llvm/test/CodeGen/X86/haddsub-shuf.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse3           | FileCheck %s --check-prefixes=SSE,SSE_SLOW,SSE3
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+sse3,fast-hops | FileCheck %s --check-prefixes=SSE,SSE_FAST,SSE3
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3           | FileCheck %s --check-prefixes=SSE,SSE_SLOW,SSSE3,SSSE3_SLOW
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+ssse3,fast-hops | FileCheck %s --check-prefixes=SSE,SSE_FAST,SSSE3,SSSE3_FAST
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx             | FileCheck %s --check-prefixes=AVX,AVX1,AVX1_SLOW
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx,fast-hops   | FileCheck %s --check-prefixes=AVX,AVX1,AVX1_FAST
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2            | FileCheck %s --check-prefixes=AVX,AVX2,AVX2_SLOW
; RUN: llc < %s -mtriple=x86_64-unknown -mattr=+avx2,fast-hops  | FileCheck %s --check-prefixes=AVX,AVX2,AVX2_FAST

; The next 8 tests check for matching the horizontal op and eliminating the shuffle.
; PR34111 - https://bugs.llvm.org/show_bug.cgi?id=34111

define <4 x float> @hadd_v4f32(<4 x float> %a) {
; SSE-LABEL: hadd_v4f32:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm0, %xmm0
; SSE-NEXT:    retq
;
; AVX-LABEL: hadd_v4f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
; AVX-NEXT:    retq
  %a02 = shufflevector <4 x float> %a, <4 x float> undef, <2 x i32> <i32 0, i32 2>
  %a13 = shufflevector <4 x float> %a, <4 x float> undef, <2 x i32> <i32 1, i32 3>
  %hop = fadd <2 x float> %a02, %a13
  %shuf = shufflevector <2 x float> %hop, <2 x float> undef, <4 x i32> <i32 undef, i32 undef, i32 0, i32 1>
  ret <4 x float> %shuf
}

define <8 x float> @hadd_v8f32a(<8 x float> %a) {
; SSE_SLOW-LABEL: hadd_v8f32a:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movaps %xmm0, %xmm2
; SSE_SLOW-NEXT:    haddps %xmm1, %xmm2
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm2[0,0]
; SSE_SLOW-NEXT:    movaps %xmm2, %xmm1
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hadd_v8f32a:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    movaps %xmm0, %xmm2
; SSE_FAST-NEXT:    haddps %xmm1, %xmm2
; SSE_FAST-NEXT:    haddps %xmm0, %xmm0
; SSE_FAST-NEXT:    movaps %xmm2, %xmm1
; SSE_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hadd_v8f32a:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_SLOW-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
; AVX1_SLOW-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hadd_v8f32a:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX1_FAST-NEXT:    vhaddps %ymm0, %ymm1, %ymm0
; AVX1_FAST-NEXT:    retq
;
; AVX2-LABEL: hadd_v8f32a:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,2,1]
; AVX2-NEXT:    retq
  %a0 = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %a1 = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hop = fadd <4 x float> %a0, %a1
  %shuf = shufflevector <4 x float> %hop, <4 x float> undef, <8 x i32> <i32 undef, i32 undef, i32 0, i32 1, i32 undef, i32 undef, i32 2, i32 3>
  ret <8 x float> %shuf
}

define <8 x float> @hadd_v8f32b(<8 x float> %a) {
; SSE-LABEL: hadd_v8f32b:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm0, %xmm0
; SSE-NEXT:    haddps %xmm1, %xmm1
; SSE-NEXT:    retq
;
; AVX-LABEL: hadd_v8f32b:
; AVX:       # %bb.0:
; AVX-NEXT:    vhaddps %ymm0, %ymm0, %ymm0
; AVX-NEXT:    retq
  %a0 = shufflevector <8 x float> %a, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 4, i32 6, i32 undef, i32 undef>
  %a1 = shufflevector <8 x float> %a, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 5, i32 7, i32 undef, i32 undef>
  %hop = fadd <8 x float> %a0, %a1
  %shuf = shufflevector <8 x float> %hop, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
  ret <8 x float> %shuf
}

define <4 x float> @hsub_v4f32(<4 x float> %a) {
; SSE-LABEL: hsub_v4f32:
; SSE:       # %bb.0:
; SSE-NEXT:    hsubps %xmm0, %xmm0
; SSE-NEXT:    retq
;
; AVX-LABEL: hsub_v4f32:
; AVX:       # %bb.0:
; AVX-NEXT:    vhsubps %xmm0, %xmm0, %xmm0
; AVX-NEXT:    retq
  %a02 = shufflevector <4 x float> %a, <4 x float> undef, <2 x i32> <i32 0, i32 2>
  %a13 = shufflevector <4 x float> %a, <4 x float> undef, <2 x i32> <i32 1, i32 3>
  %hop = fsub <2 x float> %a02, %a13
  %shuf = shufflevector <2 x float> %hop, <2 x float> undef, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
  ret <4 x float> %shuf
}

define <8 x float> @hsub_v8f32a(<8 x float> %a) {
; SSE_SLOW-LABEL: hsub_v8f32a:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movaps %xmm0, %xmm2
; SSE_SLOW-NEXT:    hsubps %xmm1, %xmm2
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm2[0,0]
; SSE_SLOW-NEXT:    movaps %xmm2, %xmm1
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hsub_v8f32a:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    movaps %xmm0, %xmm2
; SSE_FAST-NEXT:    hsubps %xmm1, %xmm2
; SSE_FAST-NEXT:    hsubps %xmm0, %xmm0
; SSE_FAST-NEXT:    movaps %xmm2, %xmm1
; SSE_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hsub_v8f32a:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_SLOW-NEXT:    vhsubps %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vmovddup {{.*#+}} xmm1 = xmm0[0,0]
; AVX1_SLOW-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hsub_v8f32a:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm1
; AVX1_FAST-NEXT:    vhsubps %ymm0, %ymm1, %ymm0
; AVX1_FAST-NEXT:    retq
;
; AVX2-LABEL: hsub_v8f32a:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vhsubps %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,0,2,1]
; AVX2-NEXT:    retq
  %a0 = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %a1 = shufflevector <8 x float> %a, <8 x float> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hop = fsub <4 x float> %a0, %a1
  %shuf = shufflevector <4 x float> %hop, <4 x float> undef, <8 x i32> <i32 undef, i32 undef, i32 0, i32 1, i32 undef, i32 undef, i32 2, i32 3>
  ret <8 x float> %shuf
}

define <8 x float> @hsub_v8f32b(<8 x float> %a) {
; SSE-LABEL: hsub_v8f32b:
; SSE:       # %bb.0:
; SSE-NEXT:    hsubps %xmm0, %xmm0
; SSE-NEXT:    hsubps %xmm1, %xmm1
; SSE-NEXT:    retq
;
; AVX-LABEL: hsub_v8f32b:
; AVX:       # %bb.0:
; AVX-NEXT:    vhsubps %ymm0, %ymm0, %ymm0
; AVX-NEXT:    retq
  %a0 = shufflevector <8 x float> %a, <8 x float> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 4, i32 6, i32 undef, i32 undef>
  %a1 = shufflevector <8 x float> %a, <8 x float> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 5, i32 7, i32 undef, i32 undef>
  %hop = fsub <8 x float> %a0, %a1
  %shuf = shufflevector <8 x float> %hop, <8 x float> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
  ret <8 x float> %shuf
}

