test/CodeGen/X86/avx512-shuffles/in_lane_permute.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc -mtriple=x86_64-unknown-linux-gnu -mattr=+avx512f,+avx512vl %s -o - | FileCheck %s

; FIXME: The non immediate <16 x float> test cases should be fixed by PR34382

define <4 x float> @test_4xfloat_perm_mask0(<4 x float> %vec) {
; CHECK-LABEL: test_4xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[2,1,3,1]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 2, i32 1, i32 3, i32 1>
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_perm_mask0(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm1 {%k1} = xmm0[2,1,3,1]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 2, i32 1, i32 3, i32 1>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mask0(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = xmm0[2,1,3,1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 2, i32 1, i32 3, i32 1>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_perm_mask1(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm1 {%k1} = xmm0[1,2,3,2]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 2>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mask1(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = xmm0[1,2,3,2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 2>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_perm_mask2(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm1 {%k1} = xmm0[1,3,2,1]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 2, i32 1>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mask2(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = xmm0[1,3,2,1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 2, i32 1>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_perm_mask3(<4 x float> %vec) {
; CHECK-LABEL: test_4xfloat_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 = xmm0[1,2,3,2]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 2>
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_perm_mask3(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm1 {%k1} = xmm0[1,2,3,2]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 2>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mask3(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = xmm0[1,2,3,2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 2, i32 3, i32 2>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_perm_mem_mask0(<4 x float>* %vp) {
; CHECK-LABEL: test_4xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 = mem[3,3,1,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %res = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 3>
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_perm_mem_mask0(<4 x float>* %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} = mem[3,3,1,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 3>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mem_mask0(<4 x float>* %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = mem[3,3,1,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 3, i32 3, i32 1, i32 3>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}

define <4 x float> @test_masked_4xfloat_perm_mem_mask1(<4 x float>* %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} = mem[1,3,2,0]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mem_mask1(<4 x float>* %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = mem[1,3,2,0]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 3, i32 2, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}

define <4 x float> @test_masked_4xfloat_perm_mem_mask2(<4 x float>* %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} = mem[2,1,3,2]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 2, i32 1, i32 3, i32 2>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mem_mask2(<4 x float>* %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = mem[2,1,3,2]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 2, i32 1, i32 3, i32 2>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_perm_mem_mask3(<4 x float>* %vp) {
; CHECK-LABEL: test_4xfloat_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 = mem[0,1,3,0]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %res = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 3, i32 0>
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_perm_mem_mask3(<4 x float>* %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} = mem[0,1,3,0]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 3, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec2
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_4xfloat_perm_mem_mask3(<4 x float>* %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} xmm0 {%k1} {z} = mem[0,1,3,0]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, <4 x float>* %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 0, i32 1, i32 3, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}

