llvm-for-llvmta/test/CodeGen/X86/avx512-shuffles/unpack.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

define <4 x float> @test_4xfloat_unpack_low_mask0(<4 x float> %vec1, <4 x float> %vec2) {
; CHECK-LABEL: test_4xfloat_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_low_mask0(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm2 {%k1} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mask0(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_masked_unpack_low_mask1(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm2 {%k1} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mask1(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_masked_unpack_low_mask2(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm2 {%k1} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mask2(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_unpack_low_mask3(<4 x float> %vec1, <4 x float> %vec2) {
; CHECK-LABEL: test_4xfloat_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_low_mask3(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm2 {%k1} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mask3(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_unpack_low_mem_mask0(<4 x float> %vec1, <4 x float>* %vec2p) {
; CHECK-LABEL: test_4xfloat_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_low_mem_mask0(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm1 {%k1} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mem_mask0(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_masked_unpack_low_mem_mask1(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm1 {%k1} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mem_mask1(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_masked_unpack_low_mem_mask2(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm1 {%k1} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mem_mask2(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_unpack_low_mem_mask3(<4 x float> %vec1, <4 x float>* %vec2p) {
; CHECK-LABEL: test_4xfloat_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_low_mem_mask3(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm1 {%k1} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_low_mem_mask3(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} xmm0 {%k1} {z} = xmm0[0],mem[0],xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 0, i32 4, i32 1, i32 5>
  %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_unpack_low_mask0(<8 x float> %vec1, <8 x float> %vec2) {
; CHECK-LABEL: test_8xfloat_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_low_mask0(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mask0(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_masked_unpack_low_mask1(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mask1(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_masked_unpack_low_mask2(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mask2(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_unpack_low_mask3(<8 x float> %vec1, <8 x float> %vec2) {
; CHECK-LABEL: test_8xfloat_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_low_mask3(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mask3(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[4],ymm1[4],ymm0[5],ymm1[5]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_unpack_low_mem_mask0(<8 x float> %vec1, <8 x float>* %vec2p) {
; CHECK-LABEL: test_8xfloat_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_low_mem_mask0(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mem_mask0(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_masked_unpack_low_mem_mask1(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mem_mask1(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_masked_unpack_low_mem_mask2(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mem_mask2(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_unpack_low_mem_mask3(<8 x float> %vec1, <8 x float>* %vec2p) {
; CHECK-LABEL: test_8xfloat_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_low_mem_mask3(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_low_mem_mask3(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[1],mem[1],ymm0[4],mem[4],ymm0[5],mem[5]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13>
  %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_unpack_low_mask0(<16 x float> %vec1, <16 x float> %vec2) {
; CHECK-LABEL: test_16xfloat_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_low_mask0(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mask0(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_masked_unpack_low_mask1(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mask1(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_masked_unpack_low_mask2(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mask2(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_unpack_low_mask3(<16 x float> %vec1, <16 x float> %vec2) {
; CHECK-LABEL: test_16xfloat_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_low_mask3(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mask3(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[1],zmm1[1],zmm0[4],zmm1[4],zmm0[5],zmm1[5],zmm0[8],zmm1[8],zmm0[9],zmm1[9],zmm0[12],zmm1[12],zmm0[13],zmm1[13]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_unpack_low_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p) {
; CHECK-LABEL: test_16xfloat_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_low_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_masked_unpack_low_mem_mask1(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mem_mask1(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_masked_unpack_low_mem_mask2(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mem_mask2(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_unpack_low_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p) {
; CHECK-LABEL: test_16xfloat_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_low_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_low_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklps {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[1],mem[1],zmm0[4],mem[4],zmm0[5],mem[5],zmm0[8],mem[8],zmm0[9],mem[9],zmm0[12],mem[12],zmm0[13],mem[13]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 0, i32 16, i32 1, i32 17, i32 4, i32 20, i32 5, i32 21, i32 8, i32 24, i32 9, i32 25, i32 12, i32 28, i32 13, i32 29>
