x86_avx512_mask_pmovus_qw_512(<8 x i64> %x0, <8 x i16> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_qw_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovusqw %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovusqw %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovusqw %zmm0, %xmm0
; CHECK-NEXT:    vpaddw %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddw %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <8 x i16> @llvm.x86.avx512.mask.pmovus.qw.512(<8 x i64> %x0, <8 x i16> %x1, i8 -1)
    %res1 = call <8 x i16> @llvm.x86.avx512.mask.pmovus.qw.512(<8 x i64> %x0, <8 x i16> %x1, i8 %x2)
    %res2 = call <8 x i16> @llvm.x86.avx512.mask.pmovus.qw.512(<8 x i64> %x0, <8 x i16> zeroinitializer, i8 %x2)
    %res3 = add <8 x i16> %res0, %res1
    %res4 = add <8 x i16> %res3, %res2
    ret <8 x i16> %res4
}

declare void @llvm.x86.avx512.mask.pmovus.qw.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmovus_qw_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_qw_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovusqw %zmm0, (%rdi)
; CHECK-NEXT:    vpmovusqw %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovus.qw.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmovus.qw.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <8 x i32> @llvm.x86.avx512.mask.pmov.qd.512(<8 x i64>, <8 x i32>, i8)

define <8 x i32>@test_int_x86_avx512_mask_pmov_qd_512(<8 x i64> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_qd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovqd %zmm0, %ymm2 {%k1} {z}
; CHECK-NEXT:    vpmovqd %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vpmovqd %zmm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; CHECK-NEXT:    retq
    %res0 = call <8 x i32> @llvm.x86.avx512.mask.pmov.qd.512(<8 x i64> %x0, <8 x i32> %x1, i8 -1)
    %res1 = call <8 x i32> @llvm.x86.avx512.mask.pmov.qd.512(<8 x i64> %x0, <8 x i32> %x1, i8 %x2)
    %res2 = call <8 x i32> @llvm.x86.avx512.mask.pmov.qd.512(<8 x i64> %x0, <8 x i32> zeroinitializer, i8 %x2)
    %res3 = add <8 x i32> %res0, %res1
    %res4 = add <8 x i32> %res3, %res2
    ret <8 x i32> %res4
}

declare void @llvm.x86.avx512.mask.pmov.qd.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmov_qd_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_qd_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovqd %zmm0, (%rdi)
; CHECK-NEXT:    vpmovqd %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmov.qd.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmov.qd.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <8 x i32> @llvm.x86.avx512.mask.pmovs.qd.512(<8 x i64>, <8 x i32>, i8)

define <8 x i32>@test_int_x86_avx512_mask_pmovs_qd_512(<8 x i64> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_qd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovsqd %zmm0, %ymm2 {%k1} {z}
; CHECK-NEXT:    vpmovsqd %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vpmovsqd %zmm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; CHECK-NEXT:    retq
    %res0 = call <8 x i32> @llvm.x86.avx512.mask.pmovs.qd.512(<8 x i64> %x0, <8 x i32> %x1, i8 -1)
    %res1 = call <8 x i32> @llvm.x86.avx512.mask.pmovs.qd.512(<8 x i64> %x0, <8 x i32> %x1, i8 %x2)
    %res2 = call <8 x i32> @llvm.x86.avx512.mask.pmovs.qd.512(<8 x i64> %x0, <8 x i32> zeroinitializer, i8 %x2)
    %res3 = add <8 x i32> %res0, %res1
    %res4 = add <8 x i32> %res3, %res2
    ret <8 x i32> %res4
}

declare void @llvm.x86.avx512.mask.pmovs.qd.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmovs_qd_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_qd_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovsqd %zmm0, (%rdi)
; CHECK-NEXT:    vpmovsqd %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovs.qd.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmovs.qd.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <8 x i32> @llvm.x86.avx512.mask.pmovus.qd.512(<8 x i64>, <8 x i32>, i8)

define <8 x i32>@test_int_x86_avx512_mask_pmovus_qd_512(<8 x i64> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_qd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovusqd %zmm0, %ymm2 {%k1} {z}
; CHECK-NEXT:    vpmovusqd %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vpmovusqd %zmm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm1, %ymm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm2, %ymm0, %ymm0
; CHECK-NEXT:    retq
    %res0 = call <8 x i32> @llvm.x86.avx512.mask.pmovus.qd.512(<8 x i64> %x0, <8 x i32> %x1, i8 -1)
    %res1 = call <8 x i32> @llvm.x86.avx512.mask.pmovus.qd.512(<8 x i64> %x0, <8 x i32> %x1, i8 %x2)
    %res2 = call <8 x i32> @llvm.x86.avx512.mask.pmovus.qd.512(<8 x i64> %x0, <8 x i32> zeroinitializer, i8 %x2)
    %res3 = add <8 x i32> %res0, %res1
    %res4 = add <8 x i32> %res3, %res2
    ret <8 x i32> %res4
}

declare void @llvm.x86.avx512.mask.pmovus.qd.mem.512(i8* %ptr, <8 x i64>, i8)

define void @test_int_x86_avx512_mask_pmovus_qd_mem_512(i8* %ptr, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_qd_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovusqd %zmm0, (%rdi)
; CHECK-NEXT:    vpmovusqd %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovus.qd.mem.512(i8* %ptr, <8 x i64> %x1, i8 -1)
    call void @llvm.x86.avx512.mask.pmovus.qd.mem.512(i8* %ptr, <8 x i64> %x1, i8 %x2)
    ret void
}

declare <16 x i8> @llvm.x86.avx512.mask.pmov.db.512(<16 x i32>, <16 x i8>, i16)

define <16 x i8>@test_int_x86_avx512_mask_pmov_db_512(<16 x i32> %x0, <16 x i8> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_db_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovdb %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovdb %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovdb %zmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i8> @llvm.x86.avx512.mask.pmov.db.512(<16 x i32> %x0, <16 x i8> %x1, i16 -1)
    %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmov.db.512(<16 x i32> %x0, <16 x i8> %x1, i16 %x2)
    %res2 = call <16 x i8> @llvm.x86.avx512.mask.pmov.db.512(<16 x i32> %x0, <16 x i8> zeroinitializer, i16 %x2)
    %res3 = add <16 x i8> %res0, %res1
    %res4 = add <16 x i8> %res3, %res2
    ret <16 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmov.db.mem.512(i8* %ptr, <16 x i32>, i16)

define void @test_int_x86_avx512_mask_pmov_db_mem_512(i8* %ptr, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_db_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovdb %zmm0, (%rdi)
; CHECK-NEXT:    vpmovdb %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmov.db.mem.512(i8* %ptr, <16 x i32> %x1, i16 -1)
    call void @llvm.x86.avx512.mask.pmov.db.mem.512(i8* %ptr, <16 x i32> %x1, i16 %x2)
    ret void
}

declare <16 x i8> @llvm.x86.avx512.mask.pmovs.db.512(<16 x i32>, <16 x i8>, i16)

define <16 x i8>@test_int_x86_avx512_mask_pmovs_db_512(<16 x i32> %x0, <16 x i8> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_db_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovsdb %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovsdb %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovsdb %zmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i8> @llvm.x86.avx512.mask.pmovs.db.512(<16 x i32> %x0, <16 x i8> %x1, i16 -1)
    %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmovs.db.512(<16 x i32> %x0, <16 x i8> %x1, i16 %x2)
    %res2 = call <16 x i8> @llvm.x86.avx512.mask.pmovs.db.512(<16 x i32> %x0, <16 x i8> zeroinitializer, i16 %x2)
    %res3 = add <16 x i8> %res0, %res1
    %res4 = add <16 x i8> %res3, %res2
    ret <16 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmovs.db.mem.512(i8* %ptr, <16 x i32>, i16)

define void @test_int_x86_avx512_mask_pmovs_db_mem_512(i8* %ptr, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_db_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovsdb %zmm0, (%rdi)
; CHECK-NEXT:    vpmovsdb %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovs.db.mem.512(i8* %ptr, <16 x i32> %x1, i16 -1)
    call void @llvm.x86.avx512.mask.pmovs.db.mem.512(i8* %ptr, <16 x i32> %x1, i16 %x2)
    ret void
}

declare <16 x i8> @llvm.x86.avx512.mask.pmovus.db.512(<16 x i32>, <16 x i8>, i16)

define <16 x i8>@test_int_x86_avx512_mask_pmovus_db_512(<16 x i32> %x0, <16 x i8> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_db_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovusdb %zmm0, %xmm2 {%k1} {z}
; CHECK-NEXT:    vpmovusdb %zmm0, %xmm1 {%k1}
; CHECK-NEXT:    vpmovusdb %zmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vpaddb %xmm2, %xmm0, %xmm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i8> @llvm.x86.avx512.mask.pmovus.db.512(<16 x i32> %x0, <16 x i8> %x1, i16 -1)
    %res1 = call <16 x i8> @llvm.x86.avx512.mask.pmovus.db.512(<16 x i32> %x0, <16 x i8> %x1, i16 %x2)
    %res2 = call <16 x i8> @llvm.x86.avx512.mask.pmovus.db.512(<16 x i32> %x0, <16 x i8> zeroinitializer, i16 %x2)
    %res3 = add <16 x i8> %res0, %res1
    %res4 = add <16 x i8> %res3, %res2
    ret <16 x i8> %res4
}

declare void @llvm.x86.avx512.mask.pmovus.db.mem.512(i8* %ptr, <16 x i32>, i16)

define void @test_int_x86_avx512_mask_pmovus_db_mem_512(i8* %ptr, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_db_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovusdb %zmm0, (%rdi)
; CHECK-NEXT:    vpmovusdb %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovus.db.mem.512(i8* %ptr, <16 x i32> %x1, i16 -1)
    call void @llvm.x86.avx512.mask.pmovus.db.mem.512(i8* %ptr, <16 x i32> %x1, i16 %x2)
    ret void
}

declare <16 x i16> @llvm.x86.avx512.mask.pmov.dw.512(<16 x i32>, <16 x i16>, i16)

define <16 x i16>@test_int_x86_avx512_mask_pmov_dw_512(<16 x i32> %x0, <16 x i16> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_dw_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovdw %zmm0, %ymm2 {%k1} {z}
; CHECK-NEXT:    vpmovdw %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vpmovdw %zmm0, %ymm0
; CHECK-NEXT:    vpaddw %ymm1, %ymm0, %ymm0
; CHECK-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i16> @llvm.x86.avx512.mask.pmov.dw.512(<16 x i32> %x0, <16 x i16> %x1, i16 -1)
    %res1 = call <16 x i16> @llvm.x86.avx512.mask.pmov.dw.512(<16 x i32> %x0, <16 x i16> %x1, i16 %x2)
    %res2 = call <16 x i16> @llvm.x86.avx512.mask.pmov.dw.512(<16 x i32> %x0, <16 x i16> zeroinitializer, i16 %x2)
    %res3 = add <16 x i16> %res0, %res1
    %res4 = add <16 x i16> %res3, %res2
    ret <16 x i16> %res4
}

declare void @llvm.x86.avx512.mask.pmov.dw.mem.512(i8* %ptr, <16 x i32>, i16)

define void @test_int_x86_avx512_mask_pmov_dw_mem_512(i8* %ptr, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmov_dw_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovdw %zmm0, (%rdi)
; CHECK-NEXT:    vpmovdw %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmov.dw.mem.512(i8* %ptr, <16 x i32> %x1, i16 -1)
    call void @llvm.x86.avx512.mask.pmov.dw.mem.512(i8* %ptr, <16 x i32> %x1, i16 %x2)
    ret void
}

declare <16 x i16> @llvm.x86.avx512.mask.pmovs.dw.512(<16 x i32>, <16 x i16>, i16)

define <16 x i16>@test_int_x86_avx512_mask_pmovs_dw_512(<16 x i32> %x0, <16 x i16> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_dw_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovsdw %zmm0, %ymm2 {%k1} {z}
; CHECK-NEXT:    vpmovsdw %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vpmovsdw %zmm0, %ymm0
; CHECK-NEXT:    vpaddw %ymm1, %ymm0, %ymm0
; CHECK-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i16> @llvm.x86.avx512.mask.pmovs.dw.512(<16 x i32> %x0, <16 x i16> %x1, i16 -1)
    %res1 = call <16 x i16> @llvm.x86.avx512.mask.pmovs.dw.512(<16 x i32> %x0, <16 x i16> %x1, i16 %x2)
    %res2 = call <16 x i16> @llvm.x86.avx512.mask.pmovs.dw.512(<16 x i32> %x0, <16 x i16> zeroinitializer, i16 %x2)
    %res3 = add <16 x i16> %res0, %res1
    %res4 = add <16 x i16> %res3, %res2
    ret <16 x i16> %res4
}

