; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx,+f16c -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX1
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX2
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX2
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx2,+f16c,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX2
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX512
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX512
; RUN: llc < %s -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl,+fast-variable-perlane-shuffle -verify-machineinstrs | FileCheck %s --check-prefixes=ALL,AVX512

;
; Half to Float
;

define float @cvt_i16_to_f32(i16 %a0) nounwind {
; ALL-LABEL: cvt_i16_to_f32:
; ALL:       # %bb.0:
; ALL-NEXT:    movzwl %di, %eax
; ALL-NEXT:    vmovd %eax, %xmm0
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = bitcast i16 %a0 to half
  %2 = fpext half %1 to float
  ret float %2
}

define <4 x float> @cvt_4i16_to_4f32(<4 x i16> %a0) nounwind {
; ALL-LABEL: cvt_4i16_to_4f32:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = bitcast <4 x i16> %a0 to <4 x half>
  %2 = fpext <4 x half> %1 to <4 x float>
  ret <4 x float> %2
}

define <4 x float> @cvt_8i16_to_4f32(<8 x i16> %a0) nounwind {
; ALL-LABEL: cvt_8i16_to_4f32:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %2 = bitcast <4 x i16> %1 to <4 x half>
  %3 = fpext <4 x half> %2 to <4 x float>
  ret <4 x float> %3
}

define <8 x float> @cvt_8i16_to_8f32(<8 x i16> %a0) nounwind {
; ALL-LABEL: cvt_8i16_to_8f32:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %ymm0
; ALL-NEXT:    retq
  %1 = bitcast <8 x i16> %a0 to <8 x half>
  %2 = fpext <8 x half> %1 to <8 x float>
  ret <8 x float> %2
}

define <16 x float> @cvt_16i16_to_16f32(<16 x i16> %a0) nounwind {
; AVX1-LABEL: cvt_16i16_to_16f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtph2ps %xmm0, %ymm2
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vcvtph2ps %xmm0, %ymm1
; AVX1-NEXT:    vmovaps %ymm2, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: cvt_16i16_to_16f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtph2ps %xmm0, %ymm2
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vcvtph2ps %xmm0, %ymm1
; AVX2-NEXT:    vmovaps %ymm2, %ymm0
; AVX2-NEXT:    retq
;
; AVX512-LABEL: cvt_16i16_to_16f32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtph2ps %ymm0, %zmm0
; AVX512-NEXT:    retq
  %1 = bitcast <16 x i16> %a0 to <16 x half>
  %2 = fpext <16 x half> %1 to <16 x float>
  ret <16 x float> %2
}

define <2 x float> @cvt_2i16_to_2f32_constrained(<2 x i16> %a0) nounwind strictfp {
; ALL-LABEL: cvt_2i16_to_2f32_constrained:
; ALL:       # %bb.0:
; ALL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = bitcast <2 x i16> %a0 to <2 x half>
  %2 = call <2 x float> @llvm.experimental.constrained.fpext.v2f32.v2f16(<2 x half> %1, metadata !"fpexcept.strict") strictfp
  ret <2 x float> %2
}
declare <2 x float> @llvm.experimental.constrained.fpext.v2f32.v2f16(<2 x half>, metadata) strictfp

define <4 x float> @cvt_4i16_to_4f32_constrained(<4 x i16> %a0) nounwind strictfp {
; ALL-LABEL: cvt_4i16_to_4f32_constrained:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = bitcast <4 x i16> %a0 to <4 x half>
  %2 = call <4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half> %1, metadata !"fpexcept.strict") strictfp
  ret <4 x float> %2
}
declare <4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half>, metadata) strictfp

define <8 x float> @cvt_8i16_to_8f32_constrained(<8 x i16> %a0) nounwind strictfp {
; ALL-LABEL: cvt_8i16_to_8f32_constrained:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %ymm0
; ALL-NEXT:    retq
  %1 = bitcast <8 x i16> %a0 to <8 x half>
  %2 = call <8 x float> @llvm.experimental.constrained.fpext.v8f32.v8f16(<8 x half> %1, metadata !"fpexcept.strict") strictfp
  ret <8 x float> %2
}
declare <8 x float> @llvm.experimental.constrained.fpext.v8f32.v8f16(<8 x half>, metadata) strictfp

define <16 x float> @cvt_16i16_to_16f32_constrained(<16 x i16> %a0) nounwind strictfp {
; AVX1-LABEL: cvt_16i16_to_16f32_constrained:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vcvtph2ps %xmm1, %ymm1
; AVX1-NEXT:    vcvtph2ps %xmm0, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: cvt_16i16_to_16f32_constrained:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vcvtph2ps %xmm1, %ymm1
; AVX2-NEXT:    vcvtph2ps %xmm0, %ymm0
; AVX2-NEXT:    retq
;
; AVX512-LABEL: cvt_16i16_to_16f32_constrained:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtph2ps %ymm0, %zmm0
; AVX512-NEXT:    retq
  %1 = bitcast <16 x i16> %a0 to <16 x half>
  %2 = call <16 x float> @llvm.experimental.constrained.fpext.v16f32.v16f16(<16 x half> %1, metadata !"fpexcept.strict") strictfp
  ret <16 x float> %2
}
declare <16 x float> @llvm.experimental.constrained.fpext.v16f32.v16f16(<16 x half>, metadata) strictfp