  %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_unpack_low_mask0(<2 x double> %vec1, <2 x double> %vec2) {
; CHECK-LABEL: test_2xdouble_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vmovlhps {{.*#+}} xmm0 = xmm0[0],xmm1[0]
; CHECK-NEXT:    retq
  %res = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  ret <2 x double> %res
}
define <2 x double> @test_2xdouble_masked_unpack_low_mask0(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm2 {%k1} = xmm0[0],xmm1[0]
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_low_mask0(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm0 {%k1} {z} = xmm0[0],xmm1[0]
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %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_masked_unpack_low_mask1(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm2 {%k1} = xmm0[0],xmm1[0]
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_low_mask1(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm0 {%k1} {z} = xmm0[0],xmm1[0]
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %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_unpack_low_mem_mask0(<2 x double> %vec1, <2 x double>* %vec2p) {
; CHECK-LABEL: test_2xdouble_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm0[0],mem[0]
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %res = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  ret <2 x double> %res
}
define <2 x double> @test_2xdouble_masked_unpack_low_mem_mask0(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm1 {%k1} = xmm0[0],mem[0]
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_low_mem_mask0(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm0 {%k1} {z} = xmm0[0],mem[0]
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %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_masked_unpack_low_mem_mask1(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm1 {%k1} = xmm0[0],mem[0]
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_low_mem_mask1(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} xmm0 {%k1} {z} = xmm0[0],mem[0]
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 0, i32 2>
  %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_unpack_low_mask0(<4 x double> %vec1, <4 x double> %vec2) {
; CHECK-LABEL: test_4xdouble_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_low_mask0(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mask0(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_masked_unpack_low_mask1(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mask1(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_masked_unpack_low_mask2(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mask2(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_unpack_low_mask3(<4 x double> %vec1, <4 x double> %vec2) {
; CHECK-LABEL: test_4xdouble_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_low_mask3(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm2 {%k1} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mask3(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],ymm1[0],ymm0[2],ymm1[2]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_unpack_low_mem_mask0(<4 x double> %vec1, <4 x double>* %vec2p) {
; CHECK-LABEL: test_4xdouble_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_low_mem_mask0(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mem_mask0(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_masked_unpack_low_mem_mask1(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mem_mask1(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_masked_unpack_low_mem_mask2(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mem_mask2(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_unpack_low_mem_mask3(<4 x double> %vec1, <4 x double>* %vec2p) {
; CHECK-LABEL: test_4xdouble_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_low_mem_mask3(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm1 {%k1} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_low_mem_mask3(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} ymm0 {%k1} {z} = ymm0[0],mem[0],ymm0[2],mem[2]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %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_unpack_low_mask0(<8 x double> %vec1, <8 x double> %vec2) {
; CHECK-LABEL: test_8xdouble_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_low_mask0(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mask0(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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_masked_unpack_low_mask1(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mask1(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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_masked_unpack_low_mask2(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mask2(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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_unpack_low_mask3(<8 x double> %vec1, <8 x double> %vec2) {
; CHECK-LABEL: test_8xdouble_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_low_mask3(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm2 {%k1} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mask3(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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_unpack_low_mem_mask0(<8 x double> %vec1, <8 x double>* %vec2p) {
; CHECK-LABEL: test_8xdouble_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_low_mem_mask0(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mem_mask0(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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_masked_unpack_low_mem_mask1(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mem_mask1(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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_masked_unpack_low_mem_mask2(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mem_mask2(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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_unpack_low_mem_mask3(<8 x double> %vec1, <8 x double>* %vec2p) {
; CHECK-LABEL: test_8xdouble_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_low_mem_mask3(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm1 {%k1} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_low_mem_mask3(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_low_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpcklpd {{.*#+}} zmm0 {%k1} {z} = zmm0[0],mem[0],zmm0[2],mem[2],zmm0[4],mem[4],zmm0[6],mem[6]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14>