define <8 x float> @test_8xfloat_perm_mask0(<8 x float> %vec) {
; CHECK-LABEL: test_8xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,1,2,3,4,6,6,6]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 6, i32 6, i32 6>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_mask0(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[0,1,2,3,4,6,6,6]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 6, i32 6, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mask0(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[0,1,2,3,4,6,6,6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 6, i32 6, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_imm_mask1(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[3,2,3,2,7,6,7,6]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 2, i32 3, i32 2, i32 7, i32 6, i32 7, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mask1(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[3,2,3,2,7,6,7,6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 2, i32 3, i32 2, i32 7, i32 6, i32 7, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_mask2(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[2,1,2,1,6,5,4,4]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 1, i32 2, i32 1, i32 6, i32 5, i32 4, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mask2(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[2,1,2,1,6,5,4,4]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 1, i32 2, i32 1, i32 6, i32 5, i32 4, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_perm_imm_mask3(<8 x float> %vec) {
; CHECK-LABEL: test_8xfloat_perm_imm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[2,2,1,0,6,6,5,4]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 2, i32 1, i32 0, i32 6, i32 6, i32 5, i32 4>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_imm_mask3(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[2,2,1,0,6,6,5,4]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 2, i32 1, i32 0, i32 6, i32 6, i32 5, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mask3(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[2,2,1,0,6,6,5,4]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 2, i32 1, i32 0, i32 6, i32 6, i32 5, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_mask4(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[3,3,3,3,7,7,6,5]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 7, i32 7, i32 6, i32 5>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mask4(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[3,3,3,3,7,7,6,5]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 3, i32 3, i32 3, i32 7, i32 7, i32 6, i32 5>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_imm_mask5(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[2,1,3,3,6,5,7,7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 1, i32 3, i32 3, i32 6, i32 5, i32 7, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mask5(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[2,1,3,3,6,5,7,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 1, i32 3, i32 3, i32 6, i32 5, i32 7, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_perm_mask6(<8 x float> %vec) {
; CHECK-LABEL: test_8xfloat_perm_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,2,3,2,5,6,7,7]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 2, i32 3, i32 2, i32 5, i32 6, i32 7, i32 7>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_mask6(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[3,2,3,2,5,6,7,7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 2, i32 3, i32 2, i32 5, i32 6, i32 7, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mask6(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[3,2,3,2,5,6,7,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 2, i32 3, i32 2, i32 5, i32 6, i32 7, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_imm_mask7(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm1 {%k1} = ymm0[3,0,2,1,7,4,6,5]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 0, i32 2, i32 1, i32 7, i32 4, i32 6, i32 5>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mask7(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm0[3,0,2,1,7,4,6,5]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 0, i32 2, i32 1, i32 7, i32 4, i32 6, i32 5>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_perm_mem_mask0(<8 x float>* %vp) {
; CHECK-LABEL: test_8xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm0
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[3,0,0,2,4,6,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 0, i32 0, i32 2, i32 4, i32 6, i32 7, i32 6>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_mem_mask0(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = ymm2[3,0,0,2,4,6,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 0, i32 0, i32 2, i32 4, i32 6, i32 7, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mem_mask0(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm1[3,0,0,2,4,6,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 3, i32 0, i32 0, i32 2, i32 4, i32 6, i32 7, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <8 x float> @test_masked_8xfloat_perm_imm_mem_mask1(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = mem[2,0,2,2,6,4,6,6]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 0, i32 2, i32 2, i32 6, i32 4, i32 6, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mem_mask1(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = mem[2,0,2,2,6,4,6,6]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 0, i32 2, i32 2, i32 6, i32 4, i32 6, i32 6>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <8 x float> @test_masked_8xfloat_perm_mem_mask2(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = ymm2[2,1,1,3,4,4,7,4]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 1, i32 1, i32 3, i32 4, i32 4, i32 7, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mem_mask2(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm1[2,1,1,3,4,4,7,4]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 1, i32 1, i32 3, i32 4, i32 4, i32 7, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_perm_imm_mem_mask3(<8 x float>* %vp) {
; CHECK-LABEL: test_8xfloat_perm_imm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 = mem[0,0,3,3,4,4,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 3, i32 3, i32 4, i32 4, i32 7, i32 7>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_imm_mem_mask3(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = mem[0,0,3,3,4,4,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 3, i32 3, i32 4, i32 4, i32 7, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mem_mask3(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = mem[0,0,3,3,4,4,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 0, i32 3, i32 3, i32 4, i32 4, i32 7, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <8 x float> @test_masked_8xfloat_perm_mem_mask4(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = ymm2[0,1,0,1,4,6,5,4]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 6, i32 5, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mem_mask4(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm1[0,1,0,1,4,6,5,4]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 6, i32 5, i32 4>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <8 x float> @test_masked_8xfloat_perm_imm_mem_mask5(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mem_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = mem[2,0,0,3,6,4,4,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 0, i32 0, i32 3, i32 6, i32 4, i32 4, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mem_mask5(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mem_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = mem[2,0,0,3,6,4,4,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 2, i32 0, i32 0, i32 3, i32 6, i32 4, i32 4, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_perm_mem_mask6(<8 x float>* %vp) {
; CHECK-LABEL: test_8xfloat_perm_mem_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm0
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 = ymm0[0,1,2,3,7,4,6,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 6, i32 7>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_perm_mem_mask6(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_mem_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = ymm2[0,1,2,3,7,4,6,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 6, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_mem_mask6(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_mem_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %ymm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = ymm1[0,1,2,3,7,4,6,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 7, i32 4, i32 6, i32 7>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <8 x float> @test_masked_8xfloat_perm_imm_mem_mask7(<8 x float>* %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_perm_imm_mem_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} = mem[0,2,3,1,4,6,7,5]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 3, i32 1, i32 4, i32 6, i32 7, i32 5>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec2
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_8xfloat_perm_imm_mem_mask7(<8 x float>* %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_perm_imm_mem_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} ymm0 {%k1} {z} = mem[0,2,3,1,4,6,7,5]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, <8 x float>* %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 3, i32 1, i32 4, i32 6, i32 7, i32 5>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}