define <2 x double> @hadd_v2f64(<2 x double> %a) {
; SSE_SLOW-LABEL: hadd_v2f64:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm1
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
; SSE_SLOW-NEXT:    addsd %xmm0, %xmm1
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm1[0,0]
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hadd_v2f64:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    haddpd %xmm0, %xmm0
; SSE_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hadd_v2f64:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX1_SLOW-NEXT:    vaddsd %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hadd_v2f64:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    retq
;
; AVX2_SLOW-LABEL: hadd_v2f64:
; AVX2_SLOW:       # %bb.0:
; AVX2_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX2_SLOW-NEXT:    vaddsd %xmm1, %xmm0, %xmm0
; AVX2_SLOW-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX2_SLOW-NEXT:    retq
;
; AVX2_FAST-LABEL: hadd_v2f64:
; AVX2_FAST:       # %bb.0:
; AVX2_FAST-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
; AVX2_FAST-NEXT:    retq
  %a0 = shufflevector <2 x double> %a, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
  %a1 = shufflevector <2 x double> %a, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
  %hop = fadd <2 x double> %a0, %a1
  %shuf = shufflevector <2 x double> %hop, <2 x double> undef, <2 x i32> <i32 0, i32 0>
  ret <2 x double> %shuf
}

define <2 x double> @hadd_v2f64_scalar_splat(<2 x double> %a) {
; SSE_SLOW-LABEL: hadd_v2f64_scalar_splat:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm1
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
; SSE_SLOW-NEXT:    addsd %xmm0, %xmm1
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm1[0,0]
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hadd_v2f64_scalar_splat:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    haddpd %xmm0, %xmm0
; SSE_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hadd_v2f64_scalar_splat:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX1_SLOW-NEXT:    vaddsd %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hadd_v2f64_scalar_splat:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    retq
;
; AVX2_SLOW-LABEL: hadd_v2f64_scalar_splat:
; AVX2_SLOW:       # %bb.0:
; AVX2_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX2_SLOW-NEXT:    vaddsd %xmm1, %xmm0, %xmm0
; AVX2_SLOW-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX2_SLOW-NEXT:    retq
;
; AVX2_FAST-LABEL: hadd_v2f64_scalar_splat:
; AVX2_FAST:       # %bb.0:
; AVX2_FAST-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
; AVX2_FAST-NEXT:    retq
  %a0 = extractelement <2 x double> %a, i32 0
  %a1 = extractelement <2 x double> %a, i32 1
  %hop = fadd double %a0, %a1
  %ins = insertelement <2 x double> undef, double %hop, i32 0
  %shuf = shufflevector <2 x double> %ins, <2 x double> undef, <2 x i32> <i32 0, i32 0>
  ret <2 x double> %shuf
}

define <4 x double> @hadd_v4f64_scalar_splat(<4 x double> %a) {
; SSE_SLOW-LABEL: hadd_v4f64_scalar_splat:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm2
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm0[1]
; SSE_SLOW-NEXT:    addsd %xmm0, %xmm2
; SSE_SLOW-NEXT:    movapd %xmm1, %xmm3
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm3 = xmm3[1],xmm1[1]
; SSE_SLOW-NEXT:    addsd %xmm1, %xmm3
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm2[0,0]
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm1 = xmm3[0,0]
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hadd_v4f64_scalar_splat:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    haddpd %xmm0, %xmm0
; SSE_FAST-NEXT:    haddpd %xmm1, %xmm1
; SSE_FAST-NEXT:    retq
;
; AVX-LABEL: hadd_v4f64_scalar_splat:
; AVX:       # %bb.0:
; AVX-NEXT:    vhaddpd %ymm0, %ymm0, %ymm0
; AVX-NEXT:    retq
  %a0 = extractelement <4 x double> %a, i32 0
  %a1 = extractelement <4 x double> %a, i32 1
  %hop0 = fadd double %a0, %a1
  %a2 = extractelement <4 x double> %a, i32 2
  %a3 = extractelement <4 x double> %a, i32 3
  %hop1 = fadd double %a2, %a3
  %ins = insertelement <4 x double> undef, double %hop0, i32 0
  %ins2 = insertelement <4 x double> %ins,  double %hop1, i32 2
  %shuf = shufflevector <4 x double> %ins2, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
  ret <4 x double> %shuf
}

define <4 x double> @hadd_v4f64_scalar_broadcast(<4 x double> %a) {
; SSE_SLOW-LABEL: hadd_v4f64_scalar_broadcast:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm1
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
; SSE_SLOW-NEXT:    addsd %xmm0, %xmm1
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm1[0,0]
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm1
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hadd_v4f64_scalar_broadcast:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    haddpd %xmm0, %xmm0
; SSE_FAST-NEXT:    movapd %xmm0, %xmm1
; SSE_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hadd_v4f64_scalar_broadcast:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX1_SLOW-NEXT:    vaddsd %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX1_SLOW-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hadd_v4f64_scalar_broadcast:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
; AVX1_FAST-NEXT:    retq
;
; AVX2_SLOW-LABEL: hadd_v4f64_scalar_broadcast:
; AVX2_SLOW:       # %bb.0:
; AVX2_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX2_SLOW-NEXT:    vaddsd %xmm1, %xmm0, %xmm0
; AVX2_SLOW-NEXT:    vbroadcastsd %xmm0, %ymm0
; AVX2_SLOW-NEXT:    retq
;
; AVX2_FAST-LABEL: hadd_v4f64_scalar_broadcast:
; AVX2_FAST:       # %bb.0:
; AVX2_FAST-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
; AVX2_FAST-NEXT:    vbroadcastsd %xmm0, %ymm0
; AVX2_FAST-NEXT:    retq
  %a0 = extractelement <4 x double> %a, i32 0
  %a1 = extractelement <4 x double> %a, i32 1
  %hop0 = fadd double %a0, %a1
  %a2 = extractelement <4 x double> %a, i32 2
  %a3 = extractelement <4 x double> %a, i32 3
  %hop1 = fadd double %a2, %a3
  %ins = insertelement <4 x double> undef, double %hop0, i32 0
  %ins2 = insertelement <4 x double> %ins,  double %hop1, i32 2
  %shuf = shufflevector <4 x double> %ins2, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0>
  ret <4 x double> %shuf
}

define <4 x double> @hadd_v4f64(<4 x double> %a) {
; SSE_SLOW-LABEL: hadd_v4f64:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm2
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm0[1]
; SSE_SLOW-NEXT:    addsd %xmm0, %xmm2
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm2[0,0]
; SSE_SLOW-NEXT:    movapd %xmm1, %xmm2
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm1[1]
; SSE_SLOW-NEXT:    addsd %xmm1, %xmm2
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm1 = xmm2[0,0]
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hadd_v4f64:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    haddpd %xmm0, %xmm0
; SSE_FAST-NEXT:    haddpd %xmm1, %xmm1
; SSE_FAST-NEXT:    retq
;
; AVX-LABEL: hadd_v4f64:
; AVX:       # %bb.0:
; AVX-NEXT:    vhaddpd %ymm0, %ymm0, %ymm0
; AVX-NEXT:    retq
  %a0 = shufflevector <4 x double> %a, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 2, i32 undef>
  %a1 = shufflevector <4 x double> %a, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 3, i32 undef>
  %hop = fadd <4 x double> %a0, %a1
  %shuf = shufflevector <4 x double> %hop, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
  ret <4 x double> %shuf
}