declare void @llvm.x86.avx512.mask.pmovs.dw.mem.512(i8* %ptr, <16 x i32>, i16)

define void @test_int_x86_avx512_mask_pmovs_dw_mem_512(i8* %ptr, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovs_dw_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovsdw %zmm0, (%rdi)
; CHECK-NEXT:    vpmovsdw %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovs.dw.mem.512(i8* %ptr, <16 x i32> %x1, i16 -1)
    call void @llvm.x86.avx512.mask.pmovs.dw.mem.512(i8* %ptr, <16 x i32> %x1, i16 %x2)
    ret void
}

declare <16 x i16> @llvm.x86.avx512.mask.pmovus.dw.512(<16 x i32>, <16 x i16>, i16)

define <16 x i16>@test_int_x86_avx512_mask_pmovus_dw_512(<16 x i32> %x0, <16 x i16> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_dw_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpmovusdw %zmm0, %ymm2 {%k1} {z}
; CHECK-NEXT:    vpmovusdw %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vpmovusdw %zmm0, %ymm0
; CHECK-NEXT:    vpaddw %ymm1, %ymm0, %ymm0
; CHECK-NEXT:    vpaddw %ymm2, %ymm0, %ymm0
; CHECK-NEXT:    retq
    %res0 = call <16 x i16> @llvm.x86.avx512.mask.pmovus.dw.512(<16 x i32> %x0, <16 x i16> %x1, i16 -1)
    %res1 = call <16 x i16> @llvm.x86.avx512.mask.pmovus.dw.512(<16 x i32> %x0, <16 x i16> %x1, i16 %x2)
    %res2 = call <16 x i16> @llvm.x86.avx512.mask.pmovus.dw.512(<16 x i32> %x0, <16 x i16> zeroinitializer, i16 %x2)
    %res3 = add <16 x i16> %res0, %res1
    %res4 = add <16 x i16> %res3, %res2
    ret <16 x i16> %res4
}

declare void @llvm.x86.avx512.mask.pmovus.dw.mem.512(i8* %ptr, <16 x i32>, i16)

define void @test_int_x86_avx512_mask_pmovus_dw_mem_512(i8* %ptr, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_pmovus_dw_mem_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vpmovusdw %zmm0, (%rdi)
; CHECK-NEXT:    vpmovusdw %zmm0, (%rdi) {%k1}
; CHECK-NEXT:    retq
    call void @llvm.x86.avx512.mask.pmovus.dw.mem.512(i8* %ptr, <16 x i32> %x1, i16 -1)
    call void @llvm.x86.avx512.mask.pmovus.dw.mem.512(i8* %ptr, <16 x i32> %x1, i16 %x2)
    ret void
}

declare <16 x float> @llvm.x86.avx512.mask.cvtdq2ps.512(<16 x i32>, <16 x float>, i16, i32)

define <16 x float>@test_int_x86_avx512_mask_cvt_dq2ps_512(<16 x i32> %x0, <16 x float> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_dq2ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtdq2ps %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vcvtdq2ps {rn-sae}, %zmm0, %zmm0
; CHECK-NEXT:    vaddps %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.cvtdq2ps.512(<16 x i32> %x0, <16 x float> %x1, i16 %x2, i32 4)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.cvtdq2ps.512(<16 x i32> %x0, <16 x float> %x1, i16 -1, i32 0)
  %res2 = fadd <16 x float> %res, %res1
  ret <16 x float> %res2
}

declare <8 x i32> @llvm.x86.avx512.mask.cvtpd2dq.512(<8 x double>, <8 x i32>, i8, i32)

define <8 x i32>@test_int_x86_avx512_mask_cvt_pd2dq_512(<8 x double> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_pd2dq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtpd2dq %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vcvtpd2dq {rn-sae}, %zmm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
; CHECK-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.mask.cvtpd2dq.512(<8 x double> %x0, <8 x i32> %x1, i8 %x2, i32 4)
  %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvtpd2dq.512(<8 x double> %x0, <8 x i32> %x1, i8 -1, i32 0)
  %res2 = add <8 x i32> %res, %res1
  ret <8 x i32> %res2
}

declare <8 x float> @llvm.x86.avx512.mask.cvtpd2ps.512(<8 x double>, <8 x float>, i8, i32)

define <8 x float>@test_int_x86_avx512_mask_cvt_pd2ps_512(<8 x double> %x0, <8 x float> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_pd2ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtpd2ps %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vcvtpd2ps {ru-sae}, %zmm0, %ymm0
; CHECK-NEXT:    vaddps %ymm0, %ymm1, %ymm0
; CHECK-NEXT:    retq
  %res = call <8 x float> @llvm.x86.avx512.mask.cvtpd2ps.512(<8 x double> %x0, <8 x float> %x1, i8 %x2, i32 4)
  %res1 = call <8 x float> @llvm.x86.avx512.mask.cvtpd2ps.512(<8 x double> %x0, <8 x float> %x1, i8 -1, i32 2)
  %res2 = fadd <8 x float> %res, %res1
  ret <8 x float> %res2
}

declare <8 x i32> @llvm.x86.avx512.mask.cvtpd2udq.512(<8 x double>, <8 x i32>, i8, i32)

define <8 x i32>@test_int_x86_avx512_mask_cvt_pd2udq_512(<8 x double> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_pd2udq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtpd2udq {ru-sae}, %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vcvtpd2udq {rn-sae}, %zmm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
; CHECK-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.mask.cvtpd2udq.512(<8 x double> %x0, <8 x i32> %x1, i8 %x2, i32 2)
  %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvtpd2udq.512(<8 x double> %x0, <8 x i32> %x1, i8 -1, i32 0)
  %res2 = add <8 x i32> %res, %res1
  ret <8 x i32> %res2
}

declare <16 x i32> @llvm.x86.avx512.mask.cvtps2dq.512(<16 x float>, <16 x i32>, i16, i32)

define <16 x i32>@test_int_x86_avx512_mask_cvt_ps2dq_512(<16 x float> %x0, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_ps2dq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtps2dq {ru-sae}, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vcvtps2dq {rn-sae}, %zmm0, %zmm0
; CHECK-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.cvtps2dq.512(<16 x float> %x0, <16 x i32> %x1, i16 %x2, i32 2)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.cvtps2dq.512(<16 x float> %x0, <16 x i32> %x1, i16 -1, i32 0)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <8 x double> @llvm.x86.avx512.mask.cvtps2pd.512(<8 x float>, <8 x double>, i8, i32)

define <8 x double>@test_int_x86_avx512_mask_cvt_ps2pd_512(<8 x float> %x0, <8 x double> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_ps2pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtps2pd %ymm0, %zmm1 {%k1}
; CHECK-NEXT:    vcvtps2pd {sae}, %ymm0, %zmm0
; CHECK-NEXT:    vaddpd %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.cvtps2pd.512(<8 x float> %x0, <8 x double> %x1, i8 %x2, i32 4)
  %res1 = call <8 x double> @llvm.x86.avx512.mask.cvtps2pd.512(<8 x float> %x0, <8 x double> %x1, i8 -1, i32 8)
  %res2 = fadd <8 x double> %res, %res1
  ret <8 x double> %res2
}

declare <16 x i32> @llvm.x86.avx512.mask.cvtps2udq.512(<16 x float>, <16 x i32>, i16, i32)

define <16 x i32>@test_int_x86_avx512_mask_cvt_ps2udq_512(<16 x float> %x0, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_ps2udq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtps2udq {ru-sae}, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vcvtps2udq {rn-sae}, %zmm0, %zmm0
; CHECK-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.cvtps2udq.512(<16 x float> %x0, <16 x i32> %x1, i16 %x2, i32 2)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.cvtps2udq.512(<16 x float> %x0, <16 x i32> %x1, i16 -1, i32 0)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <8 x i32> @llvm.x86.avx512.mask.cvttpd2dq.512(<8 x double>, <8 x i32>, i8, i32)

define <8 x i32>@test_int_x86_avx512_mask_cvtt_pd2dq_512(<8 x double> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvtt_pd2dq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvttpd2dq %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vcvttpd2dq {sae}, %zmm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
; CHECK-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2dq.512(<8 x double> %x0, <8 x i32> %x1, i8 %x2, i32 4)
  %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2dq.512(<8 x double> %x0, <8 x i32> %x1, i8 -1, i32 8)
  %res2 = add <8 x i32> %res, %res1
  ret <8 x i32> %res2
}

declare <16 x float> @llvm.x86.avx512.mask.cvtudq2ps.512(<16 x i32>, <16 x float>, i16, i32)

define <16 x float>@test_int_x86_avx512_mask_cvt_udq2ps_512(<16 x i32> %x0, <16 x float> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_udq2ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtudq2ps %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vcvtudq2ps {rn-sae}, %zmm0, %zmm0
; CHECK-NEXT:    vaddps %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.cvtudq2ps.512(<16 x i32> %x0, <16 x float> %x1, i16 %x2, i32 4)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.cvtudq2ps.512(<16 x i32> %x0, <16 x float> %x1, i16 -1, i32 0)
  %res2 = fadd <16 x float> %res, %res1
  ret <16 x float> %res2
}

declare <8 x i32> @llvm.x86.avx512.mask.cvttpd2udq.512(<8 x double>, <8 x i32>, i8, i32)

define <8 x i32>@test_int_x86_avx512_mask_cvtt_pd2udq_512(<8 x double> %x0, <8 x i32> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvtt_pd2udq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvttpd2udq %zmm0, %ymm1 {%k1}
; CHECK-NEXT:    vcvttpd2udq {sae}, %zmm0, %ymm0
; CHECK-NEXT:    vpaddd %ymm0, %ymm1, %ymm0
; CHECK-NEXT:    retq
  %res = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2udq.512(<8 x double> %x0, <8 x i32> %x1, i8 %x2, i32 4)
  %res1 = call <8 x i32> @llvm.x86.avx512.mask.cvttpd2udq.512(<8 x double> %x0, <8 x i32> %x1, i8 -1, i32 8)
  %res2 = add <8 x i32> %res, %res1
  ret <8 x i32> %res2
}

declare <16 x i32> @llvm.x86.avx512.mask.cvttps2dq.512(<16 x float>, <16 x i32>, i16, i32)

define <16 x i32>@test_int_x86_avx512_mask_cvtt_ps2dq_512(<16 x float> %x0, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvtt_ps2dq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvttps2dq %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vcvttps2dq {sae}, %zmm0, %zmm0
; CHECK-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.cvttps2dq.512(<16 x float> %x0, <16 x i32> %x1, i16 %x2, i32 4)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.cvttps2dq.512(<16 x float> %x0, <16 x i32> %x1, i16 -1, i32 8)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <16 x i32> @llvm.x86.avx512.mask.cvttps2udq.512(<16 x float>, <16 x i32>, i16, i32)

define <16 x i32>@test_int_x86_avx512_mask_cvtt_ps2udq_512(<16 x float> %x0, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvtt_ps2udq_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvttps2udq %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vcvttps2udq {sae}, %zmm0, %zmm0
; CHECK-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.cvttps2udq.512(<16 x float> %x0, <16 x i32> %x1, i16 %x2, i32 4)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.cvttps2udq.512(<16 x float> %x0, <16 x i32> %x1, i16 -1, i32 8)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32) nounwind readnone

define <4 x float> @test_getexp_ss(<4 x float> %a0, <4 x float> %a1, <4 x float> %a2, i8 %mask) {
; CHECK-LABEL: test_getexp_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm2, %xmm3
; CHECK-NEXT:    vgetexpss %xmm1, %xmm0, %xmm3 {%k1}
; CHECK-NEXT:    vgetexpss {sae}, %xmm1, %xmm0, %xmm4 {%k1} {z}
; CHECK-NEXT:    vgetexpss {sae}, %xmm1, %xmm0, %xmm5
; CHECK-NEXT:    vgetexpss {sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddps %xmm2, %xmm3, %xmm0
; CHECK-NEXT:    vaddps %xmm5, %xmm4, %xmm1
; CHECK-NEXT:    vaddps %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res0 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> %a2, i8 %mask, i32 8)
  %res2 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 %mask, i32 8)
  %res3 = call <4 x float> @llvm.x86.avx512.mask.getexp.ss(<4 x float>%a0, <4 x float> %a1, <4 x float> zeroinitializer, i8 -1, i32 8)