  %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 <4 x float> @test_4xfloat_unpack_high_mask0(<4 x float> %vec1, <4 x float> %vec2) {
; CHECK-LABEL: test_4xfloat_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_high_mask0(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm2 {%k1} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mask0(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_masked_unpack_high_mask1(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm2 {%k1} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mask1(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_masked_unpack_high_mask2(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm2 {%k1} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mask2(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_unpack_high_mask3(<4 x float> %vec1, <4 x float> %vec2) {
; CHECK-LABEL: test_4xfloat_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_high_mask3(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm2 {%k1} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    vmovaps %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mask3(<4 x float> %vec1, <4 x float> %vec2, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],xmm1[2],xmm0[3],xmm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_unpack_high_mem_mask0(<4 x float> %vec1, <4 x float>* %vec2p) {
; CHECK-LABEL: test_4xfloat_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_high_mem_mask0(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm1 {%k1} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mem_mask0(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_masked_unpack_high_mem_mask1(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm1 {%k1} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mem_mask1(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_masked_unpack_high_mem_mask2(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm1 {%k1} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mem_mask2(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_unpack_high_mem_mask3(<4 x float> %vec1, <4 x float>* %vec2p) {
; CHECK-LABEL: test_4xfloat_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %res = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  ret <4 x float> %res
}
define <4 x float> @test_4xfloat_masked_unpack_high_mem_mask3(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %vec3, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm1 {%k1} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %vec3
  ret <4 x float> %res
}

define <4 x float> @test_4xfloat_zero_masked_unpack_high_mem_mask3(<4 x float> %vec1, <4 x float>* %vec2p, <4 x float> %mask) {
; CHECK-LABEL: test_4xfloat_zero_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} xmm0 {%k1} {z} = xmm0[2],mem[2],xmm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x float>, <4 x float>* %vec2p
  %shuf = shufflevector <4 x float> %vec1, <4 x float> %vec2, <4 x i32> <i32 2, i32 6, i32 3, i32 7>
  %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_unpack_high_mask0(<8 x float> %vec1, <8 x float> %vec2) {
; CHECK-LABEL: test_8xfloat_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_high_mask0(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm2 {%k1} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mask0(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_masked_unpack_high_mask1(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm2 {%k1} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mask1(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_masked_unpack_high_mask2(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm2 {%k1} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mask2(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_unpack_high_mask3(<8 x float> %vec1, <8 x float> %vec2) {
; CHECK-LABEL: test_8xfloat_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_high_mask3(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm2 {%k1} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    vmovaps %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mask3(<8 x float> %vec1, <8 x float> %vec2, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[6],ymm1[6],ymm0[7],ymm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_unpack_high_mem_mask0(<8 x float> %vec1, <8 x float>* %vec2p) {
; CHECK-LABEL: test_8xfloat_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_high_mem_mask0(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm1 {%k1} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mem_mask0(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_masked_unpack_high_mem_mask1(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm1 {%k1} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mem_mask1(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_masked_unpack_high_mem_mask2(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm1 {%k1} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mem_mask2(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_unpack_high_mem_mask3(<8 x float> %vec1, <8 x float>* %vec2p) {
; CHECK-LABEL: test_8xfloat_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %res = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  ret <8 x float> %res
}
define <8 x float> @test_8xfloat_masked_unpack_high_mem_mask3(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %vec3, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm1 {%k1} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %vec3
  ret <8 x float> %res
}

define <8 x float> @test_8xfloat_zero_masked_unpack_high_mem_mask3(<8 x float> %vec1, <8 x float>* %vec2p, <8 x float> %mask) {
; CHECK-LABEL: test_8xfloat_zero_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} ymm0 {%k1} {z} = ymm0[2],mem[2],ymm0[3],mem[3],ymm0[6],mem[6],ymm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x float>, <8 x float>* %vec2p
  %shuf = shufflevector <8 x float> %vec1, <8 x float> %vec2, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15>
  %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_unpack_high_mask0(<16 x float> %vec1, <16 x float> %vec2) {