define <16 x float> @test_16xfloat_perm_mask0(<16 x float> %vec) {
; CHECK-LABEL: test_16xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 = zmm0[1,1,3,1,6,4,6,5,8,9,8,11,13,13,13,15]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 1, i32 6, i32 4, i32 6, i32 5, i32 8, i32 9, i32 8, i32 11, i32 13, i32 13, i32 13, i32 15>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_mask0(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[1,1,3,1,6,4,6,5,8,9,8,11,13,13,13,15]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 1, i32 6, i32 4, i32 6, i32 5, i32 8, i32 9, i32 8, i32 11, i32 13, i32 13, i32 13, i32 15>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mask0(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[1,1,3,1,6,4,6,5,8,9,8,11,13,13,13,15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 1, i32 6, i32 4, i32 6, i32 5, i32 8, i32 9, i32 8, i32 11, i32 13, i32 13, i32 13, i32 15>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_imm_mask1(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[2,2,2,1,6,6,6,5,10,10,10,9,14,14,14,13]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 2, i32 2, i32 1, i32 6, i32 6, i32 6, i32 5, i32 10, i32 10, i32 10, i32 9, i32 14, i32 14, i32 14, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mask1(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[2,2,2,1,6,6,6,5,10,10,10,9,14,14,14,13]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 2, i32 2, i32 1, i32 6, i32 6, i32 6, i32 5, i32 10, i32 10, i32 10, i32 9, i32 14, i32 14, i32 14, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_mask2(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[1,2,0,0,5,4,6,5,11,10,9,9,14,13,14,12]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 0, i32 0, i32 5, i32 4, i32 6, i32 5, i32 11, i32 10, i32 9, i32 9, i32 14, i32 13, i32 14, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mask2(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[1,2,0,0,5,4,6,5,11,10,9,9,14,13,14,12]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 0, i32 0, i32 5, i32 4, i32 6, i32 5, i32 11, i32 10, i32 9, i32 9, i32 14, i32 13, i32 14, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_perm_imm_mask3(<16 x float> %vec) {
; CHECK-LABEL: test_16xfloat_perm_imm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 = zmm0[1,1,0,2,5,5,4,6,9,9,8,10,13,13,12,14]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 0, i32 2, i32 5, i32 5, i32 4, i32 6, i32 9, i32 9, i32 8, i32 10, i32 13, i32 13, i32 12, i32 14>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_imm_mask3(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[1,1,0,2,5,5,4,6,9,9,8,10,13,13,12,14]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 0, i32 2, i32 5, i32 5, i32 4, i32 6, i32 9, i32 9, i32 8, i32 10, i32 13, i32 13, i32 12, i32 14>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mask3(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[1,1,0,2,5,5,4,6,9,9,8,10,13,13,12,14]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 0, i32 2, i32 5, i32 5, i32 4, i32 6, i32 9, i32 9, i32 8, i32 10, i32 13, i32 13, i32 12, i32 14>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_mask4(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[1,2,3,3,5,5,5,7,11,11,8,11,14,12,14,15]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 3, i32 3, i32 5, i32 5, i32 5, i32 7, i32 11, i32 11, i32 8, i32 11, i32 14, i32 12, i32 14, i32 15>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mask4(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[1,2,3,3,5,5,5,7,11,11,8,11,14,12,14,15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 3, i32 3, i32 5, i32 5, i32 5, i32 7, i32 11, i32 11, i32 8, i32 11, i32 14, i32 12, i32 14, i32 15>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_imm_mask5(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[1,2,1,0,5,6,5,4,9,10,9,8,13,14,13,12]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 1, i32 0, i32 5, i32 6, i32 5, i32 4, i32 9, i32 10, i32 9, i32 8, i32 13, i32 14, i32 13, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mask5(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[1,2,1,0,5,6,5,4,9,10,9,8,13,14,13,12]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 1, i32 0, i32 5, i32 6, i32 5, i32 4, i32 9, i32 10, i32 9, i32 8, i32 13, i32 14, i32 13, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_perm_mask6(<16 x float> %vec) {