;
; Half to Float (Load)
;

define float @load_cvt_i16_to_f32(i16* %a0) nounwind {
; ALL-LABEL: load_cvt_i16_to_f32:
; ALL:       # %bb.0:
; ALL-NEXT:    movzwl (%rdi), %eax
; ALL-NEXT:    vmovd %eax, %xmm0
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = load i16, i16* %a0
  %2 = bitcast i16 %1 to half
  %3 = fpext half %2 to float
  ret float %3
}

define <4 x float> @load_cvt_4i16_to_4f32(<4 x i16>* %a0) nounwind {
; ALL-LABEL: load_cvt_4i16_to_4f32:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps (%rdi), %xmm0
; ALL-NEXT:    retq
  %1 = load <4 x i16>, <4 x i16>* %a0
  %2 = bitcast <4 x i16> %1 to <4 x half>
  %3 = fpext <4 x half> %2 to <4 x float>
  ret <4 x float> %3
}

define <4 x float> @load_cvt_8i16_to_4f32(<8 x i16>* %a0) nounwind {
; ALL-LABEL: load_cvt_8i16_to_4f32:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps (%rdi), %xmm0
; ALL-NEXT:    retq
  %1 = load <8 x i16>, <8 x i16>* %a0
  %2 = shufflevector <8 x i16> %1, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %3 = bitcast <4 x i16> %2 to <4 x half>
  %4 = fpext <4 x half> %3 to <4 x float>
  ret <4 x float> %4
}

define <8 x float> @load_cvt_8i16_to_8f32(<8 x i16>* %a0) nounwind {
; ALL-LABEL: load_cvt_8i16_to_8f32:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps (%rdi), %ymm0
; ALL-NEXT:    retq
  %1 = load <8 x i16>, <8 x i16>* %a0
  %2 = bitcast <8 x i16> %1 to <8 x half>
  %3 = fpext <8 x half> %2 to <8 x float>
  ret <8 x float> %3
}

define <16 x float> @load_cvt_16i16_to_16f32(<16 x i16>* %a0) nounwind {
; AVX1-LABEL: load_cvt_16i16_to_16f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtph2ps (%rdi), %ymm0
; AVX1-NEXT:    vcvtph2ps 16(%rdi), %ymm1
; AVX1-NEXT:    retq
;
; AVX2-LABEL: load_cvt_16i16_to_16f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtph2ps (%rdi), %ymm0
; AVX2-NEXT:    vcvtph2ps 16(%rdi), %ymm1
; AVX2-NEXT:    retq
;
; AVX512-LABEL: load_cvt_16i16_to_16f32:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtph2ps (%rdi), %zmm0
; AVX512-NEXT:    retq
  %1 = load <16 x i16>, <16 x i16>* %a0
  %2 = bitcast <16 x i16> %1 to <16 x half>
  %3 = fpext <16 x half> %2 to <16 x float>
  ret <16 x float> %3
}

define <4 x float> @load_cvt_4i16_to_4f32_constrained(<4 x i16>* %a0) nounwind strictfp {
; ALL-LABEL: load_cvt_4i16_to_4f32_constrained:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps (%rdi), %xmm0
; ALL-NEXT:    retq
  %1 = load <4 x i16>, <4 x i16>* %a0
  %2 = bitcast <4 x i16> %1 to <4 x half>
  %3 = call <4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half> %2, metadata !"fpexcept.strict") strictfp
  ret <4 x float> %3
}

define <4 x float> @load_cvt_8i16_to_4f32_constrained(<8 x i16>* %a0) nounwind strictfp {
; ALL-LABEL: load_cvt_8i16_to_4f32_constrained:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps (%rdi), %xmm0
; ALL-NEXT:    retq
  %1 = load <8 x i16>, <8 x i16>* %a0
  %2 = shufflevector <8 x i16> %1, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %3 = bitcast <4 x i16> %2 to <4 x half>
  %4 = call <4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half> %3, metadata !"fpexcept.strict") strictfp
  ret <4 x float> %4
}

;
; Half to Double
;

define double @cvt_i16_to_f64(i16 %a0) nounwind {
; ALL-LABEL: cvt_i16_to_f64:
; ALL:       # %bb.0:
; ALL-NEXT:    movzwl %di, %eax
; ALL-NEXT:    vmovd %eax, %xmm0
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = bitcast i16 %a0 to half
  %2 = fpext half %1 to double
  ret double %2
}

define <2 x double> @cvt_2i16_to_2f64(<2 x i16> %a0) nounwind {
; ALL-LABEL: cvt_2i16_to_2f64:
; ALL:       # %bb.0:
; ALL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = bitcast <2 x i16> %a0 to <2 x half>
  %2 = fpext <2 x half> %1 to <2 x double>
  ret <2 x double> %2
}

define <4 x double> @cvt_4i16_to_4f64(<4 x i16> %a0) nounwind {
; ALL-LABEL: cvt_4i16_to_4f64:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %ymm0
; ALL-NEXT:    retq
  %1 = bitcast <4 x i16> %a0 to <4 x half>
  %2 = fpext <4 x half> %1 to <4 x double>
  ret <4 x double> %2
}

define <2 x double> @cvt_8i16_to_2f64(<8 x i16> %a0) nounwind {
; ALL-LABEL: cvt_8i16_to_2f64:
; ALL:       # %bb.0:
; ALL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
  %2 = bitcast <2 x i16> %1 to <2 x half>
  %3 = fpext <2 x half> %2 to <2 x double>
  ret <2 x double> %3
}

define <4 x double> @cvt_8i16_to_4f64(<8 x i16> %a0) nounwind {
; ALL-LABEL: cvt_8i16_to_4f64:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %ymm0
; ALL-NEXT:    retq
  %1 = shufflevector <8 x i16> %a0, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %2 = bitcast <4 x i16> %1 to <4 x half>
  %3 = fpext <4 x half> %2 to <4 x double>
  ret <4 x double> %3
}

define <8 x double> @cvt_8i16_to_8f64(<8 x i16> %a0) nounwind {
; AVX1-LABEL: cvt_8i16_to_8f64:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtph2ps %xmm0, %ymm1
; AVX1-NEXT:    vcvtps2pd %xmm1, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
; AVX1-NEXT:    vcvtps2pd %xmm1, %ymm1
; AVX1-NEXT:    retq
;
; AVX2-LABEL: cvt_8i16_to_8f64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtph2ps %xmm0, %ymm1
; AVX2-NEXT:    vcvtps2pd %xmm1, %ymm0
; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm1
; AVX2-NEXT:    vcvtps2pd %xmm1, %ymm1
; AVX2-NEXT:    retq
;
; AVX512-LABEL: cvt_8i16_to_8f64:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtph2ps %xmm0, %ymm0
; AVX512-NEXT:    vcvtps2pd %ymm0, %zmm0
; AVX512-NEXT:    retq
  %1 = bitcast <8 x i16> %a0 to <8 x half>
  %2 = fpext <8 x half> %1 to <8 x double>
  ret <8 x double> %2
}