  %res.1 = fadd <4 x float> %res0, %res1
  %res.2 = fadd <4 x float> %res2, %res3
  %res   = fadd <4 x float> %res.1, %res.2
  ret <4 x float> %res
}

declare <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32) nounwind readnone

define <2 x double> @test_getexp_sd(<2 x double> %a0, <2 x double> %a1, <2 x double> %a2, i8 %mask) {
; CHECK-LABEL: test_getexp_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vgetexpsd %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm2, %xmm4
; CHECK-NEXT:    vgetexpsd %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vgetexpsd {sae}, %xmm1, %xmm0, %xmm5 {%k1} {z}
; CHECK-NEXT:    vgetexpsd {sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddpd %xmm2, %xmm4, %xmm0
; CHECK-NEXT:    vaddpd %xmm3, %xmm5, %xmm1
; CHECK-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res0 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> %a2, i8 %mask, i32 8)
  %res2 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 %mask, i32 8)
  %res3 = call <2 x double> @llvm.x86.avx512.mask.getexp.sd(<2 x double>%a0, <2 x double> %a1, <2 x double> zeroinitializer, i8 -1, i32 4)

  %res.1 = fadd <2 x double> %res0, %res1
  %res.2 = fadd <2 x double> %res2, %res3
  %res   = fadd <2 x double> %res.1, %res.2
  ret <2 x double> %res
}

declare i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double>, <2 x double>, i32, i8, i32)

define i8@test_int_x86_avx512_mask_cmp_sd(<2 x double> %x0, <2 x double> %x1, i8 %x3, i32 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_cmp_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcmpnltsd {sae}, %xmm1, %xmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
; CHECK-NEXT:    retq

  %res4 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 5, i8 %x3, i32 8)
  ret i8 %res4
}

define i8@test_int_x86_avx512_mask_cmp_sd_all(<2 x double> %x0, <2 x double> %x1, i8 %x3, i32 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_cmp_sd_all:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmplesd %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %ecx
; CHECK-NEXT:    vcmpunordsd {sae}, %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %edx
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcmpneqsd %xmm1, %xmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %esi
; CHECK-NEXT:    vcmpnltsd {sae}, %xmm1, %xmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    orb %cl, %dl
; CHECK-NEXT:    orb %sil, %al
; CHECK-NEXT:    orb %dl, %al
; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
; CHECK-NEXT:    retq

  %res1 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 2, i8 -1, i32 4)
  %res2 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 3, i8 -1, i32 8)
  %res3 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 4, i8 %x3, i32 4)
  %res4 = call i8 @llvm.x86.avx512.mask.cmp.sd(<2 x double> %x0, <2 x double> %x1, i32 5, i8 %x3, i32 8)

  %res11 = or i8 %res1, %res2
  %res12 = or i8 %res3, %res4
  %res13 = or i8 %res11, %res12
  ret i8 %res13
}

declare i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float>, <4 x float>, i32, i8, i32)

define i8@test_int_x86_avx512_mask_cmp_ss(<4 x float> %x0, <4 x float> %x1, i8 %x3, i32 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_cmp_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcmpunordss %xmm1, %xmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
; CHECK-NEXT:    retq

  %res2 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 3, i8 %x3, i32 4)
  ret i8 %res2
}


define i8@test_int_x86_avx512_mask_cmp_ss_all(<4 x float> %x0, <4 x float> %x1, i8 %x3, i32 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_cmp_ss_all:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpless %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %ecx
; CHECK-NEXT:    vcmpunordss {sae}, %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %edx
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcmpneqss %xmm1, %xmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %esi
; CHECK-NEXT:    vcmpnltss {sae}, %xmm1, %xmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    andb %cl, %dl
; CHECK-NEXT:    andb %sil, %al
; CHECK-NEXT:    andb %dl, %al
; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %res1 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 2, i8 -1, i32 4)
  %res2 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 3, i8 -1, i32 8)
  %res3 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 4, i8 %x3, i32 4)
  %res4 = call i8 @llvm.x86.avx512.mask.cmp.ss(<4 x float> %x0, <4 x float> %x1, i32 5, i8 %x3, i32 8)

  %res11 = and i8 %res1, %res2
  %res12 = and i8 %res3, %res4
  %res13 = and i8 %res11, %res12
  ret i8 %res13
}

declare <16 x float> @llvm.x86.avx512.mask.shuf.f32x4(<16 x float>, <16 x float>, i32, <16 x float>, i16)

define <16 x float>@test_int_x86_avx512_mask_shuf_f32x4(<16 x float> %x0, <16 x float> %x1, <16 x float> %x3, i16 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_shuf_f32x4:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vshuff32x4 {{.*#+}} zmm2 {%k1} = zmm0[8,9,10,11,4,5,6,7],zmm1[4,5,6,7,0,1,2,3]
; CHECK-NEXT:    vshuff32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,4,5,6,7],zmm1[4,5,6,7,0,1,2,3]
; CHECK-NEXT:    vaddps %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.shuf.f32x4(<16 x float> %x0, <16 x float> %x1, i32 22, <16 x float> %x3, i16 %x4)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.shuf.f32x4(<16 x float> %x0, <16 x float> %x1, i32 22, <16 x float> %x3, i16 -1)
  %res2 = fadd <16 x float> %res, %res1
  ret <16 x float> %res2
}

declare <8 x double> @llvm.x86.avx512.mask.shuf.f64x2(<8 x double>, <8 x double>, i32, <8 x double>, i8)

define <8 x double>@test_int_x86_avx512_mask_shuf_f64x2(<8 x double> %x0, <8 x double> %x1, <8 x double> %x3, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_shuf_f64x2:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vshuff64x2 {{.*#+}} zmm3 {%k1} {z} = zmm0[4,5,2,3],zmm1[2,3,0,1]
; CHECK-NEXT:    vshuff64x2 {{.*#+}} zmm2 {%k1} = zmm0[4,5,2,3],zmm1[2,3,0,1]
; CHECK-NEXT:    vshuff64x2 {{.*#+}} zmm0 = zmm0[4,5,2,3],zmm1[2,3,0,1]
; CHECK-NEXT:    vaddpd %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    vaddpd %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.shuf.f64x2(<8 x double> %x0, <8 x double> %x1, i32 22, <8 x double> %x3, i8 %x4)
  %res1 = call <8 x double> @llvm.x86.avx512.mask.shuf.f64x2(<8 x double> %x0, <8 x double> %x1, i32 22, <8 x double> %x3, i8 -1)
  %res2 = call <8 x double> @llvm.x86.avx512.mask.shuf.f64x2(<8 x double> %x0, <8 x double> %x1, i32 22, <8 x double> zeroinitializer, i8 %x4)

  %res3 = fadd <8 x double> %res, %res1
  %res4 = fadd <8 x double> %res3, %res2
  ret <8 x double> %res4
}

declare <16 x i32> @llvm.x86.avx512.mask.shuf.i32x4(<16 x i32>, <16 x i32>, i32, <16 x i32>, i16)

define <16 x i32>@test_int_x86_avx512_mask_shuf_i32x4(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x3, i16 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_shuf_i32x4:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vshufi32x4 {{.*#+}} zmm2 {%k1} = zmm0[8,9,10,11,4,5,6,7],zmm1[4,5,6,7,0,1,2,3]
; CHECK-NEXT:    vshufi32x4 {{.*#+}} zmm0 = zmm0[8,9,10,11,4,5,6,7],zmm1[4,5,6,7,0,1,2,3]
; CHECK-NEXT:    vpaddd %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.shuf.i32x4(<16 x i32> %x0, <16 x i32> %x1, i32 22, <16 x i32> %x3, i16 %x4)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.shuf.i32x4(<16 x i32> %x0, <16 x i32> %x1, i32 22, <16 x i32> %x3, i16 -1)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <8 x i64> @llvm.x86.avx512.mask.shuf.i64x2(<8 x i64>, <8 x i64>, i32, <8 x i64>, i8)

define <8 x i64>@test_int_x86_avx512_mask_shuf_i64x2(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x3, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_shuf_i64x2:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vshufi64x2 {{.*#+}} zmm2 {%k1} = zmm0[4,5,2,3],zmm1[2,3,0,1]
; CHECK-NEXT:    vshufi64x2 {{.*#+}} zmm0 = zmm0[4,5,2,3],zmm1[2,3,0,1]
; CHECK-NEXT:    vpaddq %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.mask.shuf.i64x2(<8 x i64> %x0, <8 x i64> %x1, i32 22, <8 x i64> %x3, i8 %x4)
  %res1 = call <8 x i64> @llvm.x86.avx512.mask.shuf.i64x2(<8 x i64> %x0, <8 x i64> %x1, i32 22, <8 x i64> %x3, i8 -1)
  %res2 = add <8 x i64> %res, %res1
  ret <8 x i64> %res2
}

declare <8 x double> @llvm.x86.avx512.mask.getmant.pd.512(<8 x double>, i32, <8 x double>, i8, i32)

define <8 x double>@test_int_x86_avx512_mask_getmant_pd_512(<8 x double> %x0, <8 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_getmant_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vgetmantpd $11, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vgetmantpd $11, {sae}, %zmm0, %zmm0
; CHECK-NEXT:    vaddpd %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.getmant.pd.512(<8 x double> %x0, i32 11, <8 x double> %x2, i8 %x3, i32 4)
  %res1 = call <8 x double> @llvm.x86.avx512.mask.getmant.pd.512(<8 x double> %x0, i32 11, <8 x double> %x2, i8 -1, i32 8)
  %res2 = fadd <8 x double> %res, %res1
  ret <8 x double> %res2
}

declare <16 x float> @llvm.x86.avx512.mask.getmant.ps.512(<16 x float>, i32, <16 x float>, i16, i32)

define <16 x float>@test_int_x86_avx512_mask_getmant_ps_512(<16 x float> %x0, <16 x float> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_getmant_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vgetmantps $11, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vgetmantps $11, {sae}, %zmm0, %zmm0
; CHECK-NEXT:    vaddps %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.getmant.ps.512(<16 x float> %x0, i32 11, <16 x float> %x2, i16 %x3, i32 4)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.getmant.ps.512(<16 x float> %x0, i32 11, <16 x float> %x2, i16 -1, i32 8)
  %res2 = fadd <16 x float> %res, %res1
  ret <16 x float> %res2
}

declare <2 x double> @llvm.x86.avx512.mask.getmant.sd(<2 x double>, <2 x double>, i32, <2 x double>, i8, i32)

define <2 x double>@test_int_x86_avx512_mask_getmant_sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_getmant_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vgetmantsd $11, %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm2, %xmm4
; CHECK-NEXT:    vgetmantsd $11, %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vgetmantsd $11, %xmm1, %xmm0, %xmm5 {%k1} {z}
; CHECK-NEXT:    vgetmantsd $11, {sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddpd %xmm5, %xmm4, %xmm0
; CHECK-NEXT:    vaddpd %xmm3, %xmm2, %xmm1
; CHECK-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res  = call <2 x double> @llvm.x86.avx512.mask.getmant.sd(<2 x double> %x0, <2 x double> %x1, i32 11, <2 x double> %x2, i8 %x3, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask.getmant.sd(<2 x double> %x0, <2 x double> %x1, i32 11, <2 x double> zeroinitializer, i8 %x3, i32 4)
  %res2 = call <2 x double> @llvm.x86.avx512.mask.getmant.sd(<2 x double> %x0, <2 x double> %x1, i32 11, <2 x double> %x2, i8 %x3, i32 8)
  %res3 = call <2 x double> @llvm.x86.avx512.mask.getmant.sd(<2 x double> %x0, <2 x double> %x1, i32 11, <2 x double> %x2, i8 -1, i32 4)
  %res11 = fadd <2 x double> %res, %res1
  %res12 = fadd <2 x double> %res2, %res3
  %res13 = fadd <2 x double> %res11, %res12
  ret <2 x double> %res13
}