; CHECK-LABEL: test_16xfloat_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_high_mask0(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm2 {%k1} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mask0(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_masked_unpack_high_mask1(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm2 {%k1} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mask1(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_masked_unpack_high_mask2(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm2 {%k1} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mask2(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_unpack_high_mask3(<16 x float> %vec1, <16 x float> %vec2) {
; CHECK-LABEL: test_16xfloat_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    retq
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_high_mask3(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqps %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm2 {%k1} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    vmovaps %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mask3(<16 x float> %vec1, <16 x float> %vec2, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],zmm1[2],zmm0[3],zmm1[3],zmm0[6],zmm1[6],zmm0[7],zmm1[7],zmm0[10],zmm1[10],zmm0[11],zmm1[11],zmm0[14],zmm1[14],zmm0[15],zmm1[15]
; CHECK-NEXT:    retq
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_unpack_high_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p) {
; CHECK-LABEL: test_16xfloat_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_high_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm1 {%k1} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mem_mask0(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_masked_unpack_high_mem_mask1(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm1 {%k1} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mem_mask1(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_masked_unpack_high_mem_mask2(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm1 {%k1} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mem_mask2(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_unpack_high_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p) {
; CHECK-LABEL: test_16xfloat_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %res = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  ret <16 x float> %res
}
define <16 x float> @test_16xfloat_masked_unpack_high_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %vec3, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm1 {%k1} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %vec3
  ret <16 x float> %res
}

define <16 x float> @test_16xfloat_zero_masked_unpack_high_mem_mask3(<16 x float> %vec1, <16 x float>* %vec2p, <16 x float> %mask) {
; CHECK-LABEL: test_16xfloat_zero_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhps {{.*#+}} zmm0 {%k1} {z} = zmm0[2],mem[2],zmm0[3],mem[3],zmm0[6],mem[6],zmm0[7],mem[7],zmm0[10],mem[10],zmm0[11],mem[11],zmm0[14],mem[14],zmm0[15],mem[15]
; CHECK-NEXT:    retq
  %vec2 = load <16 x float>, <16 x float>* %vec2p
  %shuf = shufflevector <16 x float> %vec1, <16 x float> %vec2, <16 x i32> <i32 2, i32 18, i32 3, i32 19, i32 6, i32 22, i32 7, i32 23, i32 10, i32 26, i32 11, i32 27, i32 14, i32 30, i32 15, i32 31>
  %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_unpack_high_mask0(<2 x double> %vec1, <2 x double> %vec2) {
; CHECK-LABEL: test_2xdouble_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %res = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  ret <2 x double> %res
}
define <2 x double> @test_2xdouble_masked_unpack_high_mask0(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm2 {%k1} = xmm0[1],xmm1[1]
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_high_mask0(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm0 {%k1} {z} = xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %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_masked_unpack_high_mask1(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %xmm4, %xmm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm2 {%k1} = xmm0[1],xmm1[1]
; CHECK-NEXT:    vmovapd %xmm2, %xmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_high_mask1(<2 x double> %vec1, <2 x double> %vec2, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm0 {%k1} {z} = xmm0[1],xmm1[1]
; CHECK-NEXT:    retq
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %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_unpack_high_mem_mask0(<2 x double> %vec1, <2 x double>* %vec2p) {
; CHECK-LABEL: test_2xdouble_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm0 = xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %res = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  ret <2 x double> %res
}
define <2 x double> @test_2xdouble_masked_unpack_high_mem_mask0(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm1 {%k1} = xmm0[1],mem[1]
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_high_mem_mask0(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm0 {%k1} {z} = xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %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_masked_unpack_high_mem_mask1(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %vec3, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %xmm3, %xmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm1 {%k1} = xmm0[1],mem[1]
; CHECK-NEXT:    vmovapd %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %cmp = fcmp oeq <2 x double> %mask, zeroinitializer
  %res = select <2 x i1> %cmp, <2 x double> %shuf, <2 x double> %vec3
  ret <2 x double> %res
}

define <2 x double> @test_2xdouble_zero_masked_unpack_high_mem_mask1(<2 x double> %vec1, <2 x double>* %vec2p, <2 x double> %mask) {
; CHECK-LABEL: test_2xdouble_zero_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %xmm2, %xmm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} xmm0 {%k1} {z} = xmm0[1],mem[1]
; CHECK-NEXT:    retq
  %vec2 = load <2 x double>, <2 x double>* %vec2p
  %shuf = shufflevector <2 x double> %vec1, <2 x double> %vec2, <2 x i32> <i32 1, i32 3>
  %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_unpack_high_mask0(<4 x double> %vec1, <4 x double> %vec2) {