define <2 x double> @cvt_2i16_to_2f64_constrained(<2 x i16> %a0) nounwind strictfp {
; ALL-LABEL: cvt_2i16_to_2f64_constrained:
; ALL:       # %bb.0:
; ALL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = bitcast <2 x i16> %a0 to <2 x half>
  %2 = call <2 x double> @llvm.experimental.constrained.fpext.v2f64.v2f16(<2 x half> %1, metadata !"fpexcept.strict") strictfp
  ret <2 x double> %2
}
declare <2 x double> @llvm.experimental.constrained.fpext.v2f64.v2f16(<2 x half>, metadata) strictfp

define <4 x double> @cvt_4i16_to_4f64_constrained(<4 x i16> %a0) nounwind strictfp {
; ALL-LABEL: cvt_4i16_to_4f64_constrained:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %ymm0
; ALL-NEXT:    retq
  %1 = bitcast <4 x i16> %a0 to <4 x half>
  %2 = call <4 x double> @llvm.experimental.constrained.fpext.v4f64.v4f16(<4 x half> %1, metadata !"fpexcept.strict") strictfp
  ret <4 x double> %2
}
declare <4 x double> @llvm.experimental.constrained.fpext.v4f64.v4f16(<4 x half>, metadata) strictfp

define <8 x double> @cvt_8i16_to_8f64_constrained(<8 x i16> %a0) nounwind strictfp {
; AVX1-LABEL: cvt_8i16_to_8f64_constrained:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtph2ps %xmm0, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vcvtps2pd %xmm1, %ymm1
; AVX1-NEXT:    vcvtps2pd %xmm0, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: cvt_8i16_to_8f64_constrained:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtph2ps %xmm0, %ymm0
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vcvtps2pd %xmm1, %ymm1
; AVX2-NEXT:    vcvtps2pd %xmm0, %ymm0
; AVX2-NEXT:    retq
;
; AVX512-LABEL: cvt_8i16_to_8f64_constrained:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtph2ps %xmm0, %ymm0
; AVX512-NEXT:    vcvtps2pd %ymm0, %zmm0
; AVX512-NEXT:    retq
  %1 = bitcast <8 x i16> %a0 to <8 x half>
  %2 = call <8 x double> @llvm.experimental.constrained.fpext.v8f64.v8f16(<8 x half> %1, metadata !"fpexcept.strict") strictfp
  ret <8 x double> %2
}
declare <8 x double> @llvm.experimental.constrained.fpext.v8f64.v8f16(<8 x half>, metadata) strictfp

;
; Half to Double (Load)
;

define double @load_cvt_i16_to_f64(i16* %a0) nounwind {
; ALL-LABEL: load_cvt_i16_to_f64:
; ALL:       # %bb.0:
; ALL-NEXT:    movzwl (%rdi), %eax
; ALL-NEXT:    vmovd %eax, %xmm0
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtss2sd %xmm0, %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = load i16, i16* %a0
  %2 = bitcast i16 %1 to half
  %3 = fpext half %2 to double
  ret double %3
}

define <2 x double> @load_cvt_2i16_to_2f64(<2 x i16>* %a0) nounwind {
; ALL-LABEL: load_cvt_2i16_to_2f64:
; ALL:       # %bb.0:
; ALL-NEXT:    vmovd {{.*#+}} xmm0 = mem[0],zero,zero,zero
; ALL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
; ALL-NEXT:    vcvtph2ps %xmm0, %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = load <2 x i16>, <2 x i16>* %a0
  %2 = bitcast <2 x i16> %1 to <2 x half>
  %3 = fpext <2 x half> %2 to <2 x double>
  ret <2 x double> %3
}

define <4 x double> @load_cvt_4i16_to_4f64(<4 x i16>* %a0) nounwind {
; ALL-LABEL: load_cvt_4i16_to_4f64:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps (%rdi), %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %ymm0
; ALL-NEXT:    retq
  %1 = load <4 x i16>, <4 x i16>* %a0
  %2 = bitcast <4 x i16> %1 to <4 x half>
  %3 = fpext <4 x half> %2 to <4 x double>
  ret <4 x double> %3
}

define <4 x double> @load_cvt_8i16_to_4f64(<8 x i16>* %a0) nounwind {
; ALL-LABEL: load_cvt_8i16_to_4f64:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtph2ps (%rdi), %xmm0
; ALL-NEXT:    vcvtps2pd %xmm0, %ymm0
; ALL-NEXT:    retq
  %1 = load <8 x i16>, <8 x i16>* %a0
  %2 = shufflevector <8 x i16> %1, <8 x i16> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %3 = bitcast <4 x i16> %2 to <4 x half>
  %4 = fpext <4 x half> %3 to <4 x double>
  ret <4 x double> %4
}

define <8 x double> @load_cvt_8i16_to_8f64(<8 x i16>* %a0) nounwind {
; AVX1-LABEL: load_cvt_8i16_to_8f64:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtph2ps (%rdi), %ymm1
; AVX1-NEXT:    vcvtps2pd %xmm1, %ymm0
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
; AVX1-NEXT:    vcvtps2pd %xmm1, %ymm1
; AVX1-NEXT:    retq
;
; AVX2-LABEL: load_cvt_8i16_to_8f64:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtph2ps (%rdi), %ymm1
; AVX2-NEXT:    vcvtps2pd %xmm1, %ymm0
; AVX2-NEXT:    vextractf128 $1, %ymm1, %xmm1
; AVX2-NEXT:    vcvtps2pd %xmm1, %ymm1
; AVX2-NEXT:    retq
;
; AVX512-LABEL: load_cvt_8i16_to_8f64:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtph2ps (%rdi), %ymm0
; AVX512-NEXT:    vcvtps2pd %ymm0, %zmm0
; AVX512-NEXT:    retq
  %1 = load <8 x i16>, <8 x i16>* %a0
  %2 = bitcast <8 x i16> %1 to <8 x half>
  %3 = fpext <8 x half> %2 to <8 x double>
  ret <8 x double> %3
}