; CHECK-LABEL: test_16xfloat_perm_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 = zmm0[2,0,3,2,4,4,6,7,9,11,8,11,13,12,13,13]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 0, i32 3, i32 2, i32 4, i32 4, i32 6, i32 7, i32 9, i32 11, i32 8, i32 11, i32 13, i32 12, i32 13, i32 13>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_mask6(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[2,0,3,2,4,4,6,7,9,11,8,11,13,12,13,13]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 0, i32 3, i32 2, i32 4, i32 4, i32 6, i32 7, i32 9, i32 11, i32 8, i32 11, i32 13, i32 12, i32 13, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mask6(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[2,0,3,2,4,4,6,7,9,11,8,11,13,12,13,13]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 0, i32 3, i32 2, i32 4, i32 4, i32 6, i32 7, i32 9, i32 11, i32 8, i32 11, i32 13, i32 12, i32 13, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_imm_mask7(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm1 {%k1} = zmm0[3,3,0,2,7,7,4,6,11,11,8,10,15,15,12,14]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 3, i32 3, i32 0, i32 2, i32 7, i32 7, i32 4, i32 6, i32 11, i32 11, i32 8, i32 10, i32 15, i32 15, i32 12, i32 14>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mask7(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm0[3,3,0,2,7,7,4,6,11,11,8,10,15,15,12,14]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 3, i32 3, i32 0, i32 2, i32 7, i32 7, i32 4, i32 6, i32 11, i32 11, i32 8, i32 10, i32 15, i32 15, i32 12, i32 14>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_perm_mem_mask0(<16 x float>* %vp) {
; CHECK-LABEL: test_16xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm0
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 = zmm0[3,3,3,0,6,6,6,6,11,10,9,10,12,14,12,12]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 3, i32 3, i32 3, i32 0, i32 6, i32 6, i32 6, i32 6, i32 11, i32 10, i32 9, i32 10, i32 12, i32 14, i32 12, i32 12>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_mem_mask0(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = zmm2[3,3,3,0,6,6,6,6,11,10,9,10,12,14,12,12]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 3, i32 3, i32 3, i32 0, i32 6, i32 6, i32 6, i32 6, i32 11, i32 10, i32 9, i32 10, i32 12, i32 14, i32 12, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mem_mask0(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm1[3,3,3,0,6,6,6,6,11,10,9,10,12,14,12,12]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 3, i32 3, i32 3, i32 0, i32 6, i32 6, i32 6, i32 6, i32 11, i32 10, i32 9, i32 10, i32 12, i32 14, i32 12, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <16 x float> @test_masked_16xfloat_perm_imm_mem_mask1(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = mem[1,3,2,1,5,7,6,5,9,11,10,9,13,15,14,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 3, i32 2, i32 1, i32 5, i32 7, i32 6, i32 5, i32 9, i32 11, i32 10, i32 9, i32 13, i32 15, i32 14, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mem_mask1(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = mem[1,3,2,1,5,7,6,5,9,11,10,9,13,15,14,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 3, i32 2, i32 1, i32 5, i32 7, i32 6, i32 5, i32 9, i32 11, i32 10, i32 9, i32 13, i32 15, i32 14, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <16 x float> @test_masked_16xfloat_perm_mem_mask2(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = zmm2[2,0,0,3,5,5,6,5,9,8,8,8,14,12,13,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 0, i32 0, i32 3, i32 5, i32 5, i32 6, i32 5, i32 9, i32 8, i32 8, i32 8, i32 14, i32 12, i32 13, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mem_mask2(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm1[2,0,0,3,5,5,6,5,9,8,8,8,14,12,13,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 0, i32 0, i32 3, i32 5, i32 5, i32 6, i32 5, i32 9, i32 8, i32 8, i32 8, i32 14, i32 12, i32 13, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_perm_imm_mem_mask3(<16 x float>* %vp) {