define <2 x double> @hsub_v2f64(<2 x double> %a) {
; SSE_SLOW-LABEL: hsub_v2f64:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm1
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1]
; SSE_SLOW-NEXT:    subsd %xmm1, %xmm0
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm0[0,0]
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hsub_v2f64:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    hsubpd %xmm0, %xmm0
; SSE_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hsub_v2f64:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX1_SLOW-NEXT:    vsubsd %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hsub_v2f64:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vhsubpd %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    retq
;
; AVX2_SLOW-LABEL: hsub_v2f64:
; AVX2_SLOW:       # %bb.0:
; AVX2_SLOW-NEXT:    vshufpd {{.*#+}} xmm1 = xmm0[1,0]
; AVX2_SLOW-NEXT:    vsubsd %xmm1, %xmm0, %xmm0
; AVX2_SLOW-NEXT:    vmovddup {{.*#+}} xmm0 = xmm0[0,0]
; AVX2_SLOW-NEXT:    retq
;
; AVX2_FAST-LABEL: hsub_v2f64:
; AVX2_FAST:       # %bb.0:
; AVX2_FAST-NEXT:    vhsubpd %xmm0, %xmm0, %xmm0
; AVX2_FAST-NEXT:    retq
  %a0 = shufflevector <2 x double> %a, <2 x double> undef, <2 x i32> <i32 0, i32 undef>
  %a1 = shufflevector <2 x double> %a, <2 x double> undef, <2 x i32> <i32 1, i32 undef>
  %hop = fsub <2 x double> %a0, %a1
  %shuf = shufflevector <2 x double> %hop, <2 x double> undef, <2 x i32> <i32 undef, i32 0>
  ret <2 x double> %shuf
}

define <4 x double> @hsub_v4f64(<4 x double> %a) {
; SSE_SLOW-LABEL: hsub_v4f64:
; SSE_SLOW:       # %bb.0:
; SSE_SLOW-NEXT:    movapd %xmm0, %xmm2
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm0[1]
; SSE_SLOW-NEXT:    subsd %xmm2, %xmm0
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm0 = xmm0[0,0]
; SSE_SLOW-NEXT:    movapd %xmm1, %xmm2
; SSE_SLOW-NEXT:    unpckhpd {{.*#+}} xmm2 = xmm2[1],xmm1[1]
; SSE_SLOW-NEXT:    subsd %xmm2, %xmm1
; SSE_SLOW-NEXT:    movddup {{.*#+}} xmm1 = xmm1[0,0]
; SSE_SLOW-NEXT:    retq
;
; SSE_FAST-LABEL: hsub_v4f64:
; SSE_FAST:       # %bb.0:
; SSE_FAST-NEXT:    hsubpd %xmm0, %xmm0
; SSE_FAST-NEXT:    hsubpd %xmm1, %xmm1
; SSE_FAST-NEXT:    retq
;
; AVX-LABEL: hsub_v4f64:
; AVX:       # %bb.0:
; AVX-NEXT:    vhsubpd %ymm0, %ymm0, %ymm0
; AVX-NEXT:    retq
  %a0 = shufflevector <4 x double> %a, <4 x double> undef, <4 x i32> <i32 0, i32 undef, i32 2, i32 undef>
  %a1 = shufflevector <4 x double> %a, <4 x double> undef, <4 x i32> <i32 1, i32 undef, i32 3, i32 undef>
  %hop = fsub <4 x double> %a0, %a1
  %shuf = shufflevector <4 x double> %hop, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 2, i32 2>
  ret <4 x double> %shuf
}

define <4 x i32> @hadd_v4i32(<4 x i32> %a) {
; SSE3-LABEL: hadd_v4i32:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[1,1,2,3]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,2]
; SSE3-NEXT:    paddd %xmm1, %xmm0
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hadd_v4i32:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddd %xmm0, %xmm0
; SSSE3-NEXT:    retq
;
; AVX-LABEL: hadd_v4i32:
; AVX:       # %bb.0:
; AVX-NEXT:    vphaddd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    retq
  %a02 = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
  %a13 = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
  %hop = add <4 x i32> %a02, %a13
  %shuf = shufflevector <4 x i32> %hop, <4 x i32> undef, <4 x i32> <i32 0, i32 undef, i32 undef, i32 1>
  ret <4 x i32> %shuf
}

define <8 x i32> @hadd_v8i32a(<8 x i32> %a) {
; SSE3-LABEL: hadd_v8i32a:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movaps %xmm0, %xmm2
; SSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm1[0,2]
; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
; SSE3-NEXT:    paddd %xmm0, %xmm2
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSE3-NEXT:    movdqa %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3_SLOW-LABEL: hadd_v8i32a:
; SSSE3_SLOW:       # %bb.0:
; SSSE3_SLOW-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_SLOW-NEXT:    phaddd %xmm1, %xmm2
; SSSE3_SLOW-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSSE3_SLOW-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_SLOW-NEXT:    retq
;
; SSSE3_FAST-LABEL: hadd_v8i32a:
; SSSE3_FAST:       # %bb.0:
; SSSE3_FAST-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_FAST-NEXT:    phaddd %xmm1, %xmm2
; SSSE3_FAST-NEXT:    phaddd %xmm0, %xmm0
; SSSE3_FAST-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hadd_v8i32a:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_SLOW-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[0,1,0,1]
; AVX1_SLOW-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hadd_v8i32a:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_FAST-NEXT:    vphaddd %xmm1, %xmm0, %xmm1
; AVX1_FAST-NEXT:    vphaddd %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1_FAST-NEXT:    retq
;
; AVX2-LABEL: hadd_v8i32a:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
; AVX2-NEXT:    retq
  %a0 = shufflevector <8 x i32> %a, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %a1 = shufflevector <8 x i32> %a, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hop = add <4 x i32> %a0, %a1
  %shuf = shufflevector <4 x i32> %hop, <4 x i32> undef, <8 x i32> <i32 undef, i32 undef, i32 0, i32 1, i32 undef, i32 undef, i32 2, i32 3>
  ret <8 x i32> %shuf
}

define <8 x i32> @hadd_v8i32b(<8 x i32> %a) {
; SSE3-LABEL: hadd_v8i32b:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,3,1,3]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,0,2]
; SSE3-NEXT:    paddd %xmm2, %xmm0
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,1,3]
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,0,2]
; SSE3-NEXT:    paddd %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hadd_v8i32b:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddd %xmm0, %xmm0
; SSSE3-NEXT:    phaddd %xmm1, %xmm1
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hadd_v8i32b:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vphaddd %xmm0, %xmm0, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vphaddd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hadd_v8i32b:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vphaddd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    retq
  %a0 = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 4, i32 6, i32 undef, i32 undef>
  %a1 = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 5, i32 7, i32 undef, i32 undef>
  %hop = add <8 x i32> %a0, %a1
  %shuf = shufflevector <8 x i32> %hop, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
  ret <8 x i32> %shuf
}