;
; Float to Half
;

define i16 @cvt_f32_to_i16(float %a0) nounwind {
; ALL-LABEL: cvt_f32_to_i16:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; ALL-NEXT:    vmovd %xmm0, %eax
; ALL-NEXT:    # kill: def $ax killed $ax killed $eax
; ALL-NEXT:    retq
  %1 = fptrunc float %a0 to half
  %2 = bitcast half %1 to i16
  ret i16 %2
}

define <4 x i16> @cvt_4f32_to_4i16(<4 x float> %a0) nounwind {
; ALL-LABEL: cvt_4f32_to_4i16:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = fptrunc <4 x float> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  ret <4 x i16> %2
}

define <8 x i16> @cvt_4f32_to_8i16_undef(<4 x float> %a0) nounwind {
; ALL-LABEL: cvt_4f32_to_8i16_undef:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = fptrunc <4 x float> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  ret <8 x i16> %3
}

define <8 x i16> @cvt_4f32_to_8i16_zero(<4 x float> %a0) nounwind {
; ALL-LABEL: cvt_4f32_to_8i16_zero:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; ALL-NEXT:    retq
  %1 = fptrunc <4 x float> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  ret <8 x i16> %3
}

define <8 x i16> @cvt_8f32_to_8i16(<8 x float> %a0) nounwind {
; ALL-LABEL: cvt_8f32_to_8i16:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %ymm0, %xmm0
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    retq
  %1 = fptrunc <8 x float> %a0 to <8 x half>
  %2 = bitcast <8 x half> %1 to <8 x i16>
  ret <8 x i16> %2
}

define <16 x i16> @cvt_16f32_to_16i16(<16 x float> %a0) nounwind {
; AVX1-LABEL: cvt_16f32_to_16i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtps2ph $4, %ymm0, %xmm0
; AVX1-NEXT:    vcvtps2ph $4, %ymm1, %xmm1
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: cvt_16f32_to_16i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtps2ph $4, %ymm0, %xmm0
; AVX2-NEXT:    vcvtps2ph $4, %ymm1, %xmm1
; AVX2-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX2-NEXT:    retq
;
; AVX512-LABEL: cvt_16f32_to_16i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtps2ph $4, %zmm0, %ymm0
; AVX512-NEXT:    retq
  %1 = fptrunc <16 x float> %a0 to <16 x half>
  %2 = bitcast <16 x half> %1 to <16 x i16>
  ret <16 x i16> %2
}

;
; Float to Half (Store)
;

define void @store_cvt_f32_to_i16(float %a0, i16* %a1) nounwind {
; ALL-LABEL: store_cvt_f32_to_i16:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; ALL-NEXT:    vpextrw $0, %xmm0, (%rdi)
; ALL-NEXT:    retq
  %1 = fptrunc float %a0 to half
  %2 = bitcast half %1 to i16
  store i16 %2, i16* %a1
  ret void
}

define void @store_cvt_4f32_to_4i16(<4 x float> %a0, <4 x i16>* %a1) nounwind {
; ALL-LABEL: store_cvt_4f32_to_4i16:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, (%rdi)
; ALL-NEXT:    retq
  %1 = fptrunc <4 x float> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  store <4 x i16> %2, <4 x i16>* %a1
  ret void
}

define void @store_cvt_4f32_to_8i16_undef(<4 x float> %a0, <8 x i16>* %a1) nounwind {
; ALL-LABEL: store_cvt_4f32_to_8i16_undef:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; ALL-NEXT:    vmovaps %xmm0, (%rdi)
; ALL-NEXT:    retq
  %1 = fptrunc <4 x float> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  store <8 x i16> %3, <8 x i16>* %a1
  ret void
}

define void @store_cvt_4f32_to_8i16_zero(<4 x float> %a0, <8 x i16>* %a1) nounwind {
; ALL-LABEL: store_cvt_4f32_to_8i16_zero:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %xmm0, %xmm0
; ALL-NEXT:    vmovaps %xmm0, (%rdi)
; ALL-NEXT:    retq
  %1 = fptrunc <4 x float> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  store <8 x i16> %3, <8 x i16>* %a1
  ret void
}

define void @store_cvt_8f32_to_8i16(<8 x float> %a0, <8 x i16>* %a1) nounwind {
; ALL-LABEL: store_cvt_8f32_to_8i16:
; ALL:       # %bb.0:
; ALL-NEXT:    vcvtps2ph $4, %ymm0, (%rdi)
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    retq
  %1 = fptrunc <8 x float> %a0 to <8 x half>
  %2 = bitcast <8 x half> %1 to <8 x i16>
  store <8 x i16> %2, <8 x i16>* %a1
  ret void
}

define void @store_cvt_16f32_to_16i16(<16 x float> %a0, <16 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_16f32_to_16i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtps2ph $4, %ymm1, 16(%rdi)
; AVX1-NEXT:    vcvtps2ph $4, %ymm0, (%rdi)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: store_cvt_16f32_to_16i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtps2ph $4, %ymm1, 16(%rdi)
; AVX2-NEXT:    vcvtps2ph $4, %ymm0, (%rdi)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX512-LABEL: store_cvt_16f32_to_16i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtps2ph $4, %zmm0, (%rdi)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %1 = fptrunc <16 x float> %a0 to <16 x half>
  %2 = bitcast <16 x half> %1 to <16 x i16>
  store <16 x i16> %2, <16 x i16>* %a1
  ret void
}

