; 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

define <4 x float> @test_4xfloat_dup_high(<4 x float> %vec) {
; CHECK-LABEL: test_4xfloat_dup_high:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, i32 3>
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_dup_high_mask0(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm1 {%k1} = xmm0[1,1,3,3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask0(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = xmm0[1,1,3,3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask1(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm1 {%k1} = xmm0[1,1,3,3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask1(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = xmm0[1,1,3,3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask2(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm1 {%k1} = xmm0[1,1,3,3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask2(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = xmm0[1,1,3,3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask3(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm1 {%k1} = xmm0[1,1,3,3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask3(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = xmm0[1,1,3,3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask4(<4 x float> %vec, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm1 {%k1} = xmm0[1,1,3,3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mask4(<4 x float> %vec, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = xmm0[1,1,3,3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_4xfloat_dup_high_mem(ptr %vp) {
; CHECK-LABEL: test_4xfloat_dup_high_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %res = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, i32 3>
  ret <4 x float> %res
}
define <4 x float> @test_masked_4xfloat_dup_high_mem_mask0(ptr %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask0(ptr %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask1(ptr %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask1(ptr %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask2(ptr %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask2(ptr %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask3(ptr %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask3(ptr %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask4(ptr %vp, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_masked_4xfloat_dup_high_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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_dup_high_mem_mask4(ptr %vp, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_4xfloat_dup_high_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} xmm0 {%k1} {z} = mem[1,1,3,3]
; CHECK-NEXT:    retq
  %vec = load <4 x float>, ptr %vp
  %shuf = shufflevector <4 x float> %vec, <4 x float> undef, <4 x i32> <i32 1, i32 1, i32 3, 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 <8 x float> @test_8xfloat_dup_high(<8 x float> %vec) {
; CHECK-LABEL: test_8xfloat_dup_high:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_dup_high_mask0(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm1 {%k1} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mask0(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mask1(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm1 {%k1} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mask1(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mask2(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm1 {%k1} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mask2(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mask3(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm1 {%k1} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mask3(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mask4(<8 x float> %vec, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm1 {%k1} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mask4(<8 x float> %vec, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = ymm0[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mem(ptr %vp) {
; CHECK-LABEL: test_8xfloat_dup_high_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %res = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7>
  ret <8 x float> %res
}
define <8 x float> @test_masked_8xfloat_dup_high_mem_mask0(ptr %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mem_mask0(ptr %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mem_mask1(ptr %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mem_mask1(ptr %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mem_mask2(ptr %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mem_mask2(ptr %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mem_mask3(ptr %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mem_mask3(ptr %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_masked_8xfloat_dup_high_mem_mask4(ptr %vp, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_masked_8xfloat_dup_high_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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_dup_high_mem_mask4(ptr %vp, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_8xfloat_dup_high_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} ymm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7]
; CHECK-NEXT:    retq
  %vec = load <8 x float>, ptr %vp
  %shuf = shufflevector <8 x float> %vec, <8 x float> undef, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, 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 <16 x float> @test_16xfloat_dup_high(<16 x float> %vec) {
; CHECK-LABEL: test_16xfloat_dup_high:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, i32 15>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_dup_high_mask0(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm1 {%k1} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; 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 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask0(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask1(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm1 {%k1} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; 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 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask1(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask2(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm1 {%k1} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; 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 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask2(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask3(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm1 {%k1} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; 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 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask3(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask4(<16 x float> %vec, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm1 {%k1} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; 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 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mask4(<16 x float> %vec, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = zmm0[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_16xfloat_dup_high_mem(ptr %vp) {
; CHECK-LABEL: test_16xfloat_dup_high_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %res = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, i32 15>
  ret <16 x float> %res
}
define <16 x float> @test_masked_16xfloat_dup_high_mem_mask0(ptr %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask0(ptr %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask1(ptr %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask1(ptr %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask2(ptr %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask2(ptr %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask3(ptr %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask3(ptr %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask4(ptr %vp, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_masked_16xfloat_dup_high_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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_dup_high_mem_mask4(ptr %vp, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_16xfloat_dup_high_mem_mask4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vmovshdup {{.*#+}} zmm0 {%k1} {z} = mem[1,1,3,3,5,5,7,7,9,9,11,11,13,13,15,15]
; CHECK-NEXT:    retq
  %vec = load <16 x float>, ptr %vp
  %shuf = shufflevector <16 x float> %vec, <16 x float> undef, <16 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7, i32 9, i32 9, i32 11, i32 11, i32 13, i32 13, i32 15, 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
}