define <4 x i32> @hsub_v4i32(<4 x i32> %a) {
; SSE3-LABEL: hsub_v4i32:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[0,3,1,3]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,0,3]
; SSE3-NEXT:    psubd %xmm1, %xmm0
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hsub_v4i32:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phsubd %xmm0, %xmm0
; SSSE3-NEXT:    retq
;
; AVX-LABEL: hsub_v4i32:
; AVX:       # %bb.0:
; AVX-NEXT:    vphsubd %xmm0, %xmm0, %xmm0
; AVX-NEXT:    retq
  %a02 = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 0, i32 2, i32 undef, i32 undef>
  %a13 = shufflevector <4 x i32> %a, <4 x i32> undef, <4 x i32> <i32 1, i32 3, i32 undef, i32 undef>
  %hop = sub <4 x i32> %a02, %a13
  %shuf = shufflevector <4 x i32> %hop, <4 x i32> undef, <4 x i32> <i32 undef, i32 1, i32 0, i32 undef>
  ret <4 x i32> %shuf
}

define <8 x i32> @hsub_v8i32a(<8 x i32> %a) {
; SSE3-LABEL: hsub_v8i32a:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movaps %xmm0, %xmm2
; SSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[0,2],xmm1[0,2]
; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
; SSE3-NEXT:    psubd %xmm0, %xmm2
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSE3-NEXT:    movdqa %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3_SLOW-LABEL: hsub_v8i32a:
; SSSE3_SLOW:       # %bb.0:
; SSSE3_SLOW-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_SLOW-NEXT:    phsubd %xmm1, %xmm2
; SSSE3_SLOW-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSSE3_SLOW-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_SLOW-NEXT:    retq
;
; SSSE3_FAST-LABEL: hsub_v8i32a:
; SSSE3_FAST:       # %bb.0:
; SSSE3_FAST-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_FAST-NEXT:    phsubd %xmm1, %xmm2
; SSSE3_FAST-NEXT:    phsubd %xmm0, %xmm0
; SSSE3_FAST-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hsub_v8i32a:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_SLOW-NEXT:    vphsubd %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[0,1,0,1]
; AVX1_SLOW-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hsub_v8i32a:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_FAST-NEXT:    vphsubd %xmm1, %xmm0, %xmm1
; AVX1_FAST-NEXT:    vphsubd %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1_FAST-NEXT:    retq
;
; AVX2-LABEL: hsub_v8i32a:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vphsubd %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
; AVX2-NEXT:    retq
  %a0 = shufflevector <8 x i32> %a, <8 x i32> undef, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %a1 = shufflevector <8 x i32> %a, <8 x i32> undef, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hop = sub <4 x i32> %a0, %a1
  %shuf = shufflevector <4 x i32> %hop, <4 x i32> undef, <8 x i32> <i32 undef, i32 undef, i32 0, i32 1, i32 undef, i32 undef, i32 2, i32 3>
  ret <8 x i32> %shuf
}

define <8 x i32> @hsub_v8i32b(<8 x i32> %a) {
; SSE3-LABEL: hsub_v8i32b:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[1,3,1,3]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,0,2]
; SSE3-NEXT:    psubd %xmm2, %xmm0
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm1[1,3,1,3]
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,2,0,2]
; SSE3-NEXT:    psubd %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hsub_v8i32b:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phsubd %xmm0, %xmm0
; SSSE3-NEXT:    phsubd %xmm1, %xmm1
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hsub_v8i32b:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vphsubd %xmm0, %xmm0, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vphsubd %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hsub_v8i32b:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vphsubd %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    retq
  %a0 = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> <i32 0, i32 2, i32 undef, i32 undef, i32 4, i32 6, i32 undef, i32 undef>
  %a1 = shufflevector <8 x i32> %a, <8 x i32> undef, <8 x i32> <i32 1, i32 3, i32 undef, i32 undef, i32 5, i32 7, i32 undef, i32 undef>
  %hop = sub <8 x i32> %a0, %a1
  %shuf = shufflevector <8 x i32> %hop, <8 x i32> undef, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5>
  ret <8 x i32> %shuf
}

define <8 x i16> @hadd_v8i16(<8 x i16> %a) {
; SSE3-LABEL: hadd_v8i16:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[0,2,2,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,0,2]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[3,1,2,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,5,6,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,2,0]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,5,4]
; SSE3-NEXT:    paddw %xmm1, %xmm0
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hadd_v8i16:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddw %xmm0, %xmm0
; SSSE3-NEXT:    retq
;
; AVX-LABEL: hadd_v8i16:
; AVX:       # %bb.0:
; AVX-NEXT:    vphaddw %xmm0, %xmm0, %xmm0
; AVX-NEXT:    retq
  %a0246 = shufflevector <8 x i16> %a, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
  %a1357 = shufflevector <8 x i16> %a, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
  %hop = add <8 x i16> %a0246, %a1357
  %shuf = shufflevector <8 x i16> %hop, <8 x i16> undef, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 0, i32 1, i32 2, i32 3>
  ret <8 x i16> %shuf
}

define <16 x i16> @hadd_v16i16a(<16 x i16> %a) {
; SSE3-LABEL: hadd_v16i16a:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movdqa %xmm1, %xmm3
; SSE3-NEXT:    pslld $16, %xmm3
; SSE3-NEXT:    psrad $16, %xmm3
; SSE3-NEXT:    movdqa %xmm0, %xmm2
; SSE3-NEXT:    pslld $16, %xmm2
; SSE3-NEXT:    psrad $16, %xmm2
; SSE3-NEXT:    packssdw %xmm3, %xmm2
; SSE3-NEXT:    psrad $16, %xmm1
; SSE3-NEXT:    psrad $16, %xmm0
; SSE3-NEXT:    packssdw %xmm1, %xmm0
; SSE3-NEXT:    paddw %xmm0, %xmm2
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSE3-NEXT:    movdqa %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3_SLOW-LABEL: hadd_v16i16a:
; SSSE3_SLOW:       # %bb.0:
; SSSE3_SLOW-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_SLOW-NEXT:    phaddw %xmm1, %xmm2
; SSSE3_SLOW-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSSE3_SLOW-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_SLOW-NEXT:    retq
;
; SSSE3_FAST-LABEL: hadd_v16i16a:
; SSSE3_FAST:       # %bb.0:
; SSSE3_FAST-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_FAST-NEXT:    phaddw %xmm1, %xmm2
; SSSE3_FAST-NEXT:    phaddw %xmm0, %xmm0
; SSSE3_FAST-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hadd_v16i16a:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_SLOW-NEXT:    vphaddw %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[0,1,0,1]
; AVX1_SLOW-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hadd_v16i16a:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_FAST-NEXT:    vphaddw %xmm1, %xmm0, %xmm1
; AVX1_FAST-NEXT:    vphaddw %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1_FAST-NEXT:    retq
;
; AVX2-LABEL: hadd_v16i16a:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vphaddw %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
; AVX2-NEXT:    retq
  %a0 = shufflevector <16 x i16> %a, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
  %a1 = shufflevector <16 x i16> %a, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
  %hop = add <8 x i16> %a0, %a1
  %shuf = shufflevector <8 x i16> %hop, <8 x i16> undef, <16 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 4, i32 5, i32 6, i32 7>
  ret <16 x i16> %shuf
}