;
; Double to Half
;

define i16 @cvt_f64_to_i16(double %a0) nounwind {
; ALL-LABEL: cvt_f64_to_i16:
; ALL:       # %bb.0:
; ALL-NEXT:    jmp __truncdfhf2@PLT # TAILCALL
  %1 = fptrunc double %a0 to half
  %2 = bitcast half %1 to i16
  ret i16 %2
}

define <2 x i16> @cvt_2f64_to_2i16(<2 x double> %a0) nounwind {
; ALL-LABEL: cvt_2f64_to_2i16:
; ALL:       # %bb.0:
; ALL-NEXT:    subq $40, %rsp
; ALL-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovaps (%rsp), %xmm0
; ALL-NEXT:    addq $40, %rsp
; ALL-NEXT:    retq
  %1 = fptrunc <2 x double> %a0 to <2 x half>
  %2 = bitcast <2 x half> %1 to <2 x i16>
  ret <2 x i16> %2
}

define <4 x i16> @cvt_4f64_to_4i16(<4 x double> %a0) nounwind {
; ALL-LABEL: cvt_4f64_to_4i16:
; ALL:       # %bb.0:
; ALL-NEXT:    subq $72, %rsp
; ALL-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; ALL-NEXT:    vextractf128 $1, %ymm0, %xmm0
; ALL-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovaps (%rsp), %xmm0
; ALL-NEXT:    addq $72, %rsp
; ALL-NEXT:    retq
  %1 = fptrunc <4 x double> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  ret <4 x i16> %2
}

define <8 x i16> @cvt_4f64_to_8i16_undef(<4 x double> %a0) nounwind {
; ALL-LABEL: cvt_4f64_to_8i16_undef:
; ALL:       # %bb.0:
; ALL-NEXT:    subq $72, %rsp
; ALL-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; ALL-NEXT:    vextractf128 $1, %ymm0, %xmm0
; ALL-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovaps (%rsp), %xmm0
; ALL-NEXT:    addq $72, %rsp
; ALL-NEXT:    retq
  %1 = fptrunc <4 x double> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  ret <8 x i16> %3
}

define <8 x i16> @cvt_4f64_to_8i16_zero(<4 x double> %a0) nounwind {
; ALL-LABEL: cvt_4f64_to_8i16_zero:
; ALL:       # %bb.0:
; ALL-NEXT:    subq $72, %rsp
; ALL-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; ALL-NEXT:    vextractf128 $1, %ymm0, %xmm0
; ALL-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; ALL-NEXT:    addq $72, %rsp
; ALL-NEXT:    retq
  %1 = fptrunc <4 x double> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  ret <8 x i16> %3
}

define <8 x i16> @cvt_8f64_to_8i16(<8 x double> %a0) nounwind {
; AVX1-LABEL: cvt_8f64_to_8i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    pushq %r15
; AVX1-NEXT:    pushq %r14
; AVX1-NEXT:    pushq %rbx
; AVX1-NEXT:    subq $64, %rsp
; AVX1-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
; AVX1-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %ebx
; AVX1-NEXT:    shll $16, %ebx
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movzwl %ax, %r15d
; AVX1-NEXT:    orl %ebx, %r15d
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %ebx
; AVX1-NEXT:    shll $16, %ebx
; AVX1-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movzwl %ax, %r14d
; AVX1-NEXT:    orl %ebx, %r14d
; AVX1-NEXT:    shlq $32, %r14
; AVX1-NEXT:    orq %r15, %r14
; AVX1-NEXT:    vpermilpd $1, (%rsp), %xmm0 # 16-byte Folded Reload
; AVX1-NEXT:    # xmm0 = mem[1,0]
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %ebx
; AVX1-NEXT:    shll $16, %ebx
; AVX1-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movzwl %ax, %r15d
; AVX1-NEXT:    orl %ebx, %r15d
; AVX1-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %ebx
; AVX1-NEXT:    shll $16, %ebx
; AVX1-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movzwl %ax, %eax
; AVX1-NEXT:    orl %ebx, %eax
; AVX1-NEXT:    shlq $32, %rax
; AVX1-NEXT:    orq %r15, %rax
; AVX1-NEXT:    vmovq %rax, %xmm0
; AVX1-NEXT:    vmovq %r14, %xmm1
; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX1-NEXT:    addq $64, %rsp
; AVX1-NEXT:    popq %rbx
; AVX1-NEXT:    popq %r14
; AVX1-NEXT:    popq %r15
; AVX1-NEXT:    retq
;
; AVX2-LABEL: cvt_8f64_to_8i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    subq $64, %rsp
; AVX2-NEXT:    vmovups %ymm1, (%rsp) # 32-byte Spill
; AVX2-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %ebx
; AVX2-NEXT:    shll $16, %ebx
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movzwl %ax, %r15d
; AVX2-NEXT:    orl %ebx, %r15d
; AVX2-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %ebx
; AVX2-NEXT:    shll $16, %ebx
; AVX2-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movzwl %ax, %r14d
; AVX2-NEXT:    orl %ebx, %r14d
; AVX2-NEXT:    shlq $32, %r14
; AVX2-NEXT:    orq %r15, %r14
; AVX2-NEXT:    vpermilpd $1, (%rsp), %xmm0 # 16-byte Folded Reload
; AVX2-NEXT:    # xmm0 = mem[1,0]
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %ebx
; AVX2-NEXT:    shll $16, %ebx
; AVX2-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movzwl %ax, %r15d
; AVX2-NEXT:    orl %ebx, %r15d
; AVX2-NEXT:    vmovupd (%rsp), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %ebx
; AVX2-NEXT:    shll $16, %ebx
; AVX2-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movzwl %ax, %eax
; AVX2-NEXT:    orl %ebx, %eax
; AVX2-NEXT:    shlq $32, %rax
; AVX2-NEXT:    orq %r15, %rax
; AVX2-NEXT:    vmovq %rax, %xmm0
; AVX2-NEXT:    vmovq %r14, %xmm1
; AVX2-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX2-NEXT:    addq $64, %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    retq
;
; AVX512-LABEL: cvt_8f64_to_8i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    pushq %r15
; AVX512-NEXT:    pushq %r14
; AVX512-NEXT:    pushq %rbx
; AVX512-NEXT:    subq $80, %rsp
; AVX512-NEXT:    vmovupd %zmm0, (%rsp) # 64-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %ebx
; AVX512-NEXT:    shll $16, %ebx
; AVX512-NEXT:    vmovups (%rsp), %zmm0 # 64-byte Reload
; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movzwl %ax, %r15d
; AVX512-NEXT:    orl %ebx, %r15d
; AVX512-NEXT:    vmovupd (%rsp), %zmm0 # 64-byte Reload
; AVX512-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX512-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %ebx
; AVX512-NEXT:    shll $16, %ebx
; AVX512-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movzwl %ax, %r14d
; AVX512-NEXT:    orl %ebx, %r14d
; AVX512-NEXT:    shlq $32, %r14
; AVX512-NEXT:    orq %r15, %r14
; AVX512-NEXT:    vmovupd (%rsp), %zmm0 # 64-byte Reload
; AVX512-NEXT:    vextractf64x4 $1, %zmm0, %ymm0
; AVX512-NEXT:    vmovupd %ymm0, (%rsp) # 32-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %ebx
; AVX512-NEXT:    shll $16, %ebx
; AVX512-NEXT:    vmovups (%rsp), %ymm0 # 32-byte Reload
; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movzwl %ax, %r15d
; AVX512-NEXT:    orl %ebx, %r15d
; AVX512-NEXT:    vmovupd (%rsp), %ymm0 # 32-byte Reload
; AVX512-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX512-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %ebx
; AVX512-NEXT:    shll $16, %ebx
; AVX512-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movzwl %ax, %eax
; AVX512-NEXT:    orl %ebx, %eax
; AVX512-NEXT:    shlq $32, %rax
; AVX512-NEXT:    orq %r15, %rax
; AVX512-NEXT:    vmovq %rax, %xmm0
; AVX512-NEXT:    vmovq %r14, %xmm1
; AVX512-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
; AVX512-NEXT:    addq $80, %rsp
; AVX512-NEXT:    popq %rbx
; AVX512-NEXT:    popq %r14
; AVX512-NEXT:    popq %r15
; AVX512-NEXT:    retq
  %1 = fptrunc <8 x double> %a0 to <8 x half>
  %2 = bitcast <8 x half> %1 to <8 x i16>
  ret <8 x i16> %2
}