declare <4 x float> @llvm.x86.avx512.mask.getmant.ss(<4 x float>, <4 x float>, i32, <4 x float>, i8, i32)

define <4 x float>@test_int_x86_avx512_mask_getmant_ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_getmant_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vgetmantss $11, %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vgetmantss $11, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vgetmantss $11, %xmm1, %xmm0, %xmm4 {%k1} {z}
; CHECK-NEXT:    vgetmantss $11, {sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vaddps %xmm4, %xmm2, %xmm1
; CHECK-NEXT:    vaddps %xmm3, %xmm0, %xmm0
; CHECK-NEXT:    vaddps %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    retq
  %res  = call <4 x float> @llvm.x86.avx512.mask.getmant.ss(<4 x float> %x0, <4 x float> %x1, i32 11, <4 x float> %x2, i8 %x3, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask.getmant.ss(<4 x float> %x0, <4 x float> %x1, i32 11, <4 x float> zeroinitializer, i8 %x3, i32 4)
  %res2 = call <4 x float> @llvm.x86.avx512.mask.getmant.ss(<4 x float> %x0, <4 x float> %x1, i32 11, <4 x float> %x2, i8 -1, i32 8)
  %res3 = call <4 x float> @llvm.x86.avx512.mask.getmant.ss(<4 x float> %x0, <4 x float> %x1, i32 11, <4 x float> %x2, i8 -1, i32 4)
  %res11 = fadd <4 x float> %res, %res1
  %res12 = fadd <4 x float> %res2, %res3
  %res13 = fadd <4 x float> %res11, %res12
  ret <4 x float> %res13
}

declare <8 x double> @llvm.x86.avx512.vpermilvar.pd.512(<8 x double>, <8 x i64>)

define <8 x double>@test_int_x86_avx512_vpermilvar_pd_512(<8 x double> %x0, <8 x i64> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpermilpd %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.vpermilvar.pd.512(<8 x double> %x0, <8 x i64> %x1)
  ret <8 x double> %res
}

define <8 x double>@test_int_x86_avx512_vpermilvar_pd_512_mask(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 %mask) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_pd_512_mask:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermilpd %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.vpermilvar.pd.512(<8 x double> %x0, <8 x i64> %x1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x double> %res, <8 x double> %x2
  ret <8 x double> %res2
}

define <8 x double>@test_int_x86_avx512_vpermilvar_pd_512_maskz(<8 x double> %x0, <8 x i64> %x1, i8 %mask) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_pd_512_maskz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermilpd %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.vpermilvar.pd.512(<8 x double> %x0, <8 x i64> %x1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x double> %res, <8 x double> zeroinitializer
  ret <8 x double> %res2
}

declare <16 x float> @llvm.x86.avx512.vpermilvar.ps.512(<16 x float>, <16 x i32>)

define <16 x float>@test_int_x86_avx512_vpermilvar_ps_512(<16 x float> %x0, <16 x i32> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpermilps %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.vpermilvar.ps.512(<16 x float> %x0, <16 x i32> %x1)
  ret <16 x float> %res
}

define <16 x float>@test_int_x86_avx512_vpermilvar_ps_512_mask(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 %mask) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_ps_512_mask:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermilps %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.vpermilvar.ps.512(<16 x float> %x0, <16 x i32> %x1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x float> %res, <16 x float> %x2
  ret <16 x float> %res2
}

define <16 x float>@test_int_x86_avx512_vpermilvar_ps_512_maskz(<16 x float> %x0, <16 x i32> %x1, i16 %mask) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_ps_512_maskz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermilps %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.vpermilvar.ps.512(<16 x float> %x0, <16 x i32> %x1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x float> %res, <16 x float> zeroinitializer
  ret <16 x float> %res2
}

; Test case to make sure we can print shuffle decode comments for constant pool loads.
define <16 x float>@test_int_x86_avx512_vpermilvar_ps_512_constant_pool(<16 x float> %x0, <16 x i32> %x1) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_ps_512_constant_pool:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 = zmm0[1,0,3,2,4,5,6,7,10,11,8,9,14,15,13,12]
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.vpermilvar.ps.512(<16 x float> %x0, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 1, i32 0>)
  ret <16 x float> %res
}

define <16 x float>@test_int_x86_avx512_vpermilvar_ps_512_constant_pool_mask(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 %mask) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_ps_512_constant_pool_mask:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm2 {%k1} = zmm0[1,0,3,2,4,5,6,7,10,11,8,9,14,15,13,12]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.vpermilvar.ps.512(<16 x float> %x0, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 1, i32 0>)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x float> %res, <16 x float> %x2
  ret <16 x float> %res2
}

define <16 x float>@test_int_x86_avx512_vpermilvar_ps_512_constant_pool_maskz(<16 x float> %x0, <16 x i32> %x1, i16 %mask) {
; CHECK-LABEL: test_int_x86_avx512_vpermilvar_ps_512_constant_pool_maskz:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[1,0,3,2,4,5,6,7,10,11,8,9,14,15,13,12]
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.vpermilvar.ps.512(<16 x float> %x0, <16 x i32> <i32 1, i32 0, i32 3, i32 2, i32 0, i32 1, i32 2, i32 3, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 1, i32 0>)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x float> %res, <16 x float> zeroinitializer
  ret <16 x float> %res2
}

declare <2 x double> @llvm.x86.avx512.mask.cvtss2sd.round(<2 x double>, <4 x float>, <2 x double>, i8, i32)

define <2 x double>@test_int_x86_avx512_mask_cvt_ss2sd_round(<2 x double> %x0,<4 x float> %x1, <2 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_ss2sd_round:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtss2sd %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vcvtss2sd {sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vaddpd %xmm0, %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.cvtss2sd.round(<2 x double> %x0, <4 x float> %x1, <2 x double> %x2, i8 %x3, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask.cvtss2sd.round(<2 x double> %x0, <4 x float> %x1, <2 x double> %x2, i8 -1, i32 8)
  %res2 = fadd <2 x double> %res, %res1
  ret <2 x double> %res2
}

declare <4 x float> @llvm.x86.avx512.mask.cvtsd2ss.round(<4 x float>, <2 x double>, <4 x float>, i8, i32)

define <4 x float>@test_int_x86_avx512_mask_cvt_sd2ss_round(<4 x float> %x0,<2 x double> %x1, <4 x float> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_cvt_sd2ss_round:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vcvtsd2ss {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vcvtsd2ss {rn-sae}, %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    vaddps %xmm0, %xmm2, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.cvtsd2ss.round(<4 x float> %x0, <2 x double> %x1, <4 x float> %x2, i8 %x3, i32 3)
  %res1 = call <4 x float> @llvm.x86.avx512.mask.cvtsd2ss.round(<4 x float> %x0, <2 x double> %x1, <4 x float> %x2, i8 -1, i32 8)
  %res2 = fadd <4 x float> %res, %res1
  ret <4 x float> %res2
}

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

define <16 x i32>@test_int_x86_avx512_mask_pternlog_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_pternlog_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm3
; CHECK-NEXT:    vpternlogd $33, %zmm2, %zmm1, %zmm3
; CHECK-NEXT:    vpternlogd $33, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT:    vpaddd %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.pternlog.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i32 33, i16 %x4)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.pternlog.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i32 33, i16 -1)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

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

define <16 x i32>@test_int_x86_avx512_maskz_pternlog_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x4) {
; CHECK-LABEL: test_int_x86_avx512_maskz_pternlog_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm3
; CHECK-NEXT:    vpternlogd $33, %zmm2, %zmm1, %zmm3
; CHECK-NEXT:    vpternlogd $33, %zmm2, %zmm1, %zmm0 {%k1} {z}
; CHECK-NEXT:    vpaddd %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.maskz.pternlog.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i32 33, i16 %x4)
  %res1 = call <16 x i32> @llvm.x86.avx512.maskz.pternlog.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i32 33, i16 -1)
  %res2 = add <16 x i32> %res, %res1
  ret <16 x i32> %res2
}

declare <8 x i64> @llvm.x86.avx512.mask.pternlog.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i32, i8)

define <8 x i64>@test_int_x86_avx512_mask_pternlog_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_pternlog_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm3
; CHECK-NEXT:    vpternlogq $33, %zmm2, %zmm1, %zmm3
; CHECK-NEXT:    vpternlogq $33, %zmm2, %zmm1, %zmm0 {%k1}
; CHECK-NEXT:    vpaddq %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.mask.pternlog.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i32 33, i8 %x4)
  %res1 = call <8 x i64> @llvm.x86.avx512.mask.pternlog.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i32 33, i8 -1)
  %res2 = add <8 x i64> %res, %res1
  ret <8 x i64> %res2
}

declare <8 x i64> @llvm.x86.avx512.maskz.pternlog.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i32, i8)

define <8 x i64>@test_int_x86_avx512_maskz_pternlog_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_maskz_pternlog_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovdqa64 %zmm0, %zmm3
; CHECK-NEXT:    vpternlogq $33, %zmm2, %zmm1, %zmm3
; CHECK-NEXT:    vpternlogq $33, %zmm2, %zmm1, %zmm0 {%k1} {z}
; CHECK-NEXT:    vpaddq %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.maskz.pternlog.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i32 33, i8 %x4)
  %res1 = call <8 x i64> @llvm.x86.avx512.maskz.pternlog.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i32 33, i8 -1)
  %res2 = add <8 x i64> %res, %res1
  ret <8 x i64> %res2
}

define i32 @test_x86_avx512_comi_sd_eq_sae(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_comi_sd_eq_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpeqsd {sae}, %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 0, i32 8)
  ret i32 %res
}

define i32 @test_x86_avx512_ucomi_sd_eq_sae(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_ucomi_sd_eq_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpeq_uqsd {sae}, %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 8, i32 8)
  ret i32 %res
}

define i32 @test_x86_avx512_comi_sd_eq(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_comi_sd_eq:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpeqsd %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 0, i32 4)
  ret i32 %res
}

define i32 @test_x86_avx512_ucomi_sd_eq(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_ucomi_sd_eq:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpeq_uqsd %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 8, i32 4)
  ret i32 %res
}

define i32 @test_x86_avx512_comi_sd_lt_sae(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_comi_sd_lt_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpltsd {sae}, %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 1, i32 8)
  ret i32 %res
}

define i32 @test_x86_avx512_ucomi_sd_lt_sae(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_ucomi_sd_lt_sae:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpngesd {sae}, %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 9, i32 8)
  ret i32 %res
}

define i32 @test_x86_avx512_comi_sd_lt(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_comi_sd_lt:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpltsd %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 1, i32 4)
  ret i32 %res
}

define i32 @test_x86_avx512_ucomi_sd_lt(<2 x double> %a0, <2 x double> %a1) {
; CHECK-LABEL: test_x86_avx512_ucomi_sd_lt:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpngesd %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.sd(<2 x double> %a0, <2 x double> %a1, i32 9, i32 4)
  ret i32 %res
}

declare i32 @llvm.x86.avx512.vcomi.sd(<2 x double>, <2 x double>, i32, i32)

define i32 @test_x86_avx512_ucomi_ss_lt(<4 x float> %a0, <4 x float> %a1) {
; CHECK-LABEL: test_x86_avx512_ucomi_ss_lt:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vcmpngess %xmm1, %xmm0, %k0
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    retq
  %res = call i32 @llvm.x86.avx512.vcomi.ss(<4 x float> %a0, <4 x float> %a1, i32 9, i32 4)
  ret i32 %res
}

declare i32 @llvm.x86.avx512.vcomi.ss(<4 x float>, <4 x float>, i32, i32)

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

define <16 x i32>@test_int_x86_avx512_mask_prorv_d_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_prorv_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vprorvd %zmm1, %zmm0, %zmm3
; CHECK-NEXT:    vprorvd %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vprorvd %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    vpaddd %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    vpaddd %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.prorv.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.prorv.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> zeroinitializer, i16 %x3)
  %res2 = call <16 x i32> @llvm.x86.avx512.mask.prorv.d.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
  %res3 = add <16 x i32> %res, %res1
  %res4 = add <16 x i32> %res3, %res2
  ret <16 x i32> %res4
}

declare <8 x i64> @llvm.x86.avx512.mask.prorv.q.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)

define <8 x i64>@test_int_x86_avx512_mask_prorv_q_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_prorv_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vprorvq %zmm1, %zmm0, %zmm3
; CHECK-NEXT:    vprorvq %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vprorvq %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    vpaddq %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    vpaddq %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.mask.prorv.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
  %res1 = call <8 x i64> @llvm.x86.avx512.mask.prorv.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3)
  %res2 = call <8 x i64> @llvm.x86.avx512.mask.prorv.q.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
  %res3 = add <8 x i64> %res, %res1
  %res4 = add <8 x i64> %res3, %res2
  ret <8 x i64> %res4
}