define <16 x i16> @hadd_v16i16b(<16 x i16> %a) {
; SSE3-LABEL: hadd_v16i16b:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm0[3,1,1,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,5,5,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,0,2,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,7,5,4]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,0,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,4]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,1,3,2,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,4,5]
; SSE3-NEXT:    paddw %xmm2, %xmm0
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[3,1,1,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,5,5,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,0,2,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,7,5,4]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,0,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,4]
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,3,2,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,4,5]
; SSE3-NEXT:    paddw %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hadd_v16i16b:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddw %xmm0, %xmm0
; SSSE3-NEXT:    phaddw %xmm1, %xmm1
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hadd_v16i16b:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vphaddw %xmm0, %xmm0, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vphaddw %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hadd_v16i16b:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vphaddw %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    retq
  %a0 = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef>
  %a1 = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef>
  %hop = add <16 x i16> %a0, %a1
  %shuf = shufflevector <16 x i16> %hop, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11, i32 8, i32 9, i32 10, i32 11>
  ret <16 x i16> %shuf
}

define <8 x i16> @hsub_v8i16(<8 x i16> %a) {
; SSE3-LABEL: hsub_v8i16:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,1,3]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm1 = xmm0[1,1,3,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,4,6,6]
; SSE3-NEXT:    psubw %xmm1, %xmm0
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hsub_v8i16:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phsubw %xmm0, %xmm0
; SSSE3-NEXT:    retq
;
; AVX-LABEL: hsub_v8i16:
; AVX:       # %bb.0:
; AVX-NEXT:    vphsubw %xmm0, %xmm0, %xmm0
; AVX-NEXT:    retq
  %a0246 = shufflevector <8 x i16> %a, <8 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef>
  %a1357 = shufflevector <8 x i16> %a, <8 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef>
  %hop = sub <8 x i16> %a0246, %a1357
  %shuf = shufflevector <8 x i16> %hop, <8 x i16> undef, <8 x i32> <i32 0, i32 undef, i32 2, i32 undef, i32 undef, i32 1, i32 undef, i32 3>
  ret <8 x i16> %shuf
}

define <16 x i16> @hsub_v16i16a(<16 x i16> %a) {
; SSE3-LABEL: hsub_v16i16a:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movdqa %xmm1, %xmm3
; SSE3-NEXT:    pslld $16, %xmm3
; SSE3-NEXT:    psrad $16, %xmm3
; SSE3-NEXT:    movdqa %xmm0, %xmm2
; SSE3-NEXT:    pslld $16, %xmm2
; SSE3-NEXT:    psrad $16, %xmm2
; SSE3-NEXT:    packssdw %xmm3, %xmm2
; SSE3-NEXT:    psrad $16, %xmm1
; SSE3-NEXT:    psrad $16, %xmm0
; SSE3-NEXT:    packssdw %xmm1, %xmm0
; SSE3-NEXT:    psubw %xmm0, %xmm2
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSE3-NEXT:    movdqa %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3_SLOW-LABEL: hsub_v16i16a:
; SSSE3_SLOW:       # %bb.0:
; SSSE3_SLOW-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_SLOW-NEXT:    phsubw %xmm1, %xmm2
; SSSE3_SLOW-NEXT:    pshufd {{.*#+}} xmm0 = xmm2[0,1,0,1]
; SSSE3_SLOW-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_SLOW-NEXT:    retq
;
; SSSE3_FAST-LABEL: hsub_v16i16a:
; SSSE3_FAST:       # %bb.0:
; SSSE3_FAST-NEXT:    movdqa %xmm0, %xmm2
; SSSE3_FAST-NEXT:    phsubw %xmm1, %xmm2
; SSSE3_FAST-NEXT:    phsubw %xmm0, %xmm0
; SSSE3_FAST-NEXT:    movdqa %xmm2, %xmm1
; SSSE3_FAST-NEXT:    retq
;
; AVX1_SLOW-LABEL: hsub_v16i16a:
; AVX1_SLOW:       # %bb.0:
; AVX1_SLOW-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_SLOW-NEXT:    vphsubw %xmm1, %xmm0, %xmm0
; AVX1_SLOW-NEXT:    vpshufd {{.*#+}} xmm1 = xmm0[0,1,0,1]
; AVX1_SLOW-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1_SLOW-NEXT:    retq
;
; AVX1_FAST-LABEL: hsub_v16i16a:
; AVX1_FAST:       # %bb.0:
; AVX1_FAST-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1_FAST-NEXT:    vphsubw %xmm1, %xmm0, %xmm1
; AVX1_FAST-NEXT:    vphsubw %xmm0, %xmm0, %xmm0
; AVX1_FAST-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1_FAST-NEXT:    retq
;
; AVX2-LABEL: hsub_v16i16a:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vphsubw %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,0,2,1]
; AVX2-NEXT:    retq
  %a0 = shufflevector <16 x i16> %a, <16 x i16> undef, <8 x i32> <i32 0, i32 2, i32 4, i32 6, i32 8, i32 10, i32 12, i32 14>
  %a1 = shufflevector <16 x i16> %a, <16 x i16> undef, <8 x i32> <i32 1, i32 3, i32 5, i32 7, i32 9, i32 11, i32 13, i32 15>
  %hop = sub <8 x i16> %a0, %a1
  %shuf = shufflevector <8 x i16> %hop, <8 x i16> undef, <16 x i32> <i32 undef, i32 undef, i32 undef, i32 undef, i32 0, i32 1, i32 2, i32 3, i32 undef, i32 undef, i32 undef, i32 undef, i32 4, i32 5, i32 6, i32 7>
  ret <16 x i16> %shuf
}

define <16 x i16> @hsub_v16i16b(<16 x i16> %a) {
; SSE3-LABEL: hsub_v16i16b:
; SSE3:       # %bb.0:
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm0[3,1,1,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,5,5,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,0,2,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,7,5,4]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,0,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,4]
; SSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm0 = xmm0[0,1,3,2,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,7,6,4,5]
; SSE3-NEXT:    psubw %xmm2, %xmm0
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm1[3,1,1,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,7,5,5,7]
; SSE3-NEXT:    pshufd {{.*#+}} xmm2 = xmm2[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm2 = xmm2[1,0,2,3,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,7,5,4]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,0,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,4]
; SSE3-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,3,2,1]
; SSE3-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[0,1,3,2,4,5,6,7]
; SSE3-NEXT:    pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,7,6,4,5]
; SSE3-NEXT:    psubw %xmm2, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hsub_v16i16b:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phsubw %xmm0, %xmm0
; SSSE3-NEXT:    phsubw %xmm1, %xmm1
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hsub_v16i16b:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vphsubw %xmm0, %xmm0, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm0
; AVX1-NEXT:    vphsubw %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm1, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hsub_v16i16b:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vphsubw %ymm0, %ymm0, %ymm0
; AVX2-NEXT:    retq
  %a0 = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 undef, i32 undef, i32 undef, i32 undef, i32 8, i32 10, i32 12, i32 14, i32 undef, i32 undef, i32 undef, i32 undef>
  %a1 = shufflevector <16 x i16> %a, <16 x i16> undef, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 undef, i32 undef, i32 undef, i32 undef, i32 9, i32 11, i32 13, i32 15, i32 undef, i32 undef, i32 undef, i32 undef>
  %hop = sub <16 x i16> %a0, %a1
  %shuf = shufflevector <16 x i16> %hop, <16 x i16> undef, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11, i32 8, i32 9, i32 10, i32 11>
  ret <16 x i16> %shuf
}