;
; Double to Half (Store)
;

define void @store_cvt_f64_to_i16(double %a0, i16* %a1) nounwind {
; ALL-LABEL: store_cvt_f64_to_i16:
; ALL:       # %bb.0:
; ALL-NEXT:    pushq %rbx
; ALL-NEXT:    movq %rdi, %rbx
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rbx)
; ALL-NEXT:    popq %rbx
; ALL-NEXT:    retq
  %1 = fptrunc double %a0 to half
  %2 = bitcast half %1 to i16
  store i16 %2, i16* %a1
  ret void
}

define void @store_cvt_2f64_to_2i16(<2 x double> %a0, <2 x i16>* %a1) nounwind {
; ALL-LABEL: store_cvt_2f64_to_2i16:
; ALL:       # %bb.0:
; ALL-NEXT:    pushq %rbp
; ALL-NEXT:    pushq %rbx
; ALL-NEXT:    subq $24, %rsp
; ALL-NEXT:    movq %rdi, %rbx
; ALL-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; ALL-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movl %eax, %ebp
; ALL-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rbx)
; ALL-NEXT:    movw %bp, 2(%rbx)
; ALL-NEXT:    addq $24, %rsp
; ALL-NEXT:    popq %rbx
; ALL-NEXT:    popq %rbp
; ALL-NEXT:    retq
  %1 = fptrunc <2 x double> %a0 to <2 x half>
  %2 = bitcast <2 x half> %1 to <2 x i16>
  store <2 x i16> %2, <2 x i16>* %a1
  ret void
}

define void @store_cvt_4f64_to_4i16(<4 x double> %a0, <4 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_4f64_to_4i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    pushq %rbp
; AVX1-NEXT:    pushq %r15
; AVX1-NEXT:    pushq %r14
; AVX1-NEXT:    pushq %rbx
; AVX1-NEXT:    subq $56, %rsp
; AVX1-NEXT:    movq %rdi, %rbx
; AVX1-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %r14d
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %r15d
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %ebp
; AVX1-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movw %ax, 4(%rbx)
; AVX1-NEXT:    movw %bp, (%rbx)
; AVX1-NEXT:    movw %r15w, 6(%rbx)
; AVX1-NEXT:    movw %r14w, 2(%rbx)
; AVX1-NEXT:    addq $56, %rsp
; AVX1-NEXT:    popq %rbx
; AVX1-NEXT:    popq %r14
; AVX1-NEXT:    popq %r15
; AVX1-NEXT:    popq %rbp
; AVX1-NEXT:    retq
;
; AVX2-LABEL: store_cvt_4f64_to_4i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    subq $56, %rsp
; AVX2-NEXT:    movq %rdi, %rbx
; AVX2-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %r14d
; AVX2-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %r15d
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %ebp
; AVX2-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movw %ax, 4(%rbx)
; AVX2-NEXT:    movw %bp, (%rbx)
; AVX2-NEXT:    movw %r15w, 6(%rbx)
; AVX2-NEXT:    movw %r14w, 2(%rbx)
; AVX2-NEXT:    addq $56, %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512-LABEL: store_cvt_4f64_to_4i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    pushq %rbp
; AVX512-NEXT:    pushq %r15
; AVX512-NEXT:    pushq %r14
; AVX512-NEXT:    pushq %rbx
; AVX512-NEXT:    subq $56, %rsp
; AVX512-NEXT:    movq %rdi, %rbx
; AVX512-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %r14d
; AVX512-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX512-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX512-NEXT:    vmovapd %xmm0, (%rsp) # 16-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %r15d
; AVX512-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %ebp
; AVX512-NEXT:    vmovaps (%rsp), %xmm0 # 16-byte Reload
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movw %ax, 4(%rbx)
; AVX512-NEXT:    movw %bp, (%rbx)
; AVX512-NEXT:    movw %r15w, 6(%rbx)
; AVX512-NEXT:    movw %r14w, 2(%rbx)
; AVX512-NEXT:    addq $56, %rsp
; AVX512-NEXT:    popq %rbx
; AVX512-NEXT:    popq %r14
; AVX512-NEXT:    popq %r15
; AVX512-NEXT:    popq %rbp
; AVX512-NEXT:    retq
  %1 = fptrunc <4 x double> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  store <4 x i16> %2, <4 x i16>* %a1
  ret void
}