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

define <16 x i32>@test_int_x86_avx512_mask_prol_d_512(<16 x i32> %x0, i32 %x1, <16 x i32> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_prol_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vprold $3, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vprold $3, %zmm0, %zmm2 {%k1} {z}
; CHECK-NEXT:    vpaddd %zmm2, %zmm1, %zmm1
; CHECK-NEXT:    vprold $3, %zmm0, %zmm0
; CHECK-NEXT:    vpaddd %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.prol.d.512(<16 x i32> %x0, i32 3, <16 x i32> %x2, i16 %x3)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.prol.d.512(<16 x i32> %x0, i32 3, <16 x i32> zeroinitializer, i16 %x3)
  %res2 = call <16 x i32> @llvm.x86.avx512.mask.prol.d.512(<16 x i32> %x0, i32 3, <16 x i32> %x2, i16 -1)
  %res3 = add <16 x i32> %res, %res1
  %res4 = add <16 x i32> %res3, %res2
  ret <16 x i32> %res4
}

declare <8 x i64> @llvm.x86.avx512.mask.prol.q.512(<8 x i64>, i32, <8 x i64>, i8)

define <8 x i64>@test_int_x86_avx512_mask_prol_q_512(<8 x i64> %x0, i32 %x1, <8 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_prol_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vprolq $3, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vprolq $3, %zmm0, %zmm2 {%k1} {z}
; CHECK-NEXT:    vpaddq %zmm2, %zmm1, %zmm1
; CHECK-NEXT:    vprolq $3, %zmm0, %zmm0
; CHECK-NEXT:    vpaddq %zmm0, %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.mask.prol.q.512(<8 x i64> %x0, i32 3, <8 x i64> %x2, i8 %x3)
  %res1 = call <8 x i64> @llvm.x86.avx512.mask.prol.q.512(<8 x i64> %x0, i32 3, <8 x i64> zeroinitializer, i8 %x3)
  %res2 = call <8 x i64> @llvm.x86.avx512.mask.prol.q.512(<8 x i64> %x0, i32 3, <8 x i64> %x2, i8 -1)
  %res3 = add <8 x i64> %res, %res1
  %res4 = add <8 x i64> %res3, %res2
  ret <8 x i64> %res4
}

declare <8 x double> @llvm.x86.avx512.mask.permvar.df.512(<8 x double>, <8 x i64>, <8 x double>, i8)

define <8 x double>@test_int_x86_avx512_mask_permvar_df_512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_permvar_df_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm3
; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm2 {%k1}
; CHECK-NEXT:    vpermpd %zmm0, %zmm1, %zmm0 {%k1} {z}
; CHECK-NEXT:    vaddpd %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    vaddpd %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.permvar.df.512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 %x3)
  %res1 = call <8 x double> @llvm.x86.avx512.mask.permvar.df.512(<8 x double> %x0, <8 x i64> %x1, <8 x double> zeroinitializer, i8 %x3)
  %res2 = call <8 x double> @llvm.x86.avx512.mask.permvar.df.512(<8 x double> %x0, <8 x i64> %x1, <8 x double> %x2, i8 -1)
  %res3 = fadd <8 x double> %res, %res1
  %res4 = fadd <8 x double> %res3, %res2
  ret <8 x double> %res4
}

declare <8 x i64> @llvm.x86.avx512.mask.permvar.di.512(<8 x i64>, <8 x i64>, <8 x i64>, i8)

define <8 x i64>@test_int_x86_avx512_mask_permvar_di_512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_permvar_di_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermq %zmm0, %zmm1, %zmm3
; CHECK-NEXT:    vpermq %zmm0, %zmm1, %zmm2 {%k1}
; CHECK-NEXT:    vpermq %zmm0, %zmm1, %zmm0 {%k1} {z}
; CHECK-NEXT:    vpaddq %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    vpaddq %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.mask.permvar.di.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 %x3)
  %res1 = call <8 x i64> @llvm.x86.avx512.mask.permvar.di.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> zeroinitializer, i8 %x3)
  %res2 = call <8 x i64> @llvm.x86.avx512.mask.permvar.di.512(<8 x i64> %x0, <8 x i64> %x1, <8 x i64> %x2, i8 -1)
  %res3 = add <8 x i64> %res, %res1
  %res4 = add <8 x i64> %res3, %res2
  ret <8 x i64> %res4
}

declare <16 x float> @llvm.x86.avx512.mask.permvar.sf.512(<16 x float>, <16 x i32>, <16 x float>, i16)

define <16 x float>@test_int_x86_avx512_mask_permvar_sf_512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_permvar_sf_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermps %zmm0, %zmm1, %zmm3
; CHECK-NEXT:    vpermps %zmm0, %zmm1, %zmm2 {%k1}
; CHECK-NEXT:    vpermps %zmm0, %zmm1, %zmm0 {%k1} {z}
; CHECK-NEXT:    vaddps %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    vaddps %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.permvar.sf.512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 %x3)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.permvar.sf.512(<16 x float> %x0, <16 x i32> %x1, <16 x float> zeroinitializer, i16 %x3)
  %res2 = call <16 x float> @llvm.x86.avx512.mask.permvar.sf.512(<16 x float> %x0, <16 x i32> %x1, <16 x float> %x2, i16 -1)
  %res3 = fadd <16 x float> %res, %res1
  %res4 = fadd <16 x float> %res3, %res2
  ret <16 x float> %res4
}

declare <16 x i32> @llvm.x86.avx512.mask.permvar.si.512(<16 x i32>, <16 x i32>, <16 x i32>, i16)

define <16 x i32>@test_int_x86_avx512_mask_permvar_si_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3) {
; CHECK-LABEL: test_int_x86_avx512_mask_permvar_si_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpermd %zmm0, %zmm1, %zmm3
; CHECK-NEXT:    vpermd %zmm0, %zmm1, %zmm2 {%k1}
; CHECK-NEXT:    vpermd %zmm0, %zmm1, %zmm0 {%k1} {z}
; CHECK-NEXT:    vpaddd %zmm0, %zmm2, %zmm0
; CHECK-NEXT:    vpaddd %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.mask.permvar.si.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 %x3)
  %res1 = call <16 x i32> @llvm.x86.avx512.mask.permvar.si.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> zeroinitializer, i16 %x3)
  %res2 = call <16 x i32> @llvm.x86.avx512.mask.permvar.si.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2, i16 -1)
  %res3 = add <16 x i32> %res, %res1
  %res4 = add <16 x i32> %res3, %res2
  ret <16 x i32> %res4
}

declare <8 x double> @llvm.x86.avx512.mask.fixupimm.pd.512(<8 x double>, <8 x double>, <8 x i64>, i32, i8, i32)

define <8 x double>@test_int_x86_avx512_mask_fixupimm_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x i64> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_fixupimm_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %zmm0, %zmm3
; CHECK-NEXT:    vfixupimmpd $4, %zmm2, %zmm1, %zmm3 {%k1}
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vfixupimmpd $5, %zmm2, %zmm1, %zmm4 {%k1} {z}
; CHECK-NEXT:    vaddpd %zmm4, %zmm3, %zmm3
; CHECK-NEXT:    vfixupimmpd $3, {sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT:    vaddpd %zmm0, %zmm3, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.mask.fixupimm.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x i64> %x2, i32 4, i8 %x4, i32 4)
  %res1 = call <8 x double> @llvm.x86.avx512.mask.fixupimm.pd.512(<8 x double> zeroinitializer, <8 x double> %x1, <8 x i64> %x2, i32 5, i8 %x4, i32 4)
  %res2 = call <8 x double> @llvm.x86.avx512.mask.fixupimm.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x i64> %x2, i32 3, i8 -1, i32 8)
  %res3 = fadd <8 x double> %res, %res1
  %res4 = fadd <8 x double> %res3, %res2
  ret <8 x double> %res4
}

declare <8 x double> @llvm.x86.avx512.maskz.fixupimm.pd.512(<8 x double>, <8 x double>, <8 x i64>, i32, i8, i32)

define <8 x double>@test_int_x86_avx512_maskz_fixupimm_pd_512(<8 x double> %x0, <8 x double> %x1, <8 x i64> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_maskz_fixupimm_pd_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %zmm0, %zmm3
; CHECK-NEXT:    vfixupimmpd $3, %zmm2, %zmm1, %zmm3 {%k1} {z}
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vmovapd %zmm0, %zmm5
; CHECK-NEXT:    vfixupimmpd $5, %zmm4, %zmm1, %zmm5 {%k1} {z}
; CHECK-NEXT:    vaddpd %zmm5, %zmm3, %zmm3
; CHECK-NEXT:    vfixupimmpd $2, {sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT:    vaddpd %zmm0, %zmm3, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x double> @llvm.x86.avx512.maskz.fixupimm.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x i64> %x2, i32 3, i8 %x4, i32 4)
  %res1 = call <8 x double> @llvm.x86.avx512.maskz.fixupimm.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x i64> zeroinitializer, i32 5, i8 %x4, i32 4)
  %res2 = call <8 x double> @llvm.x86.avx512.maskz.fixupimm.pd.512(<8 x double> %x0, <8 x double> %x1, <8 x i64> %x2, i32 2, i8 -1, i32 8)
  %res3 = fadd <8 x double> %res, %res1
  %res4 = fadd <8 x double> %res3, %res2
  ret <8 x double> %res4
}

declare <4 x float> @llvm.x86.avx512.mask.fixupimm.ss(<4 x float>, <4 x float>, <4 x i32>, i32, i8, i32)

define <4 x float>@test_int_x86_avx512_mask_fixupimm_ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_fixupimm_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm0, %xmm3
; CHECK-NEXT:    vfixupimmss $5, %xmm2, %xmm1, %xmm3 {%k1}
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vmovaps %xmm0, %xmm5
; CHECK-NEXT:    vfixupimmss $5, %xmm4, %xmm1, %xmm5 {%k1}
; CHECK-NEXT:    vfixupimmss $5, {sae}, %xmm2, %xmm1, %xmm0
; CHECK-NEXT:    vaddps %xmm5, %xmm3, %xmm1
; CHECK-NEXT:    vaddps %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.fixupimm.ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> %x2, i32 5, i8 %x4, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask.fixupimm.ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> zeroinitializer, i32 5, i8 %x4, i32 4)
  %res2 = call <4 x float> @llvm.x86.avx512.mask.fixupimm.ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> %x2, i32 5, i8 -1, i32 8)
  %res3 = fadd <4 x float> %res, %res1
  %res4 = fadd <4 x float> %res3, %res2
  ret <4 x float> %res4
}

declare <4 x float> @llvm.x86.avx512.maskz.fixupimm.ss(<4 x float>, <4 x float>, <4 x i32>, i32, i8, i32)

define <4 x float>@test_int_x86_avx512_maskz_fixupimm_ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_maskz_fixupimm_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovaps %xmm0, %xmm3
; CHECK-NEXT:    vfixupimmss $5, %xmm2, %xmm1, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm0, %xmm4
; CHECK-NEXT:    vfixupimmss $5, %xmm2, %xmm1, %xmm4 {%k1} {z}
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vfixupimmss $5, {sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
; CHECK-NEXT:    vaddps %xmm0, %xmm4, %xmm0
; CHECK-NEXT:    vaddps %xmm3, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.maskz.fixupimm.ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> %x2, i32 5, i8 %x4, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.maskz.fixupimm.ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> zeroinitializer, i32 5, i8 %x4, i32 8)
  %res2 = call <4 x float> @llvm.x86.avx512.maskz.fixupimm.ss(<4 x float> %x0, <4 x float> %x1, <4 x i32> %x2, i32 5, i8 -1, i32 4)
  %res3 = fadd <4 x float> %res, %res1
  %res4 = fadd <4 x float> %res3, %res2
  ret <4 x float> %res4
}

declare <16 x float> @llvm.x86.avx512.mask.fixupimm.ps.512(<16 x float>, <16 x float>, <16 x i32>, i32, i16, i32)