define <4 x float> @broadcast_haddps_v4f32(<4 x float> %a0) {
; SSE-LABEL: broadcast_haddps_v4f32:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm0, %xmm0
; SSE-NEXT:    movsldup {{.*#+}} xmm0 = xmm0[0,0,2,2]
; SSE-NEXT:    retq
;
; AVX1-LABEL: broadcast_haddps_v4f32:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
; AVX1-NEXT:    vmovsldup {{.*#+}} xmm0 = xmm0[0,0,2,2]
; AVX1-NEXT:    retq
;
; AVX2-LABEL: broadcast_haddps_v4f32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vhaddps %xmm0, %xmm0, %xmm0
; AVX2-NEXT:    vbroadcastss %xmm0, %xmm0
; AVX2-NEXT:    retq
  %1 = tail call <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float> %a0, <4 x float> %a0)
  %2 = shufflevector <4 x float> %1, <4 x float> undef, <4 x i32> zeroinitializer
  ret <4 x float> %2
}

declare <4 x float> @llvm.x86.sse3.hadd.ps(<4 x float>, <4 x float>)

define <4 x float> @PR34724_1(<4 x float> %a, <4 x float> %b) {
; SSE-LABEL: PR34724_1:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm1, %xmm0
; SSE-NEXT:    retq
;
; AVX-LABEL: PR34724_1:
; AVX:       # %bb.0:
; AVX-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
; AVX-NEXT:    retq
  %t0 = shufflevector <4 x float> %a, <4 x float> %b, <2 x i32> <i32 2, i32 4>
  %t1 = shufflevector <4 x float> %a, <4 x float> %b, <2 x i32> <i32 3, i32 5>
  %t2 = fadd <2 x float> %t0, %t1
  %vecinit9 = shufflevector <2 x float> %t2, <2 x float> undef, <4 x i32> <i32 undef, i32 0, i32 1, i32 undef>
  %t3 = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> <i32 undef, i32 undef, i32 undef, i32 2>
  %t4 = fadd <4 x float> %t3, %b
  %vecinit13 = shufflevector <4 x float> %vecinit9, <4 x float> %t4, <4 x i32> <i32 undef, i32 1, i32 2, i32 7>
  ret <4 x float> %vecinit13
}

define <4 x float> @PR34724_2(<4 x float> %a, <4 x float> %b) {
; SSE-LABEL: PR34724_2:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm1, %xmm0
; SSE-NEXT:    retq
;
; AVX-LABEL: PR34724_2:
; AVX:       # %bb.0:
; AVX-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
; AVX-NEXT:    retq
  %t0 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 2, i32 4, i32 undef, i32 undef>
  %t1 = shufflevector <4 x float> %a, <4 x float> %b, <4 x i32> <i32 3, i32 5, i32 undef, i32 undef>
  %t2 = fadd <4 x float> %t0, %t1
  %vecinit9 = shufflevector <4 x float> %t2, <4 x float> undef, <4 x i32> <i32 undef, i32 0, i32 1, i32 undef>
  %t3 = shufflevector <4 x float> %b, <4 x float> undef, <4 x i32> <i32 undef, i32 undef, i32 undef, i32 2>
  %t4 = fadd <4 x float> %t3, %b
  %vecinit13 = shufflevector <4 x float> %vecinit9, <4 x float> %t4, <4 x i32> <i32 undef, i32 1, i32 2, i32 7>
  ret <4 x float> %vecinit13
}

;
; fold HOP(LOSUBVECTOR(SHUFFLE(X)),HISUBVECTOR(SHUFFLE(X)))
;  --> SHUFFLE(HOP(LOSUBVECTOR(X),HISUBVECTOR(X))).
;

define <4 x float> @hadd_4f32_v8f32_shuffle(<8 x float> %a0) {
; SSE-LABEL: hadd_4f32_v8f32_shuffle:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm1, %xmm0
; SSE-NEXT:    movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE-NEXT:    retq
;
; AVX-LABEL: hadd_4f32_v8f32_shuffle:
; AVX:       # %bb.0:
; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
  %shuf256 = shufflevector <8 x float> %a0, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 6, i32 7, i32 6, i32 7>
  %lo = shufflevector <8 x float> %shuf256, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %hi = shufflevector <8 x float> %shuf256, <8 x float> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  %hadd0 = shufflevector <4 x float> %lo, <4 x float> %hi, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %hadd1 = shufflevector <4 x float> %lo, <4 x float> %hi, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hadd = fadd <4 x float> %hadd0, %hadd1
  ret <4 x float> %hadd
}

define <4 x float> @hsub_4f32_v8f32_shuffle(<8 x float> %a0) {
; SSE-LABEL: hsub_4f32_v8f32_shuffle:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm1, %xmm0
; SSE-NEXT:    movshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSE-NEXT:    retq
;
; AVX-LABEL: hsub_4f32_v8f32_shuffle:
; AVX:       # %bb.0:
; AVX-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX-NEXT:    vhaddps %xmm1, %xmm0, %xmm0
; AVX-NEXT:    vmovshdup {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX-NEXT:    vzeroupper
; AVX-NEXT:    retq
  %shuf256 = shufflevector <8 x float> %a0, <8 x float> undef, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 6, i32 7, i32 6, i32 7>
  %lo = shufflevector <8 x float> %shuf256, <8 x float> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %hi = shufflevector <8 x float> %shuf256, <8 x float> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  %hsub0 = shufflevector <4 x float> %lo, <4 x float> %hi, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %hsub1 = shufflevector <4 x float> %lo, <4 x float> %hi, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hsub = fadd <4 x float> %hsub0, %hsub1
  ret <4 x float> %hsub
}

define <4 x i32> @hadd_4i32_v8i32_shuffle(<8 x i32> %a0) {
; SSE3-LABEL: hadd_4i32_v8i32_shuffle:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movaps %xmm0, %xmm2
; SSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,2],xmm1[2,2]
; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3],xmm1[3,3]
; SSE3-NEXT:    paddd %xmm2, %xmm0
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hadd_4i32_v8i32_shuffle:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddd %xmm1, %xmm0
; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hadd_4i32_v8i32_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hadd_4i32_v8i32_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %shuf256 = shufflevector <8 x i32> %a0, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 6, i32 7, i32 6, i32 7>
  %lo = shufflevector <8 x i32> %shuf256, <8 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %hi = shufflevector <8 x i32> %shuf256, <8 x i32> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  %hadd0 = shufflevector <4 x i32> %lo, <4 x i32> %hi, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %hadd1 = shufflevector <4 x i32> %lo, <4 x i32> %hi, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hadd = add <4 x i32> %hadd0, %hadd1
  ret <4 x i32> %hadd
}