define void @store_cvt_4f64_to_8i16_undef(<4 x double> %a0, <8 x i16>* %a1) nounwind {
; ALL-LABEL: store_cvt_4f64_to_8i16_undef:
; ALL:       # %bb.0:
; ALL-NEXT:    pushq %rbx
; ALL-NEXT:    subq $64, %rsp
; ALL-NEXT:    movq %rdi, %rbx
; ALL-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; ALL-NEXT:    vextractf128 $1, %ymm0, %xmm0
; ALL-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovaps (%rsp), %xmm0
; ALL-NEXT:    vmovaps %xmm0, (%rbx)
; ALL-NEXT:    addq $64, %rsp
; ALL-NEXT:    popq %rbx
; ALL-NEXT:    retq
  %1 = fptrunc <4 x double> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> undef, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  store <8 x i16> %3, <8 x i16>* %a1
  ret void
}

define void @store_cvt_4f64_to_8i16_zero(<4 x double> %a0, <8 x i16>* %a1) nounwind {
; ALL-LABEL: store_cvt_4f64_to_8i16_zero:
; ALL:       # %bb.0:
; ALL-NEXT:    pushq %rbx
; ALL-NEXT:    subq $64, %rsp
; ALL-NEXT:    movq %rdi, %rbx
; ALL-NEXT:    vmovups %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; ALL-NEXT:    vextractf128 $1, %ymm0, %xmm0
; ALL-NEXT:    vmovaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; ALL-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; ALL-NEXT:    vzeroupper
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, (%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; ALL-NEXT:    # xmm0 = mem[1,0]
; ALL-NEXT:    callq __truncdfhf2@PLT
; ALL-NEXT:    movw %ax, {{[0-9]+}}(%rsp)
; ALL-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
; ALL-NEXT:    vmovaps %xmm0, (%rbx)
; ALL-NEXT:    addq $64, %rsp
; ALL-NEXT:    popq %rbx
; ALL-NEXT:    retq
  %1 = fptrunc <4 x double> %a0 to <4 x half>
  %2 = bitcast <4 x half> %1 to <4 x i16>
  %3 = shufflevector <4 x i16> %2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
  store <8 x i16> %3, <8 x i16>* %a1
  ret void
}