define <16 x float>@test_int_x86_avx512_mask_fixupimm_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x i32> %x2, i16 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_fixupimm_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %zmm0, %zmm3
; CHECK-NEXT:    vfixupimmps $5, %zmm2, %zmm1, %zmm3 {%k1}
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vmovaps %zmm0, %zmm5
; CHECK-NEXT:    vfixupimmps $5, %zmm4, %zmm1, %zmm5 {%k1}
; CHECK-NEXT:    vaddps %zmm5, %zmm3, %zmm3
; CHECK-NEXT:    vfixupimmps $5, {sae}, %zmm2, %zmm1, %zmm0
; CHECK-NEXT:    vaddps %zmm0, %zmm3, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.mask.fixupimm.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x i32> %x2, i32 5, i16 %x4, i32 4)
  %res1 = call <16 x float> @llvm.x86.avx512.mask.fixupimm.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x i32> zeroinitializer, i32 5, i16 %x4, i32 4)
  %res2 = call <16 x float> @llvm.x86.avx512.mask.fixupimm.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x i32> %x2, i32 5, i16 -1, i32 8)
  %res3 = fadd <16 x float> %res, %res1
  %res4 = fadd <16 x float> %res3, %res2
  ret <16 x float> %res4
}

declare <16 x float> @llvm.x86.avx512.maskz.fixupimm.ps.512(<16 x float>, <16 x float>, <16 x i32>, i32, i16, i32)

define <16 x float>@test_int_x86_avx512_maskz_fixupimm_ps_512(<16 x float> %x0, <16 x float> %x1, <16 x i32> %x2, i16 %x4) {
; CHECK-LABEL: test_int_x86_avx512_maskz_fixupimm_ps_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %zmm0, %zmm3
; CHECK-NEXT:    vfixupimmps $5, %zmm2, %zmm1, %zmm3
; CHECK-NEXT:    vmovaps %zmm0, %zmm4
; CHECK-NEXT:    vfixupimmps $5, %zmm2, %zmm1, %zmm4 {%k1} {z}
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vfixupimmps $5, {sae}, %zmm2, %zmm1, %zmm0 {%k1} {z}
; CHECK-NEXT:    vaddps %zmm0, %zmm4, %zmm0
; CHECK-NEXT:    vaddps %zmm3, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x float> @llvm.x86.avx512.maskz.fixupimm.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x i32> %x2, i32 5, i16 %x4, i32 4)
  %res1 = call <16 x float> @llvm.x86.avx512.maskz.fixupimm.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x i32> zeroinitializer, i32 5, i16 %x4, i32 8)
  %res2 = call <16 x float> @llvm.x86.avx512.maskz.fixupimm.ps.512(<16 x float> %x0, <16 x float> %x1, <16 x i32> %x2, i32 5, i16 -1, i32 4)
  %res3 = fadd <16 x float> %res, %res1
  %res4 = fadd <16 x float> %res3, %res2
  ret <16 x float> %res4
}

declare <2 x double> @llvm.x86.avx512.mask.fixupimm.sd(<2 x double>, <2 x double>, <2 x i64>, i32, i8, i32)

define <2 x double>@test_int_x86_avx512_mask_fixupimm_sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_fixupimm_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovapd %xmm0, %xmm3
; CHECK-NEXT:    vfixupimmsd $5, %xmm2, %xmm1, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm0, %xmm4
; CHECK-NEXT:    vfixupimmsd $5, %xmm2, %xmm1, %xmm4 {%k1}
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vfixupimmsd $5, {sae}, %xmm2, %xmm1, %xmm0 {%k1}
; CHECK-NEXT:    vaddpd %xmm0, %xmm4, %xmm0
; CHECK-NEXT:    vaddpd %xmm3, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.fixupimm.sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> %x2, i32 5, i8 %x4, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask.fixupimm.sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> zeroinitializer, i32 5, i8 %x4, i32 8)
  %res2 = call <2 x double> @llvm.x86.avx512.mask.fixupimm.sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> %x2, i32 5, i8 -1, i32 4)
  %res3 = fadd <2 x double> %res, %res1
  %res4 = fadd <2 x double> %res3, %res2
  ret <2 x double> %res4
}

declare <2 x double> @llvm.x86.avx512.maskz.fixupimm.sd(<2 x double>, <2 x double>, <2 x i64>, i32, i8, i32)

define <2 x double>@test_int_x86_avx512_maskz_fixupimm_sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> %x2, i8 %x4) {
; CHECK-LABEL: test_int_x86_avx512_maskz_fixupimm_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm0, %xmm3
; CHECK-NEXT:    vfixupimmsd $5, %xmm2, %xmm1, %xmm3 {%k1} {z}
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vmovapd %xmm0, %xmm5
; CHECK-NEXT:    vfixupimmsd $5, {sae}, %xmm4, %xmm1, %xmm5 {%k1} {z}
; CHECK-NEXT:    vfixupimmsd $5, {sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
; CHECK-NEXT:    vaddpd %xmm5, %xmm3, %xmm1
; CHECK-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.maskz.fixupimm.sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> %x2, i32 5, i8 %x4, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.maskz.fixupimm.sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> zeroinitializer, i32 5, i8 %x4, i32 8)
  %res2 = call <2 x double> @llvm.x86.avx512.maskz.fixupimm.sd(<2 x double> %x0, <2 x double> %x1, <2 x i64> %x2, i32 5, i8 %x4, i32 8)
  %res3 = fadd <2 x double> %res, %res1
  %res4 = fadd <2 x double> %res3, %res2
  ret <2 x double> %res4
}

declare i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32>, <16 x i32>, i16 %x2)

define i16@test_int_x86_avx512_ptestnm_d_512(<16 x i32> %x0, <16 x i32> %x1, i16 %x2) {
; CHECK-LABEL: test_int_x86_avx512_ptestnm_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vptestnmd %zmm1, %zmm0, %k0
; CHECK-NEXT:    kmovw %k0, %ecx
; CHECK-NEXT:    vptestnmd %zmm1, %zmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    addl %ecx, %eax
; CHECK-NEXT:    ## kill: %AX<def> %AX<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %res = call i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32> %x0, <16 x i32> %x1, i16 %x2)
  %res1 = call i16 @llvm.x86.avx512.ptestnm.d.512(<16 x i32> %x0, <16 x i32> %x1, i16-1)
  %res2 = add i16 %res, %res1
  ret i16 %res2
}

declare i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64>, <8 x i64>, i8 %x2)

define i8@test_int_x86_avx512_ptestnm_q_512(<8 x i64> %x0, <8 x i64> %x1, i8 %x2) {
; CHECK-LABEL: test_int_x86_avx512_ptestnm_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vptestnmq %zmm1, %zmm0, %k0
; CHECK-NEXT:    kmovw %k0, %ecx
; CHECK-NEXT:    vptestnmq %zmm1, %zmm0, %k0 {%k1}
; CHECK-NEXT:    kmovw %k0, %eax
; CHECK-NEXT:    addb %cl, %al
; CHECK-NEXT:    ## kill: %AL<def> %AL<kill> %EAX<kill>
; CHECK-NEXT:    retq
  %res = call i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64> %x0, <8 x i64> %x1, i8 %x2)
  %res1 = call i8 @llvm.x86.avx512.ptestnm.q.512(<8 x i64> %x0, <8 x i64> %x1, i8-1)
  %res2 = add i8 %res, %res1
  ret i8 %res2
}


declare <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)

define <2 x double>@test_int_x86_avx512_mask_vfmadd_sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovapd %xmm0, %xmm3
; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm0, %xmm4
; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm4 {%k1}
; CHECK-NEXT:    vmovapd %xmm0, %xmm5
; CHECK-NEXT:    vfmadd213sd {rz-sae}, %xmm2, %xmm1, %xmm5
; CHECK-NEXT:    vfmadd213sd {rz-sae}, %xmm2, %xmm1, %xmm0 {%k1}
; CHECK-NEXT:    vaddpd %xmm4, %xmm3, %xmm1
; CHECK-NEXT:    vaddpd %xmm0, %xmm5, %xmm0
; CHECK-NEXT:    vaddpd %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 4)
  %res2 = call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 3)
  %res3 = call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 3)
  %res4 = fadd <2 x double> %res, %res1
  %res5 = fadd <2 x double> %res2, %res3
  %res6 = fadd <2 x double> %res4, %res5
  ret <2 x double> %res6
}

declare <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)

define <4 x float>@test_int_x86_avx512_mask_vfmadd_ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovaps %xmm0, %xmm3
; CHECK-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm0, %xmm4
; CHECK-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm4 {%k1}
; CHECK-NEXT:    vmovaps %xmm0, %xmm5
; CHECK-NEXT:    vfmadd213ss {rz-sae}, %xmm2, %xmm1, %xmm5
; CHECK-NEXT:    vfmadd213ss {rz-sae}, %xmm2, %xmm1, %xmm0 {%k1}
; CHECK-NEXT:    vaddps %xmm4, %xmm3, %xmm1
; CHECK-NEXT:    vaddps %xmm0, %xmm5, %xmm0
; CHECK-NEXT:    vaddps %xmm0, %xmm1, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 4)
  %res2 = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 3)
  %res3 = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 3)
  %res4 = fadd <4 x float> %res, %res1
  %res5 = fadd <4 x float> %res2, %res3
  %res6 = fadd <4 x float> %res4, %res5
  ret <4 x float> %res6
}

declare <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)

define <2 x double>@test_int_x86_avx512_maskz_vfmadd_sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm0, %xmm3
; CHECK-NEXT:    vfmadd213sd %xmm2, %xmm1, %xmm3 {%k1} {z}
; CHECK-NEXT:    vfmadd213sd {rz-sae}, %xmm2, %xmm1, %xmm0 {%k1} {z}
; CHECK-NEXT:    vaddpd %xmm0, %xmm3, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 3)
  %res2 = fadd <2 x double> %res, %res1
  ret <2 x double> %res2
}

declare <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)

define <4 x float>@test_int_x86_avx512_maskz_vfmadd_ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vfmadd213ss %xmm2, %xmm1, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 3)
  %res2 = fadd <4 x float> %res, %res1
  ret <4 x float> %res
}
declare <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)

define <2 x double>@test_int_x86_avx512_mask3_vfmadd_sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovapd %xmm2, %xmm3
; CHECK-NEXT:    vfmadd231sd %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm2, %xmm4
; CHECK-NEXT:    vfmadd231sd %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm5
; CHECK-NEXT:    vfmadd231sd {rz-sae}, %xmm1, %xmm0, %xmm5
; CHECK-NEXT:    vfmadd231sd {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddpd %xmm4, %xmm3, %xmm0
; CHECK-NEXT:    vaddpd %xmm2, %xmm5, %xmm1
; CHECK-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 4)
  %res2 = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 3)
  %res3 = call <2 x double> @llvm.x86.avx512.mask3.vfmadd.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 3)
  %res4 = fadd <2 x double> %res, %res1
  %res5 = fadd <2 x double> %res2, %res3
  %res6 = fadd <2 x double> %res4, %res5
  ret <2 x double> %res6
}

declare <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)

define <4 x float>@test_int_x86_avx512_mask3_vfmadd_ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovaps %xmm2, %xmm3
; CHECK-NEXT:    vfmadd231ss %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm2, %xmm4
; CHECK-NEXT:    vfmadd231ss %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm5
; CHECK-NEXT:    vfmadd231ss {rz-sae}, %xmm1, %xmm0, %xmm5
; CHECK-NEXT:    vfmadd231ss {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddps %xmm4, %xmm3, %xmm0
; CHECK-NEXT:    vaddps %xmm2, %xmm5, %xmm1
; CHECK-NEXT:    vaddps %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 4)
  %res2 = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 3)
  %res3 = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 3)
  %res4 = fadd <4 x float> %res, %res1
  %res5 = fadd <4 x float> %res2, %res3
  %res6 = fadd <4 x float> %res4, %res5
  ret <4 x float> %res6
}

define void @fmadd_ss_mask_memfold(float* %a, float* %b, i8 %c) {
; CHECK-LABEL: fmadd_ss_mask_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; CHECK-NEXT:    kmovw %edx, %k1
; CHECK-NEXT:    vfmadd132ss (%rsi), %xmm0, %xmm0 {%k1}
; CHECK-NEXT:    vmovss %xmm0, (%rdi)
; CHECK-NEXT:    retq
  %a.val = load float, float* %a
  %av0 = insertelement <4 x float> undef, float %a.val, i32 0
  %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
  %av2 = insertelement <4 x float> %av1, float 0.000000e+00, i32 2
  %av  = insertelement <4 x float> %av2, float 0.000000e+00, i32 3