define <4 x i32> @hsub_4i32_v8i32_shuffle(<8 x i32> %a0) {
; SSE3-LABEL: hsub_4i32_v8i32_shuffle:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movaps %xmm0, %xmm2
; SSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[2,2],xmm1[2,2]
; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[3,3],xmm1[3,3]
; SSE3-NEXT:    paddd %xmm2, %xmm0
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hsub_4i32_v8i32_shuffle:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddd %xmm1, %xmm0
; SSSE3-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hsub_4i32_v8i32_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm1
; AVX1-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
; AVX1-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX1-NEXT:    vzeroupper
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hsub_4i32_v8i32_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vphaddd %xmm1, %xmm0, %xmm0
; AVX2-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %shuf256 = shufflevector <8 x i32> %a0, <8 x i32> undef, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 6, i32 7, i32 6, i32 7>
  %lo = shufflevector <8 x i32> %shuf256, <8 x i32> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
  %hi = shufflevector <8 x i32> %shuf256, <8 x i32> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7>
  %hsub0 = shufflevector <4 x i32> %lo, <4 x i32> %hi, <4 x i32> <i32 0, i32 2, i32 4, i32 6>
  %hsub1 = shufflevector <4 x i32> %lo, <4 x i32> %hi, <4 x i32> <i32 1, i32 3, i32 5, i32 7>
  %hsub = add <4 x i32> %hsub0, %hsub1
  ret <4 x i32> %hsub
}

;
; fold HOP(SHUFFLE(X,Y),SHUFFLE(X,Y)) --> SHUFFLE(HOP(X,Y)).
;

define <4 x double> @hadd_4f64_v4f64_shuffle(<4 x double> %a0, <4 x double> %a1) {
; SSE-LABEL: hadd_4f64_v4f64_shuffle:
; SSE:       # %bb.0:
; SSE-NEXT:    haddpd %xmm1, %xmm0
; SSE-NEXT:    haddpd %xmm3, %xmm2
; SSE-NEXT:    movapd %xmm2, %xmm1
; SSE-NEXT:    retq
;
; AVX1-LABEL: hadd_4f64_v4f64_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm2
; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
; AVX1-NEXT:    vhaddpd %ymm0, %ymm2, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hadd_4f64_v4f64_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vhaddpd %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT:    retq
  %shuf0 = shufflevector <4 x double> %a0, <4 x double> %a1, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
  %shuf1 = shufflevector <4 x double> %a0, <4 x double> %a1, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
  %hadd0 = shufflevector <4 x double> %shuf0, <4 x double> %shuf1, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %hadd1 = shufflevector <4 x double> %shuf0, <4 x double> %shuf1, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %hadd = fadd <4 x double> %hadd0, %hadd1
  ret <4 x double> %hadd
}

define <4 x double> @hsub_4f64_v4f64_shuffle(<4 x double> %a0, <4 x double> %a1) {
; SSE-LABEL: hsub_4f64_v4f64_shuffle:
; SSE:       # %bb.0:
; SSE-NEXT:    hsubpd %xmm1, %xmm0
; SSE-NEXT:    hsubpd %xmm3, %xmm2
; SSE-NEXT:    movapd %xmm2, %xmm1
; SSE-NEXT:    retq
;
; AVX1-LABEL: hsub_4f64_v4f64_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm2
; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
; AVX1-NEXT:    vhsubpd %ymm0, %ymm2, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hsub_4f64_v4f64_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vhsubpd %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT:    retq
  %shuf0 = shufflevector <4 x double> %a0, <4 x double> %a1, <4 x i32> <i32 0, i32 1, i32 4, i32 5>
  %shuf1 = shufflevector <4 x double> %a0, <4 x double> %a1, <4 x i32> <i32 2, i32 3, i32 6, i32 7>
  %hadd0 = shufflevector <4 x double> %shuf0, <4 x double> %shuf1, <4 x i32> <i32 0, i32 4, i32 2, i32 6>
  %hadd1 = shufflevector <4 x double> %shuf0, <4 x double> %shuf1, <4 x i32> <i32 1, i32 5, i32 3, i32 7>
  %hadd = fsub <4 x double> %hadd0, %hadd1
  ret <4 x double> %hadd
}

define <8 x float> @hadd_8f32_v8f32_shuffle(<8 x float> %a0, <8 x float> %a1) {
; SSE-LABEL: hadd_8f32_v8f32_shuffle:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm1, %xmm0
; SSE-NEXT:    haddps %xmm3, %xmm2
; SSE-NEXT:    movaps %xmm2, %xmm1
; SSE-NEXT:    retq
;
; AVX1-LABEL: hadd_8f32_v8f32_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm2
; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
; AVX1-NEXT:    vhaddps %ymm0, %ymm2, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hadd_8f32_v8f32_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vhaddps %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT:    retq
  %shuf0 = shufflevector <8 x float> %a0, <8 x float> %a1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
  %shuf1 = shufflevector <8 x float> %a0, <8 x float> %a1, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 12, i32 13, i32 14, i32 15>
  %hadd0 = shufflevector <8 x float> %shuf0, <8 x float> %shuf1, <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 6, i32 12, i32 14>
  %hadd1 = shufflevector <8 x float> %shuf0, <8 x float> %shuf1, <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 7, i32 13, i32 15>
  %hadd = fadd <8 x float> %hadd0, %hadd1
  ret <8 x float> %hadd
}

define <8 x float> @hsub_8f32_v8f32_shuffle(<8 x float> %a0, <8 x float> %a1) {
; SSE-LABEL: hsub_8f32_v8f32_shuffle:
; SSE:       # %bb.0:
; SSE-NEXT:    haddps %xmm1, %xmm0
; SSE-NEXT:    haddps %xmm3, %xmm2
; SSE-NEXT:    movaps %xmm2, %xmm1
; SSE-NEXT:    retq
;
; AVX1-LABEL: hsub_8f32_v8f32_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm2
; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm0 = ymm0[2,3],ymm1[2,3]
; AVX1-NEXT:    vhaddps %ymm0, %ymm2, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hsub_8f32_v8f32_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vhaddps %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT:    retq
  %shuf0 = shufflevector <8 x float> %a0, <8 x float> %a1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
  %shuf1 = shufflevector <8 x float> %a0, <8 x float> %a1, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 12, i32 13, i32 14, i32 15>
  %hsub0 = shufflevector <8 x float> %shuf0, <8 x float> %shuf1, <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 6, i32 12, i32 14>
  %hsub1 = shufflevector <8 x float> %shuf0, <8 x float> %shuf1, <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 7, i32 13, i32 15>
  %hsub = fadd <8 x float> %hsub0, %hsub1
  ret <8 x float> %hsub
}