; CHECK-LABEL: test_16xfloat_perm_imm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 = mem[1,0,3,1,5,4,7,5,9,8,11,9,13,12,15,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 0, i32 3, i32 1, i32 5, i32 4, i32 7, i32 5, i32 9, i32 8, i32 11, i32 9, i32 13, i32 12, i32 15, i32 13>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_imm_mem_mask3(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = mem[1,0,3,1,5,4,7,5,9,8,11,9,13,12,15,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 0, i32 3, i32 1, i32 5, i32 4, i32 7, i32 5, i32 9, i32 8, i32 11, i32 9, i32 13, i32 12, i32 15, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mem_mask3(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = mem[1,0,3,1,5,4,7,5,9,8,11,9,13,12,15,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 0, i32 3, i32 1, i32 5, i32 4, i32 7, i32 5, i32 9, i32 8, i32 11, i32 9, i32 13, i32 12, i32 15, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <16 x float> @test_masked_16xfloat_perm_mem_mask4(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = zmm2[3,3,1,1,6,5,5,6,11,11,10,9,15,14,12,12]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 3, i32 3, i32 1, i32 1, i32 6, i32 5, i32 5, i32 6, i32 11, i32 11, i32 10, i32 9, i32 15, i32 14, i32 12, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mem_mask4(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm1[3,3,1,1,6,5,5,6,11,11,10,9,15,14,12,12]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 3, i32 3, i32 1, i32 1, i32 6, i32 5, i32 5, i32 6, i32 11, i32 11, i32 10, i32 9, i32 15, i32 14, i32 12, i32 12>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <16 x float> @test_masked_16xfloat_perm_imm_mem_mask5(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mem_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = mem[2,0,0,1,6,4,4,5,10,8,8,9,14,12,12,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 0, i32 0, i32 1, i32 6, i32 4, i32 4, i32 5, i32 10, i32 8, i32 8, i32 9, i32 14, i32 12, i32 12, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mem_mask5(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mem_mask5:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = mem[2,0,0,1,6,4,4,5,10,8,8,9,14,12,12,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 0, i32 0, i32 1, i32 6, i32 4, i32 4, i32 5, i32 10, i32 8, i32 8, i32 9, i32 14, i32 12, i32 12, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_perm_mem_mask6(<16 x float>* %vp) {
; CHECK-LABEL: test_16xfloat_perm_mem_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm0
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 = zmm0[2,1,1,2,6,5,5,7,9,11,9,9,12,15,14,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 1, i32 1, i32 2, i32 6, i32 5, i32 5, i32 7, i32 9, i32 11, i32 9, i32 9, i32 12, i32 15, i32 14, i32 15>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_perm_mem_mask6(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_mem_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm2
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = zmm2[2,1,1,2,6,5,5,7,9,11,9,9,12,15,14,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 1, i32 1, i32 2, i32 6, i32 5, i32 5, i32 7, i32 9, i32 11, i32 9, i32 9, i32 12, i32 15, i32 14, i32 15>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_mem_mask6(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_mem_mask6:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovaps (%rdi), %zmm1
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = zmm1[2,1,1,2,6,5,5,7,9,11,9,9,12,15,14,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 2, i32 1, i32 1, i32 2, i32 6, i32 5, i32 5, i32 7, i32 9, i32 11, i32 9, i32 9, i32 12, i32 15, i32 14, i32 15>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <16 x float> @test_masked_16xfloat_perm_imm_mem_mask7(<16 x float>* %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_perm_imm_mem_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} = mem[1,2,0,1,5,6,4,5,9,10,8,9,13,14,12,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 0, i32 1, i32 5, i32 6, i32 4, i32 5, i32 9, i32 10, i32 8, i32 9, i32 13, i32 14, i32 12, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec2
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_16xfloat_perm_imm_mem_mask7(<16 x float>* %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_perm_imm_mem_mask7:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilps {{.*#+}} zmm0 {%k1} {z} = mem[1,2,0,1,5,6,4,5,9,10,8,9,13,14,12,13]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, <16 x float>* %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 2, i32 0, i32 1, i32 5, i32 6, i32 4, i32 5, i32 9, i32 10, i32 8, i32 9, i32 13, i32 14, i32 12, i32 13>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}