  %b.val = load float, float* %b
  %bv0 = insertelement <4 x float> undef, float %b.val, i32 0
  %bv1 = insertelement <4 x float> %bv0, float 0.000000e+00, i32 1
  %bv2 = insertelement <4 x float> %bv1, float 0.000000e+00, i32 2
  %bv  = insertelement <4 x float> %bv2, float 0.000000e+00, i32 3

  %vr = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %av, <4 x float> %bv, <4 x float> %av, i8 %c, i32 4)

  %sr = extractelement <4 x float> %vr, i32 0
  store float %sr, float* %a
  ret void
}

define void @fmadd_ss_maskz_memfold(float* %a, float* %b, i8 %c) {
; CHECK-LABEL: fmadd_ss_maskz_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero
; CHECK-NEXT:    kmovw %edx, %k1
; CHECK-NEXT:    vfmadd132ss (%rsi), %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    vmovss %xmm0, (%rdi)
; CHECK-NEXT:    retq
  %a.val = load float, float* %a
  %av0 = insertelement <4 x float> undef, float %a.val, i32 0
  %av1 = insertelement <4 x float> %av0, float 0.000000e+00, i32 1
  %av2 = insertelement <4 x float> %av1, float 0.000000e+00, i32 2
  %av  = insertelement <4 x float> %av2, float 0.000000e+00, i32 3

  %b.val = load float, float* %b
  %bv0 = insertelement <4 x float> undef, float %b.val, i32 0
  %bv1 = insertelement <4 x float> %bv0, float 0.000000e+00, i32 1
  %bv2 = insertelement <4 x float> %bv1, float 0.000000e+00, i32 2
  %bv  = insertelement <4 x float> %bv2, float 0.000000e+00, i32 3

  %vr = call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %av, <4 x float> %bv, <4 x float> %av, i8 %c, i32 4)

  %sr = extractelement <4 x float> %vr, i32 0
  store float %sr, float* %a
  ret void
}

define void @fmadd_sd_mask_memfold(double* %a, double* %b, i8 %c) {
; CHECK-LABEL: fmadd_sd_mask_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; CHECK-NEXT:    kmovw %edx, %k1
; CHECK-NEXT:    vfmadd132sd (%rsi), %xmm0, %xmm0 {%k1}
; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
; CHECK-NEXT:    retq
  %a.val = load double, double* %a
  %av0 = insertelement <2 x double> undef, double %a.val, i32 0
  %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1

  %b.val = load double, double* %b
  %bv0 = insertelement <2 x double> undef, double %b.val, i32 0
  %bv = insertelement <2 x double> %bv0, double 0.000000e+00, i32 1

  %vr = call <2 x double> @llvm.x86.avx512.mask.vfmadd.sd(<2 x double> %av, <2 x double> %bv, <2 x double> %av, i8 %c, i32 4)

  %sr = extractelement <2 x double> %vr, i32 0
  store double %sr, double* %a
  ret void
}

define void @fmadd_sd_maskz_memfold(double* %a, double* %b, i8 %c) {
; CHECK-LABEL: fmadd_sd_maskz_memfold:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; CHECK-NEXT:    kmovw %edx, %k1
; CHECK-NEXT:    vfmadd132sd (%rsi), %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    vmovlpd %xmm0, (%rdi)
; CHECK-NEXT:    retq
  %a.val = load double, double* %a
  %av0 = insertelement <2 x double> undef, double %a.val, i32 0
  %av = insertelement <2 x double> %av0, double 0.000000e+00, i32 1

  %b.val = load double, double* %b
  %bv0 = insertelement <2 x double> undef, double %b.val, i32 0
  %bv = insertelement <2 x double> %bv0, double 0.000000e+00, i32 1

  %vr = call <2 x double> @llvm.x86.avx512.maskz.vfmadd.sd(<2 x double> %av, <2 x double> %bv, <2 x double> %av, i8 %c, i32 4)

  %sr = extractelement <2 x double> %vr, i32 0
  store double %sr, double* %a
  ret void
}

declare <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)

define <2 x double>@test_int_x86_avx512_mask3_vfmsub_sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovapd %xmm2, %xmm3
; CHECK-NEXT:    vfmsub231sd %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm2, %xmm4
; CHECK-NEXT:    vfmsub231sd %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm5
; CHECK-NEXT:    vfmsub231sd {rz-sae}, %xmm1, %xmm0, %xmm5
; CHECK-NEXT:    vfmsub231sd {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddpd %xmm4, %xmm3, %xmm0
; CHECK-NEXT:    vaddpd %xmm2, %xmm5, %xmm1
; CHECK-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 4)
  %res2 = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 3)
  %res3 = call <2 x double> @llvm.x86.avx512.mask3.vfmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 3)
  %res4 = fadd <2 x double> %res, %res1
  %res5 = fadd <2 x double> %res2, %res3
  %res6 = fadd <2 x double> %res4, %res5
  ret <2 x double> %res6
}

declare <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)

define <4 x float>@test_int_x86_avx512_mask3_vfmsub_ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmsub_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovaps %xmm2, %xmm3
; CHECK-NEXT:    vfmsub231ss %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm2, %xmm4
; CHECK-NEXT:    vfmsub231ss %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm5
; CHECK-NEXT:    vfmsub231ss {rz-sae}, %xmm1, %xmm0, %xmm5
; CHECK-NEXT:    vfmsub231ss {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddps %xmm4, %xmm3, %xmm0
; CHECK-NEXT:    vaddps %xmm2, %xmm5, %xmm1
; CHECK-NEXT:    vaddps %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 4)
  %res2 = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 3)
  %res3 = call <4 x float> @llvm.x86.avx512.mask3.vfmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 3)
  %res4 = fadd <4 x float> %res, %res1
  %res5 = fadd <4 x float> %res2, %res3
  %res6 = fadd <4 x float> %res4, %res5
  ret <4 x float> %res6
}

declare <2 x double> @llvm.x86.avx512.mask3.vfnmsub.sd(<2 x double>, <2 x double>, <2 x double>, i8, i32)

define <2 x double>@test_int_x86_avx512_mask3_vfnmsub_sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfnmsub_sd:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovapd %xmm2, %xmm3
; CHECK-NEXT:    vfnmsub231sd %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovapd %xmm2, %xmm4
; CHECK-NEXT:    vfnmsub231sd %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vmovapd %xmm2, %xmm5
; CHECK-NEXT:    vfnmsub231sd {rz-sae}, %xmm1, %xmm0, %xmm5
; CHECK-NEXT:    vfnmsub231sd {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddpd %xmm4, %xmm3, %xmm0
; CHECK-NEXT:    vaddpd %xmm2, %xmm5, %xmm1
; CHECK-NEXT:    vaddpd %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <2 x double> @llvm.x86.avx512.mask3.vfnmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 4)
  %res1 = call <2 x double> @llvm.x86.avx512.mask3.vfnmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 4)
  %res2 = call <2 x double> @llvm.x86.avx512.mask3.vfnmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 -1, i32 3)
  %res3 = call <2 x double> @llvm.x86.avx512.mask3.vfnmsub.sd(<2 x double> %x0, <2 x double> %x1, <2 x double> %x2, i8 %x3, i32 3)
  %res4 = fadd <2 x double> %res, %res1
  %res5 = fadd <2 x double> %res2, %res3
  %res6 = fadd <2 x double> %res4, %res5
  ret <2 x double> %res6
}

declare <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ss(<4 x float>, <4 x float>, <4 x float>, i8, i32)

define <4 x float>@test_int_x86_avx512_mask3_vfnmsub_ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3,i32 %x4 ){
; CHECK-LABEL: test_int_x86_avx512_mask3_vfnmsub_ss:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vmovaps %xmm2, %xmm3
; CHECK-NEXT:    vfnmsub231ss %xmm1, %xmm0, %xmm3
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vmovaps %xmm2, %xmm4
; CHECK-NEXT:    vfnmsub231ss %xmm1, %xmm0, %xmm4 {%k1}
; CHECK-NEXT:    vmovaps %xmm2, %xmm5
; CHECK-NEXT:    vfnmsub231ss {rz-sae}, %xmm1, %xmm0, %xmm5
; CHECK-NEXT:    vfnmsub231ss {rz-sae}, %xmm1, %xmm0, %xmm2 {%k1}
; CHECK-NEXT:    vaddps %xmm4, %xmm3, %xmm0
; CHECK-NEXT:    vaddps %xmm2, %xmm5, %xmm1
; CHECK-NEXT:    vaddps %xmm1, %xmm0, %xmm0
; CHECK-NEXT:    retq
  %res = call <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 4)
  %res1 = call <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 4)
  %res2 = call <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 -1, i32 3)
  %res3 = call <4 x float> @llvm.x86.avx512.mask3.vfnmsub.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %x2, i8 %x3, i32 3)
  %res4 = fadd <4 x float> %res, %res1
  %res5 = fadd <4 x float> %res2, %res3
  %res6 = fadd <4 x float> %res4, %res5
  ret <4 x float> %res6
}

define <4 x float>@test_int_x86_avx512_mask3_vfmadd_ss_rm(<4 x float> %x0, <4 x float> %x1, float *%ptr_b ,i8 %x3,i32 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask3_vfmadd_ss_rm:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vfmadd231ss (%rdi), %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %q = load float, float* %ptr_b
  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
  %res = call <4 x float> @llvm.x86.avx512.mask3.vfmadd.ss(<4 x float> %x0, <4 x float> %vecinit.i, <4 x float> %x1, i8 %x3, i32 4)
  ret < 4 x float> %res
}

define <4 x float>@test_int_x86_avx512_mask_vfmadd_ss_rm(<4 x float> %x0, <4 x float> %x1,float *%ptr_b ,i8 %x3,i32 %x4) {
; CHECK-LABEL: test_int_x86_avx512_mask_vfmadd_ss_rm:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %esi, %k1
; CHECK-NEXT:    vfmadd132ss (%rdi), %xmm1, %xmm0 {%k1}
; CHECK-NEXT:    retq
  %q = load float, float* %ptr_b
  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
  %res = call <4 x float> @llvm.x86.avx512.mask.vfmadd.ss(<4 x float> %x0,<4 x float> %vecinit.i, <4 x float> %x1,  i8 %x3, i32 4)
  ret < 4 x float> %res
}


define <4 x float>@test_int_x86_avx512_maskz_vfmadd_ss_rm(<4 x float> %x0, <4 x float> %x1,float *%ptr_b ,i8 %x3,i32 %x4) {
; CHECK-LABEL: test_int_x86_avx512_maskz_vfmadd_ss_rm:
; CHECK:       ## BB#0:
; CHECK-NEXT:    xorl %eax, %eax
; CHECK-NEXT:    kmovw %eax, %k1
; CHECK-NEXT:    vfmadd213ss (%rdi), %xmm1, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %q = load float, float* %ptr_b
  %vecinit.i = insertelement <4 x float> undef, float %q, i32 0
  %res = call <4 x float> @llvm.x86.avx512.maskz.vfmadd.ss(<4 x float> %x0, <4 x float> %x1, <4 x float> %vecinit.i, i8 0, i32 4)
  ret < 4 x float> %res
}

define <16 x i32> @test_x86_avx512_psll_d_512(<16 x i32> %a0, <4 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psll_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpslld %xmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psll.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psll_d_512(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %passthru, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psll_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpslld %xmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psll.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %passthru
  ret <16 x i32> %res2
}
define <16 x i32> @test_x86_avx512_maskz_psll_d_512(<16 x i32> %a0, <4 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psll_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpslld %xmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psll.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}
declare <16 x i32> @llvm.x86.avx512.psll.d.512(<16 x i32>, <4 x i32>) nounwind readnone