; CHECK-LABEL: test_4xdouble_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_high_mask0(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm2 {%k1} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mask0(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_masked_unpack_high_mask1(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm2 {%k1} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mask1(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_masked_unpack_high_mask2(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm2 {%k1} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mask2(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_unpack_high_mask3(<4 x double> %vec1, <4 x double> %vec2) {
; CHECK-LABEL: test_4xdouble_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    retq
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_high_mask3(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %ymm4, %ymm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm2 {%k1} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    vmovapd %ymm2, %ymm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mask3(<4 x double> %vec1, <4 x double> %vec2, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],ymm1[1],ymm0[3],ymm1[3]
; CHECK-NEXT:    retq
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_unpack_high_mem_mask0(<4 x double> %vec1, <4 x double>* %vec2p) {
; CHECK-LABEL: test_4xdouble_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_high_mem_mask0(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm1 {%k1} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mem_mask0(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_masked_unpack_high_mem_mask1(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm1 {%k1} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mem_mask1(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_masked_unpack_high_mem_mask2(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm1 {%k1} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mem_mask2(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_unpack_high_mem_mask3(<4 x double> %vec1, <4 x double>* %vec2p) {
; CHECK-LABEL: test_4xdouble_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %res = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  ret <4 x double> %res
}
define <4 x double> @test_4xdouble_masked_unpack_high_mem_mask3(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %vec3, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm1 {%k1} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %vec3
  ret <4 x double> %res
}

define <4 x double> @test_4xdouble_zero_masked_unpack_high_mem_mask3(<4 x double> %vec1, <4 x double>* %vec2p, <4 x double> %mask) {
; CHECK-LABEL: test_4xdouble_zero_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} ymm0 {%k1} {z} = ymm0[1],mem[1],ymm0[3],mem[3]
; CHECK-NEXT:    retq
  %vec2 = load <4 x double>, <4 x double>* %vec2p
  %shuf = shufflevector <4 x double> %vec1, <4 x double> %vec2, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %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_unpack_high_mask0(<8 x double> %vec1, <8 x double> %vec2) {
; CHECK-LABEL: test_8xdouble_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_high_mask0(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm2 {%k1} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mask0(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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_masked_unpack_high_mask1(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm2 {%k1} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mask1(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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_masked_unpack_high_mask2(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm2 {%k1} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mask2(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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_unpack_high_mask3(<8 x double> %vec1, <8 x double> %vec2) {
; CHECK-LABEL: test_8xdouble_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    retq
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_high_mask3(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm4, %xmm4, %xmm4
; CHECK-NEXT:    vcmpeqpd %zmm4, %zmm3, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm2 {%k1} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    vmovapd %zmm2, %zmm0
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mask3(<8 x double> %vec1, <8 x double> %vec2, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7]
; CHECK-NEXT:    retq
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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_unpack_high_mem_mask0(<8 x double> %vec1, <8 x double>* %vec2p) {
; CHECK-LABEL: test_8xdouble_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_high_mem_mask0(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm1 {%k1} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mem_mask0(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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_masked_unpack_high_mem_mask1(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm1 {%k1} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mem_mask1(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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_masked_unpack_high_mem_mask2(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm1 {%k1} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mem_mask2(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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_unpack_high_mem_mask3(<8 x double> %vec1, <8 x double>* %vec2p) {
; CHECK-LABEL: test_8xdouble_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %res = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  ret <8 x double> %res
}
define <8 x double> @test_8xdouble_masked_unpack_high_mem_mask3(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %vec3, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm1 {%k1} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %vec3
  ret <8 x double> %res
}

define <8 x double> @test_8xdouble_zero_masked_unpack_high_mem_mask3(<8 x double> %vec1, <8 x double>* %vec2p, <8 x double> %mask) {
; CHECK-LABEL: test_8xdouble_zero_masked_unpack_high_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vunpckhpd {{.*#+}} zmm0 {%k1} {z} = zmm0[1],mem[1],zmm0[3],mem[3],zmm0[5],mem[5],zmm0[7],mem[7]
; CHECK-NEXT:    retq
  %vec2 = load <8 x double>, <8 x double>* %vec2p
  %shuf = shufflevector <8 x double> %vec1, <8 x double> %vec2, <8 x i32> <i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15>
  %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
}