define <2 x double> @test_2xdouble_perm_mask0(<2 x double> %vec) {
; CHECK-LABEL: test_2xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; CHECK-NEXT:    retq
  %res = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  ret <2 x double> %res
}
define <2 x double> @test_masked_2xdouble_perm_mask0(<2 x double> %vec, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_masked_2xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm1 {%k1} = xmm0[1,0]
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec2
  ret <2 x double> %res
}

define <2 x double> @test_masked_z_2xdouble_perm_mask0(<2 x double> %vec, <2 x double> %mask) {
; CHECK-LABEL: test_masked_z_2xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 {%k1} {z} = xmm0[1,0]
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> zeroinitializer
  ret <2 x double> %res
}
define <2 x double> @test_masked_2xdouble_perm_mask1(<2 x double> %vec, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_masked_2xdouble_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm1 {%k1} = xmm0[1,0]
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec2
  ret <2 x double> %res
}

define <2 x double> @test_masked_z_2xdouble_perm_mask1(<2 x double> %vec, <2 x double> %mask) {
; CHECK-LABEL: test_masked_z_2xdouble_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 {%k1} {z} = xmm0[1,0]
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> zeroinitializer
  ret <2 x double> %res
}
define <2 x double> @test_2xdouble_perm_mem_mask0(<2 x double>* %vp) {
; CHECK-LABEL: test_2xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 = mem[1,0]
; CHECK-NEXT:    retq
  %vec = load <2 x double>, <2 x double>* %vp
  %res = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  ret <2 x double> %res
}
define <2 x double> @test_masked_2xdouble_perm_mem_mask0(<2 x double>* %vp, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_masked_2xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 {%k1} = mem[1,0]
; CHECK-NEXT:    retq
  %vec = load <2 x double>, <2 x double>* %vp
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec2
  ret <2 x double> %res
}

define <2 x double> @test_masked_z_2xdouble_perm_mem_mask0(<2 x double>* %vp, <2 x double> %mask) {
; CHECK-LABEL: test_masked_z_2xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %xmm1, %xmm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 {%k1} {z} = mem[1,0]
; CHECK-NEXT:    retq
  %vec = load <2 x double>, <2 x double>* %vp
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> zeroinitializer
  ret <2 x double> %res
}

define <2 x double> @test_masked_2xdouble_perm_mem_mask1(<2 x double>* %vp, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_masked_2xdouble_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 {%k1} = mem[1,0]
; CHECK-NEXT:    retq
  %vec = load <2 x double>, <2 x double>* %vp
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec2
  ret <2 x double> %res
}

define <2 x double> @test_masked_z_2xdouble_perm_mem_mask1(<2 x double>* %vp, <2 x double> %mask) {
; CHECK-LABEL: test_masked_z_2xdouble_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %xmm1, %xmm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} xmm0 {%k1} {z} = mem[1,0]
; CHECK-NEXT:    retq
  %vec = load <2 x double>, <2 x double>* %vp
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <2 x i32> <i32 1, i32 0>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> zeroinitializer
  ret <2 x double> %res
}

define <4 x double> @test_4xdouble_perm_mask0(<4 x double> %vec) {
; CHECK-LABEL: test_4xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 2, i32 3>
  ret <4 x double> %res
}
define <4 x double> @test_masked_4xdouble_perm_mask0(<4 x double> %vec, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm1 {%k1} = ymm0[1,0,2,3]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 2, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mask0(<4 x double> %vec, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1,0,2,3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 2, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_4xdouble_perm_mask1(<4 x double> %vec, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm1 {%k1} = ymm0[1,1,2,2]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mask1(<4 x double> %vec, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1,1,2,2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_4xdouble_perm_mask2(<4 x double> %vec, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm1 {%k1} = ymm0[0,1,3,3]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 1, i32 3, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mask2(<4 x double> %vec, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0,1,3,3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 1, i32 3, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_perm_mask3(<4 x double> %vec) {
; CHECK-LABEL: test_4xdouble_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 = ymm0[1,1,2,2]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2>
  ret <4 x double> %res
}
define <4 x double> @test_masked_4xdouble_perm_mask3(<4 x double> %vec, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm1 {%k1} = ymm0[1,1,2,2]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mask3(<4 x double> %vec, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1,1,2,2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 1, i32 2, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_perm_mem_mask0(<4 x double>* %vp) {
; CHECK-LABEL: test_4xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 = mem[0,1,2,2]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %res = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
  ret <4 x double> %res
}
define <4 x double> @test_masked_4xdouble_perm_mem_mask0(<4 x double>* %vp, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} = mem[0,1,2,2]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mem_mask0(<4 x double>* %vp, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = mem[0,1,2,2]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}

define <4 x double> @test_masked_4xdouble_perm_mem_mask1(<4 x double>* %vp, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} = mem[0,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 1, i32 3, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mem_mask1(<4 x double>* %vp, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = mem[0,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 0, i32 1, i32 3, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}