define <8 x i64> @test_x86_avx512_psll_q_512(<8 x i64> %a0, <2 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psll_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsllq %xmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psll.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psll_q_512(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %passthru, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psll_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllq %xmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psll.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %passthru
  ret <8 x i64> %res2
}
define <8 x i64> @test_x86_avx512_maskz_psll_q_512(<8 x i64> %a0, <2 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psll_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllq %xmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psll.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}
declare <8 x i64> @llvm.x86.avx512.psll.q.512(<8 x i64>, <2 x i64>) nounwind readnone


define <16 x i32> @test_x86_avx512_pslli_d_512(<16 x i32> %a0) {
; CHECK-LABEL: test_x86_avx512_pslli_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpslld $7, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.pslli.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_pslli_d_512(<16 x i32> %a0, <16 x i32> %passthru, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_pslli_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpslld $7, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.pslli.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %passthru
  ret <16 x i32> %res2
}
define <16 x i32> @test_x86_avx512_maskz_pslli_d_512(<16 x i32> %a0, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_pslli_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpslld $7, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.pslli.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}
declare <16 x i32> @llvm.x86.avx512.pslli.d.512(<16 x i32>, i32) nounwind readnone


define <8 x i64> @test_x86_avx512_pslli_q_512(<8 x i64> %a0) {
; CHECK-LABEL: test_x86_avx512_pslli_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsllq $7, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.pslli.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_pslli_q_512(<8 x i64> %a0, <8 x i64> %passthru, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_pslli_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllq $7, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.pslli.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %passthru
  ret <8 x i64> %res2
}
define <8 x i64> @test_x86_avx512_maskz_pslli_q_512(<8 x i64> %a0, <8 x i64> %passthru, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_pslli_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllq $7, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.pslli.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}
declare <8 x i64> @llvm.x86.avx512.pslli.q.512(<8 x i64>, i32) nounwind readnone


define <8 x i64> @test_x86_avx512_psra_q_512(<8 x i64> %a0, <2 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psra_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsraq %xmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psra.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psra_q_512(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %passthru, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psra_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsraq %xmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psra.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %passthru
  ret <8 x i64> %res2
}
define <8 x i64> @test_x86_avx512_maskz_psra_q_512(<8 x i64> %a0, <2 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psra_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsraq %xmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psra.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}
declare <8 x i64> @llvm.x86.avx512.psra.q.512(<8 x i64>, <2 x i64>) nounwind readnone


define <16 x i32> @test_x86_avx512_psra_d_512(<16 x i32> %a0, <4 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psra_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrad %xmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psra.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psra_d_512(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %passthru, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psra_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrad %xmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psra.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %passthru
  ret <16 x i32> %res2
}
define <16 x i32> @test_x86_avx512_maskz_psra_d_512(<16 x i32> %a0, <4 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psra_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrad %xmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psra.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}
declare <16 x i32> @llvm.x86.avx512.psra.d.512(<16 x i32>, <4 x i32>) nounwind readnone


define <8 x i64> @test_x86_avx512_psrai_q_512(<8 x i64> %a0) {
; CHECK-LABEL: test_x86_avx512_psrai_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsraq $7, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrai.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psrai_q_512(<8 x i64> %a0, <8 x i64> %passthru, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrai_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsraq $7, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrai.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %passthru
  ret <8 x i64> %res2
}
define <8 x i64> @test_x86_avx512_maskz_psrai_q_512(<8 x i64> %a0, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrai_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsraq $7, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrai.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}
declare <8 x i64> @llvm.x86.avx512.psrai.q.512(<8 x i64>, i32) nounwind readnone


define <16 x i32> @test_x86_avx512_psrai_d_512(<16 x i32> %a0) {
; CHECK-LABEL: test_x86_avx512_psrai_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrad $7, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrai.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psrai_d_512(<16 x i32> %a0, <16 x i32> %passthru, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrai_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrad $7, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrai.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %passthru
  ret <16 x i32> %res2
}
define <16 x i32> @test_x86_avx512_maskz_psrai_d_512(<16 x i32> %a0, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrai_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrad $7, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrai.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}
declare <16 x i32> @llvm.x86.avx512.psrai.d.512(<16 x i32>, i32) nounwind readnone


define <16 x i32> @test_x86_avx512_psrl_d_512(<16 x i32> %a0, <4 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psrl_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrld %xmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrl.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psrl_d_512(<16 x i32> %a0, <4 x i32> %a1, <16 x i32> %passthru, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrl_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrld %xmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrl.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %passthru
  ret <16 x i32> %res2
}
define <16 x i32> @test_x86_avx512_maskz_psrl_d_512(<16 x i32> %a0, <4 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrl_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrld %xmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrl.d.512(<16 x i32> %a0, <4 x i32> %a1) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}
declare <16 x i32> @llvm.x86.avx512.psrl.d.512(<16 x i32>, <4 x i32>) nounwind readnone


define <8 x i64> @test_x86_avx512_psrl_q_512(<8 x i64> %a0, <2 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psrl_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrlq %xmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrl.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psrl_q_512(<8 x i64> %a0, <2 x i64> %a1, <8 x i64> %passthru, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrl_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlq %xmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrl.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %passthru
  ret <8 x i64> %res2
}
define <8 x i64> @test_x86_avx512_maskz_psrl_q_512(<8 x i64> %a0, <2 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrl_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlq %xmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrl.q.512(<8 x i64> %a0, <2 x i64> %a1) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}
declare <8 x i64> @llvm.x86.avx512.psrl.q.512(<8 x i64>, <2 x i64>) nounwind readnone


define <16 x i32> @test_x86_avx512_psrli_d_512(<16 x i32> %a0) {
; CHECK-LABEL: test_x86_avx512_psrli_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrld $7, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrli.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  ret <16 x i32> %res
}
define <16 x i32> @test_x86_avx512_mask_psrli_d_512(<16 x i32> %a0, <16 x i32> %passthru, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrli_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrld $7, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrli.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %passthru
  ret <16 x i32> %res2
}
define <16 x i32> @test_x86_avx512_maskz_psrli_d_512(<16 x i32> %a0, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrli_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrld $7, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrli.d.512(<16 x i32> %a0, i32 7) ; <<16 x i32>> [#uses=1]
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}
declare <16 x i32> @llvm.x86.avx512.psrli.d.512(<16 x i32>, i32) nounwind readnone


define <8 x i64> @test_x86_avx512_psrli_q_512(<8 x i64> %a0) {
; CHECK-LABEL: test_x86_avx512_psrli_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrlq $7, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrli.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  ret <8 x i64> %res
}
define <8 x i64> @test_x86_avx512_mask_psrli_q_512(<8 x i64> %a0, <8 x i64> %passthru, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrli_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlq $7, %zmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm1, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrli.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %passthru
  ret <8 x i64> %res2
}
define <8 x i64> @test_x86_avx512_maskz_psrli_q_512(<8 x i64> %a0, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrli_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlq $7, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrli.q.512(<8 x i64> %a0, i32 7) ; <<8 x i64>> [#uses=1]
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}
declare <8 x i64> @llvm.x86.avx512.psrli.q.512(<8 x i64>, i32) nounwind readnone

define <16 x i32> @test_x86_avx512_psllv_d_512(<16 x i32> %a0, <16 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psllv_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32> %a0, <16 x i32> %a1)
  ret <16 x i32> %res
}

define <16 x i32> @test_x86_avx512_mask_psllv_d_512(<16 x i32> %a0, <16 x i32> %a1, <16 x i32> %a2, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psllv_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllvd %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32> %a0, <16 x i32> %a1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %a2
  ret <16 x i32> %res2
}

define <16 x i32> @test_x86_avx512_maskz_psllv_d_512(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psllv_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllvd %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32> %a0, <16 x i32> %a1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}

declare <16 x i32> @llvm.x86.avx512.psllv.d.512(<16 x i32>, <16 x i32>) nounwind readnone

define <8 x i64> @test_x86_avx512_psllv_q_512(<8 x i64> %a0, <8 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psllv_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsllvq %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64> %a0, <8 x i64> %a1)
  ret <8 x i64> %res
}

define <8 x i64> @test_x86_avx512_mask_psllv_q_512(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> %a2, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psllv_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllvq %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64> %a0, <8 x i64> %a1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %a2
  ret <8 x i64> %res2
}

define <8 x i64> @test_x86_avx512_maskz_psllv_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psllv_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsllvq %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64> %a0, <8 x i64> %a1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}

declare <8 x i64> @llvm.x86.avx512.psllv.q.512(<8 x i64>, <8 x i64>) nounwind readnone

define <16 x i32> @test_x86_avx512_psrav_d_512(<16 x i32> %a0, <16 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psrav_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsravd %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32> %a0, <16 x i32> %a1)
  ret <16 x i32> %res
}

define <16 x i32> @test_x86_avx512_mask_psrav_d_512(<16 x i32> %a0, <16 x i32> %a1, <16 x i32> %a2, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrav_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsravd %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32> %a0, <16 x i32> %a1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %a2
  ret <16 x i32> %res2
}

define <16 x i32> @test_x86_avx512_maskz_psrav_d_512(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrav_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsravd %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32> %a0, <16 x i32> %a1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}

declare <16 x i32> @llvm.x86.avx512.psrav.d.512(<16 x i32>, <16 x i32>) nounwind readnone

define <8 x i64> @test_x86_avx512_psrav_q_512(<8 x i64> %a0, <8 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psrav_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsravq %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64> %a0, <8 x i64> %a1)
  ret <8 x i64> %res
}

define <8 x i64> @test_x86_avx512_mask_psrav_q_512(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> %a2, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrav_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsravq %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64> %a0, <8 x i64> %a1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %a2
  ret <8 x i64> %res2
}

define <8 x i64> @test_x86_avx512_maskz_psrav_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrav_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsravq %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64> %a0, <8 x i64> %a1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}

declare <8 x i64> @llvm.x86.avx512.psrav.q.512(<8 x i64>, <8 x i64>) nounwind readnone

define <16 x i32> @test_x86_avx512_psrlv_d_512(<16 x i32> %a0, <16 x i32> %a1) {
; CHECK-LABEL: test_x86_avx512_psrlv_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32> %a0, <16 x i32> %a1)
  ret <16 x i32> %res
}

define <16 x i32> @test_x86_avx512_mask_psrlv_d_512(<16 x i32> %a0, <16 x i32> %a1, <16 x i32> %a2, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrlv_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32> %a0, <16 x i32> %a1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> %a2
  ret <16 x i32> %res2
}

define <16 x i32> @test_x86_avx512_maskz_psrlv_d_512(<16 x i32> %a0, <16 x i32> %a1, i16 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrlv_d_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlvd %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32> %a0, <16 x i32> %a1)
  %mask.cast = bitcast i16 %mask to <16 x i1>
  %res2 = select <16 x i1> %mask.cast, <16 x i32> %res, <16 x i32> zeroinitializer
  ret <16 x i32> %res2
}

declare <16 x i32> @llvm.x86.avx512.psrlv.d.512(<16 x i32>, <16 x i32>) nounwind readnone

define <8 x i64> @test_x86_avx512_psrlv_q_512(<8 x i64> %a0, <8 x i64> %a1) {
; CHECK-LABEL: test_x86_avx512_psrlv_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    vpsrlvq %zmm1, %zmm0, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64> %a0, <8 x i64> %a1)
  ret <8 x i64> %res
}

define <8 x i64> @test_x86_avx512_mask_psrlv_q_512(<8 x i64> %a0, <8 x i64> %a1, <8 x i64> %a2, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_mask_psrlv_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlvq %zmm1, %zmm0, %zmm2 {%k1}
; CHECK-NEXT:    vmovdqa64 %zmm2, %zmm0
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64> %a0, <8 x i64> %a1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> %a2
  ret <8 x i64> %res2
}

define <8 x i64> @test_x86_avx512_maskz_psrlv_q_512(<8 x i64> %a0, <8 x i64> %a1, i8 %mask) {
; CHECK-LABEL: test_x86_avx512_maskz_psrlv_q_512:
; CHECK:       ## BB#0:
; CHECK-NEXT:    kmovw %edi, %k1
; CHECK-NEXT:    vpsrlvq %zmm1, %zmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %res = call <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64> %a0, <8 x i64> %a1)
  %mask.cast = bitcast i8 %mask to <8 x i1>
  %res2 = select <8 x i1> %mask.cast, <8 x i64> %res, <8 x i64> zeroinitializer
  ret <8 x i64> %res2
}

declare <8 x i64> @llvm.x86.avx512.psrlv.q.512(<8 x i64>, <8 x i64>) nounwind readnone