broadcast-scalar-fp.ll | Explore in Territory

; 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 double> @test_double_to_4(double %s) {
; CHECK-LABEL: test_double_to_4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %res = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mask0(double %s, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mask0(double %s, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mask1(double %s, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mask1(double %s, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mask2(double %s, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mask2(double %s, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mask3(double %s, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovapd %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mask3(double %s, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %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_double_to_8(double %s) {
; CHECK-LABEL: test_double_to_8:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %res = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mask0(double %s, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mask0(double %s, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mask1(double %s, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mask1(double %s, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mask2(double %s, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mask2(double %s, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mask3(double %s, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqpd %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovapd %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mask3(double %s, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %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_float_to_4(float %s) {
; CHECK-LABEL: test_float_to_4:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %res = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mask0(float %s, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mask0(float %s, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mask1(float %s, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mask1(float %s, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mask2(float %s, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mask2(float %s, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mask3(float %s, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %xmm3, %xmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm1 {%k1}
; CHECK-NEXT:    vmovaps %xmm1, %xmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mask3(float %s, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <8 x float> @test_float_to_8(float %s) {
; CHECK-LABEL: test_float_to_8:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %res = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mask0(float %s, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mask0(float %s, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mask1(float %s, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mask1(float %s, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mask2(float %s, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mask2(float %s, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mask3(float %s, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %ymm3, %ymm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm1 {%k1}
; CHECK-NEXT:    vmovaps %ymm1, %ymm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mask3(float %s, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %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_float_to_16(float %s) {
; CHECK-LABEL: test_float_to_16:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %res = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mask0(float %s, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mask0(float %s, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mask1(float %s, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mask1(float %s, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mask2(float %s, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mask2(float %s, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mask3(float %s, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm3, %xmm3, %xmm3
; CHECK-NEXT:    vcmpeqps %zmm3, %zmm2, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm1 {%k1}
; CHECK-NEXT:    vmovaps %zmm1, %zmm0
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mask3(float %s, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss %xmm0, %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %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 <4 x double> @test_double_to_4_mem(ptr %p) {
; CHECK-LABEL: test_double_to_4_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %res = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mem_mask0(ptr %p, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mem_mask0(ptr %p, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mem_mask1(ptr %p, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mem_mask1(ptr %p, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mem_mask2(ptr %p, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mem_mask2(ptr %p, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> zeroinitializer
  ret <4 x double> %res
}
define <4 x double> @test_masked_double_to_4_mem_mask3(ptr %p, <4 x double> %default, <4 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_4_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x double> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x double> %shuf, <4 x double> %default
  ret <4 x double> %res
}

define <4 x double> @test_masked_z_double_to_4_mem_mask3(ptr %p, <4 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_4_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %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_double_to_8_mem(ptr %p) {
; CHECK-LABEL: test_double_to_8_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %res = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mem_mask0(ptr %p, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mem_mask0(ptr %p, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mem_mask1(ptr %p, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mem_mask1(ptr %p, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mem_mask2(ptr %p, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mem_mask2(ptr %p, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> zeroinitializer
  ret <8 x double> %res
}
define <8 x double> @test_masked_double_to_8_mem_mask3(ptr %p, <8 x double> %default, <8 x double> %mask) {
; CHECK-LABEL: test_masked_double_to_8_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqpd %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x double> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x double> %shuf, <8 x double> %default
  ret <8 x double> %res
}

define <8 x double> @test_masked_z_double_to_8_mem_mask3(ptr %p, <8 x double> %mask) {
; CHECK-LABEL: test_masked_z_double_to_8_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorpd %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqpd %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastsd (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load double, ptr %p
  %vec = insertelement <2 x double> undef, double %s, i32 0
  %shuf = shufflevector <2 x double> %vec, <2 x double> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %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_float_to_4_mem(ptr %p) {
; CHECK-LABEL: test_float_to_4_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %res = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mem_mask0(ptr %p, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mem_mask0(ptr %p, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mem_mask1(ptr %p, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mem_mask1(ptr %p, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mem_mask2(ptr %p, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mem_mask2(ptr %p, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <4 x float> @test_masked_float_to_4_mem_mask3(ptr %p, <4 x float> %default, <4 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_4_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %xmm2, %xmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> %default
  ret <4 x float> %res
}

define <4 x float> @test_masked_z_float_to_4_mem_mask3(ptr %p, <4 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_4_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %xmm1, %xmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %xmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <4 x float> %mask, zeroinitializer
  %res = select <4 x i1> %cmp, <4 x float> %shuf, <4 x float> zeroinitializer
  ret <4 x float> %res
}
define <8 x float> @test_float_to_8_mem(ptr %p) {
; CHECK-LABEL: test_float_to_8_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %res = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mem_mask0(ptr %p, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mem_mask0(ptr %p, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mem_mask1(ptr %p, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mem_mask1(ptr %p, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mem_mask2(ptr %p, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mem_mask2(ptr %p, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> zeroinitializer
  ret <8 x float> %res
}
define <8 x float> @test_masked_float_to_8_mem_mask3(ptr %p, <8 x float> %default, <8 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_8_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %ymm2, %ymm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <8 x float> %mask, zeroinitializer
  %res = select <8 x i1> %cmp, <8 x float> %shuf, <8 x float> %default
  ret <8 x float> %res
}

define <8 x float> @test_masked_z_float_to_8_mem_mask3(ptr %p, <8 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_8_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %ymm1, %ymm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %ymm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %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_float_to_16_mem(ptr %p) {
; CHECK-LABEL: test_float_to_16_mem:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %res = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mem_mask0(ptr %p, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mem_mask0(ptr %p, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mem_mask0:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mem_mask1(ptr %p, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mem_mask1(ptr %p, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mem_mask1:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mem_mask2(ptr %p, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mem_mask2(ptr %p, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mem_mask2:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> zeroinitializer
  ret <16 x float> %res
}
define <16 x float> @test_masked_float_to_16_mem_mask3(ptr %p, <16 x float> %default, <16 x float> %mask) {
; CHECK-LABEL: test_masked_float_to_16_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm2, %xmm2, %xmm2
; CHECK-NEXT:    vcmpeqps %zmm2, %zmm1, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %cmp = fcmp oeq <16 x float> %mask, zeroinitializer
  %res = select <16 x i1> %cmp, <16 x float> %shuf, <16 x float> %default
  ret <16 x float> %res
}

define <16 x float> @test_masked_z_float_to_16_mem_mask3(ptr %p, <16 x float> %mask) {
; CHECK-LABEL: test_masked_z_float_to_16_mem_mask3:
; CHECK:       # %bb.0:
; CHECK-NEXT:    vxorps %xmm1, %xmm1, %xmm1
; CHECK-NEXT:    vcmpeqps %zmm1, %zmm0, %k1
; CHECK-NEXT:    vbroadcastss (%rdi), %zmm0 {%k1} {z}
; CHECK-NEXT:    retq
  %s = load float, ptr %p
  %vec = insertelement <2 x float> undef, float %s, i32 0
  %shuf = shufflevector <2 x float> %vec, <2 x float> undef, <16 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0>
  %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
}
llvm/llvm/test/CodeGen/X86/avx512-shuffles/broadcast-scalar-fp.ll