define <8 x i32> @hadd_8i32_v8i32_shuffle(<8 x i32> %a0, <8 x i32> %a1) {
; SSE3-LABEL: hadd_8i32_v8i32_shuffle:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movaps %xmm2, %xmm4
; SSE3-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2],xmm3[0,2]
; SSE3-NEXT:    movaps %xmm0, %xmm5
; SSE3-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2],xmm1[0,2]
; SSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,3],xmm3[1,3]
; SSE3-NEXT:    paddd %xmm2, %xmm4
; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
; SSE3-NEXT:    paddd %xmm5, %xmm0
; SSE3-NEXT:    movdqa %xmm4, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hadd_8i32_v8i32_shuffle:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddd %xmm1, %xmm0
; SSSE3-NEXT:    phaddd %xmm3, %xmm2
; SSSE3-NEXT:    movdqa %xmm2, %xmm1
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hadd_8i32_v8i32_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
; AVX1-NEXT:    vphaddd %xmm2, %xmm1, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT:    vphaddd %xmm2, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hadd_8i32_v8i32_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vphaddd %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT:    retq
  %shuf0 = shufflevector <8 x i32> %a0, <8 x i32> %a1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
  %shuf1 = shufflevector <8 x i32> %a0, <8 x i32> %a1, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 12, i32 13, i32 14, i32 15>
  %hadd0 = shufflevector <8 x i32> %shuf0, <8 x i32> %shuf1, <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 6, i32 12, i32 14>
  %hadd1 = shufflevector <8 x i32> %shuf0, <8 x i32> %shuf1, <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 7, i32 13, i32 15>
  %hadd = add <8 x i32> %hadd0, %hadd1
  ret <8 x i32> %hadd
}

define <8 x i32> @hsub_8i32_v8i32_shuffle(<8 x i32> %a0, <8 x i32> %a1) {
; SSE3-LABEL: hsub_8i32_v8i32_shuffle:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movaps %xmm2, %xmm4
; SSE3-NEXT:    shufps {{.*#+}} xmm4 = xmm4[0,2],xmm3[0,2]
; SSE3-NEXT:    movaps %xmm0, %xmm5
; SSE3-NEXT:    shufps {{.*#+}} xmm5 = xmm5[0,2],xmm1[0,2]
; SSE3-NEXT:    shufps {{.*#+}} xmm2 = xmm2[1,3],xmm3[1,3]
; SSE3-NEXT:    psubd %xmm2, %xmm4
; SSE3-NEXT:    shufps {{.*#+}} xmm0 = xmm0[1,3],xmm1[1,3]
; SSE3-NEXT:    psubd %xmm0, %xmm5
; SSE3-NEXT:    movdqa %xmm5, %xmm0
; SSE3-NEXT:    movdqa %xmm4, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hsub_8i32_v8i32_shuffle:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phsubd %xmm1, %xmm0
; SSSE3-NEXT:    phsubd %xmm3, %xmm2
; SSSE3-NEXT:    movdqa %xmm2, %xmm1
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hsub_8i32_v8i32_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
; AVX1-NEXT:    vphsubd %xmm2, %xmm1, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT:    vphsubd %xmm2, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hsub_8i32_v8i32_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vphsubd %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT:    retq
  %shuf0 = shufflevector <8 x i32> %a0, <8 x i32> %a1, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
  %shuf1 = shufflevector <8 x i32> %a0, <8 x i32> %a1, <8 x i32> <i32 4, i32 5, i32 6, i32 7, i32 12, i32 13, i32 14, i32 15>
  %hadd0 = shufflevector <8 x i32> %shuf0, <8 x i32> %shuf1, <8 x i32> <i32 0, i32 2, i32 8, i32 10, i32 4, i32 6, i32 12, i32 14>
  %hadd1 = shufflevector <8 x i32> %shuf0, <8 x i32> %shuf1, <8 x i32> <i32 1, i32 3, i32 9, i32 11, i32 5, i32 7, i32 13, i32 15>
  %hadd = sub <8 x i32> %hadd0, %hadd1
  ret <8 x i32> %hadd
}

define <16 x i16> @hadd_16i16_16i16_shuffle(<16 x i16> %a0, <16 x i16> %a1) {
; SSE3-LABEL: hadd_16i16_16i16_shuffle:
; SSE3:       # %bb.0:
; SSE3-NEXT:    movdqa %xmm3, %xmm5
; SSE3-NEXT:    pslld $16, %xmm5
; SSE3-NEXT:    psrad $16, %xmm5
; SSE3-NEXT:    movdqa %xmm2, %xmm4
; SSE3-NEXT:    pslld $16, %xmm4
; SSE3-NEXT:    psrad $16, %xmm4
; SSE3-NEXT:    packssdw %xmm5, %xmm4
; SSE3-NEXT:    movdqa %xmm1, %xmm5
; SSE3-NEXT:    pslld $16, %xmm5
; SSE3-NEXT:    psrad $16, %xmm5
; SSE3-NEXT:    movdqa %xmm0, %xmm6
; SSE3-NEXT:    pslld $16, %xmm6
; SSE3-NEXT:    psrad $16, %xmm6
; SSE3-NEXT:    packssdw %xmm5, %xmm6
; SSE3-NEXT:    psrad $16, %xmm3
; SSE3-NEXT:    psrad $16, %xmm2
; SSE3-NEXT:    packssdw %xmm3, %xmm2
; SSE3-NEXT:    paddw %xmm2, %xmm4
; SSE3-NEXT:    psrad $16, %xmm1
; SSE3-NEXT:    psrad $16, %xmm0
; SSE3-NEXT:    packssdw %xmm1, %xmm0
; SSE3-NEXT:    paddw %xmm6, %xmm0
; SSE3-NEXT:    movdqa %xmm4, %xmm1
; SSE3-NEXT:    retq
;
; SSSE3-LABEL: hadd_16i16_16i16_shuffle:
; SSSE3:       # %bb.0:
; SSSE3-NEXT:    phaddw %xmm1, %xmm0
; SSSE3-NEXT:    phaddw %xmm3, %xmm2
; SSSE3-NEXT:    movdqa %xmm2, %xmm1
; SSSE3-NEXT:    retq
;
; AVX1-LABEL: hadd_16i16_16i16_shuffle:
; AVX1:       # %bb.0:
; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
; AVX1-NEXT:    vphaddw %xmm2, %xmm1, %xmm1
; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm2
; AVX1-NEXT:    vphaddw %xmm2, %xmm0, %xmm0
; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
; AVX1-NEXT:    retq
;
; AVX2-LABEL: hadd_16i16_16i16_shuffle:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vphaddw %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,1,3]
; AVX2-NEXT:    retq
  %shuf0 = shufflevector <16 x i16> %a0, <16 x i16> %a1, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 23>
  %shuf1 = shufflevector <16 x i16> %a0, <16 x i16> %a1, <16 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15, i32 24, i32 25, i32 26, i32 27, i32 28, i32 29, i32 30, i32 31>
  %hadd0 = shufflevector <16 x i16> %shuf0, <16 x i16> %shuf1, <16 x i32> <i32 0, i32 2, i32 4, i32 6, i32 16, i32 18, i32 20, i32 22, i32 8, i32 10, i32 12, i32 14, i32 24, i32 26, i32 28, i32 30>
  %hadd1 = shufflevector <16 x i16> %shuf0, <16 x i16> %shuf1, <16 x i32> <i32 1, i32 3, i32 5, i32 7, i32 17, i32 19, i32 21, i32 23, i32 9, i32 11, i32 13, i32 15, i32 25, i32 27, i32 29, i32 31>
  %hadd = add <16 x i16> %hadd0, %hadd1
  ret <16 x i16> %hadd
}