define void @store_cvt_8f64_to_8i16(<8 x double> %a0, <8 x i16>* %a1) nounwind {
; AVX1-LABEL: store_cvt_8f64_to_8i16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    pushq %rbp
; AVX1-NEXT:    pushq %r15
; AVX1-NEXT:    pushq %r14
; AVX1-NEXT:    pushq %r13
; AVX1-NEXT:    pushq %r12
; AVX1-NEXT:    pushq %rbx
; AVX1-NEXT:    subq $120, %rsp
; AVX1-NEXT:    movq %rdi, %rbx
; AVX1-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX1-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movw %ax, {{[-0-9]+}}(%r{{[sb]}}p) # 2-byte Spill
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movw %ax, {{[-0-9]+}}(%r{{[sb]}}p) # 2-byte Spill
; AVX1-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; AVX1-NEXT:    # xmm0 = mem[1,0]
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %r12d
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX1-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %r13d
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %ebp
; AVX1-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %r14d
; AVX1-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movl %eax, %r15d
; AVX1-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX1-NEXT:    callq __truncdfhf2@PLT
; AVX1-NEXT:    movw %ax, 12(%rbx)
; AVX1-NEXT:    movw %r15w, 8(%rbx)
; AVX1-NEXT:    movw %r14w, 4(%rbx)
; AVX1-NEXT:    movw %bp, (%rbx)
; AVX1-NEXT:    movw %r13w, 14(%rbx)
; AVX1-NEXT:    movw %r12w, 10(%rbx)
; AVX1-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 2-byte Folded Reload
; AVX1-NEXT:    movw %ax, 6(%rbx)
; AVX1-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 2-byte Folded Reload
; AVX1-NEXT:    movw %ax, 2(%rbx)
; AVX1-NEXT:    addq $120, %rsp
; AVX1-NEXT:    popq %rbx
; AVX1-NEXT:    popq %r12
; AVX1-NEXT:    popq %r13
; AVX1-NEXT:    popq %r14
; AVX1-NEXT:    popq %r15
; AVX1-NEXT:    popq %rbp
; AVX1-NEXT:    retq
;
; AVX2-LABEL: store_cvt_8f64_to_8i16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    pushq %rbp
; AVX2-NEXT:    pushq %r15
; AVX2-NEXT:    pushq %r14
; AVX2-NEXT:    pushq %r13
; AVX2-NEXT:    pushq %r12
; AVX2-NEXT:    pushq %rbx
; AVX2-NEXT:    subq $120, %rsp
; AVX2-NEXT:    movq %rdi, %rbx
; AVX2-NEXT:    vmovups %ymm1, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movw %ax, {{[-0-9]+}}(%r{{[sb]}}p) # 2-byte Spill
; AVX2-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movw %ax, {{[-0-9]+}}(%r{{[sb]}}p) # 2-byte Spill
; AVX2-NEXT:    vpermilpd $1, {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Folded Reload
; AVX2-NEXT:    # xmm0 = mem[1,0]
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %r12d
; AVX2-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX2-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX2-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %r13d
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %ebp
; AVX2-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %r14d
; AVX2-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movl %eax, %r15d
; AVX2-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX2-NEXT:    callq __truncdfhf2@PLT
; AVX2-NEXT:    movw %ax, 12(%rbx)
; AVX2-NEXT:    movw %r15w, 8(%rbx)
; AVX2-NEXT:    movw %r14w, 4(%rbx)
; AVX2-NEXT:    movw %bp, (%rbx)
; AVX2-NEXT:    movw %r13w, 14(%rbx)
; AVX2-NEXT:    movw %r12w, 10(%rbx)
; AVX2-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 2-byte Folded Reload
; AVX2-NEXT:    movw %ax, 6(%rbx)
; AVX2-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 2-byte Folded Reload
; AVX2-NEXT:    movw %ax, 2(%rbx)
; AVX2-NEXT:    addq $120, %rsp
; AVX2-NEXT:    popq %rbx
; AVX2-NEXT:    popq %r12
; AVX2-NEXT:    popq %r13
; AVX2-NEXT:    popq %r14
; AVX2-NEXT:    popq %r15
; AVX2-NEXT:    popq %rbp
; AVX2-NEXT:    retq
;
; AVX512-LABEL: store_cvt_8f64_to_8i16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    pushq %rbp
; AVX512-NEXT:    pushq %r15
; AVX512-NEXT:    pushq %r14
; AVX512-NEXT:    pushq %r13
; AVX512-NEXT:    pushq %r12
; AVX512-NEXT:    pushq %rbx
; AVX512-NEXT:    subq $152, %rsp
; AVX512-NEXT:    movq %rdi, %rbx
; AVX512-NEXT:    vmovupd %zmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 64-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movw %ax, {{[-0-9]+}}(%r{{[sb]}}p) # 2-byte Spill
; AVX512-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
; AVX512-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX512-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movw %ax, {{[-0-9]+}}(%r{{[sb]}}p) # 2-byte Spill
; AVX512-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
; AVX512-NEXT:    vextractf64x4 $1, %zmm0, %ymm0
; AVX512-NEXT:    vmovupd %ymm0, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %r12d
; AVX512-NEXT:    vmovupd {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX512-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX512-NEXT:    vmovapd %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
; AVX512-NEXT:    vpermilpd {{.*#+}} xmm0 = xmm0[1,0]
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %r13d
; AVX512-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %zmm0 # 64-byte Reload
; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $zmm0
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %ebp
; AVX512-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %r14d
; AVX512-NEXT:    vmovups {{[-0-9]+}}(%r{{[sb]}}p), %ymm0 # 32-byte Reload
; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movl %eax, %r15d
; AVX512-NEXT:    vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload
; AVX512-NEXT:    callq __truncdfhf2@PLT
; AVX512-NEXT:    movw %ax, 12(%rbx)
; AVX512-NEXT:    movw %r15w, 8(%rbx)
; AVX512-NEXT:    movw %r14w, 4(%rbx)
; AVX512-NEXT:    movw %bp, (%rbx)
; AVX512-NEXT:    movw %r13w, 14(%rbx)
; AVX512-NEXT:    movw %r12w, 10(%rbx)
; AVX512-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 2-byte Folded Reload
; AVX512-NEXT:    movw %ax, 6(%rbx)
; AVX512-NEXT:    movzwl {{[-0-9]+}}(%r{{[sb]}}p), %eax # 2-byte Folded Reload
; AVX512-NEXT:    movw %ax, 2(%rbx)
; AVX512-NEXT:    addq $152, %rsp
; AVX512-NEXT:    popq %rbx
; AVX512-NEXT:    popq %r12
; AVX512-NEXT:    popq %r13
; AVX512-NEXT:    popq %r14
; AVX512-NEXT:    popq %r15
; AVX512-NEXT:    popq %rbp
; AVX512-NEXT:    retq
  %1 = fptrunc <8 x double> %a0 to <8 x half>
  %2 = bitcast <8 x half> %1 to <8 x i16>
  store <8 x i16> %2, <8 x i16>* %a1
  ret void
}

define void @store_cvt_32f32_to_32f16(<32 x float> %a0, <32 x half>* %a1) nounwind {
; AVX1-LABEL: store_cvt_32f32_to_32f16:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vcvtps2ph $4, %ymm3, 48(%rdi)
; AVX1-NEXT:    vcvtps2ph $4, %ymm2, 32(%rdi)
; AVX1-NEXT:    vcvtps2ph $4, %ymm1, 16(%rdi)
; AVX1-NEXT:    vcvtps2ph $4, %ymm0, (%rdi)
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: store_cvt_32f32_to_32f16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vcvtps2ph $4, %ymm3, 48(%rdi)
; AVX2-NEXT:    vcvtps2ph $4, %ymm2, 32(%rdi)
; AVX2-NEXT:    vcvtps2ph $4, %ymm1, 16(%rdi)
; AVX2-NEXT:    vcvtps2ph $4, %ymm0, (%rdi)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
;
; AVX512-LABEL: store_cvt_32f32_to_32f16:
; AVX512:       # %bb.0:
; AVX512-NEXT:    vcvtps2ph $4, %zmm1, 32(%rdi)
; AVX512-NEXT:    vcvtps2ph $4, %zmm0, (%rdi)
; AVX512-NEXT:    vzeroupper
; AVX512-NEXT:    retq
  %1 = fptrunc <32 x float> %a0 to <32 x half>
  store <32 x half> %1, <32 x half>* %a1
  ret void
}