define <4 x double> @test_masked_4xdouble_perm_mem_mask2(<4 x double>* %vp, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} = mem[1,0,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mem_mask2(<4 x double>* %vp, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = mem[1,0,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 3>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_perm_mem_mask3(<4 x double>* %vp) {
; CHECK-LABEL: test_4xdouble_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 = mem[1,0,3,2]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %res = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
  ret <4 x double> %res
}
define <4 x double> @test_masked_4xdouble_perm_mem_mask3(<4 x double>* %vp, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_masked_4xdouble_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} = mem[1,0,3,2]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec2
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_4xdouble_perm_mem_mask3(<4 x double>* %vp, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_4xdouble_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} ymm0 {%k1} {z} = mem[1,0,3,2]
; CHECK-NEXT:    retq
  %vec = load <4 x double>, <4 x double>* %vp
  %shuf = shufflevector <4 x double> %vec, <4 x double> undef, <4 x i32> <i32 1, i32 0, i32 3, i32 2>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}

define <8 x double> @test_8xdouble_perm_mask0(<8 x double> %vec) {
; CHECK-LABEL: test_8xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 = zmm0[0,0,3,2,4,5,7,6]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 3, i32 2, i32 4, i32 5, i32 7, i32 6>
  ret <8 x double> %res
}
define <8 x double> @test_masked_8xdouble_perm_mask0(<8 x double> %vec, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm1 {%k1} = zmm0[0,0,3,2,4,5,7,6]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 3, i32 2, i32 4, i32 5, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mask0(<8 x double> %vec, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0,0,3,2,4,5,7,6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 3, i32 2, i32 4, i32 5, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_8xdouble_perm_mask1(<8 x double> %vec, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm1 {%k1} = zmm0[0,1,2,3,4,4,7,6]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 4, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mask1(<8 x double> %vec, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0,1,2,3,4,4,7,6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 4, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_8xdouble_perm_mask2(<8 x double> %vec, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm1 {%k1} = zmm0[0,0,2,3,5,5,6,7]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 5, i32 5, i32 6, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mask2(<8 x double> %vec, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0,0,2,3,5,5,6,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 5, i32 5, i32 6, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_perm_mask3(<8 x double> %vec) {
; CHECK-LABEL: test_8xdouble_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 = zmm0[0,1,2,2,4,4,6,7]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 2, i32 4, i32 4, i32 6, i32 7>
  ret <8 x double> %res
}
define <8 x double> @test_masked_8xdouble_perm_mask3(<8 x double> %vec, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm1 {%k1} = zmm0[0,1,2,2,4,4,6,7]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 2, i32 4, i32 4, i32 6, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mask3(<8 x double> %vec, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0,1,2,2,4,4,6,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 2, i32 4, i32 4, i32 6, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_perm_mem_mask0(<8 x double>* %vp) {
; CHECK-LABEL: test_8xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 = mem[0,1,2,3,5,4,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %res = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 7, i32 6>
  ret <8 x double> %res
}
define <8 x double> @test_masked_8xdouble_perm_mem_mask0(<8 x double>* %vp, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} = mem[0,1,2,3,5,4,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mem_mask0(<8 x double>* %vp, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = mem[0,1,2,3,5,4,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}

define <8 x double> @test_masked_8xdouble_perm_mem_mask1(<8 x double>* %vp, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} = mem[0,1,3,3,4,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 3, i32 3, i32 4, i32 5, i32 7, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mem_mask1(<8 x double>* %vp, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = mem[0,1,3,3,4,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 0, i32 1, i32 3, i32 3, i32 4, i32 5, i32 7, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}

define <8 x double> @test_masked_8xdouble_perm_mem_mask2(<8 x double>* %vp, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} = mem[1,1,3,3,5,4,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 4, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mem_mask2(<8 x double>* %vp, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = mem[1,1,3,3,5,4,7,6]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 4, i32 7, i32 6>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_perm_mem_mask3(<8 x double>* %vp) {
; CHECK-LABEL: test_8xdouble_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 = mem[1,0,3,2,4,5,6,7]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %res = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>
  ret <8 x double> %res
}
define <8 x double> @test_masked_8xdouble_perm_mem_mask3(<8 x double>* %vp, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_masked_8xdouble_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} = mem[1,0,3,2,4,5,6,7]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec2
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_8xdouble_perm_mem_mask3(<8 x double>* %vp, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_8xdouble_perm_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vpermilpd {{.*#+}} zmm0 {%k1} {z} = mem[1,0,3,2,4,5,6,7]
; CHECK-NEXT:    retq
  %vec = load <8 x double>, <8 x double>* %vp
  %shuf = shufflevector <8 x double> %vec, <8 x double> undef, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}