[ORC] Add std::tuple support to SimplePackedSerialization.

[llvm-project.git] / llvm / test / CodeGen / X86 / vector-interleaved-load-i16-stride-3.ll

; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2 | FileCheck --check-prefixes=AVX2,AVX2-SLOW %s
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-crosslane-shuffle,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s
; RUN: llc < %s -mtriple=x86_64-- -mattr=+avx2,+fast-variable-perlane-shuffle | FileCheck --check-prefixes=AVX2,AVX2-FAST %s

; These patterns are produced by LoopVectorizer for interleaved loads.

define void @vf2(<6 x i16>* %in.vec, <2 x i16>* %out.vec0, <2 x i16>* %out.vec1, <2 x i16>* %out.vec2) nounwind {
; AVX2-SLOW-LABEL: vf2:
; AVX2-SLOW:       # %bb.0:
; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[0,3,2,3,4,5,6,7]
; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm2 = xmm0[0,2,2,3]
; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm2 = xmm2[1,2,2,3,4,5,6,7]
; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
; AVX2-SLOW-NEXT:    vmovd %xmm1, (%rsi)
; AVX2-SLOW-NEXT:    vmovd %xmm2, (%rdx)
; AVX2-SLOW-NEXT:    vmovd %xmm0, (%rcx)
; AVX2-SLOW-NEXT:    retq
;
; AVX2-FAST-LABEL: vf2:
; AVX2-FAST:       # %bb.0:
; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm0[0,3,2,3,4,5,6,7]
; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm2 = xmm0[2,3,8,9,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,10,11,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-FAST-NEXT:    vmovd %xmm1, (%rsi)
; AVX2-FAST-NEXT:    vmovd %xmm2, (%rdx)
; AVX2-FAST-NEXT:    vmovd %xmm0, (%rcx)
; AVX2-FAST-NEXT:    retq
  %wide.vec = load <6 x i16>, <6 x i16>* %in.vec, align 32

  %strided.vec0 = shufflevector <6 x i16> %wide.vec, <6 x i16> poison, <2 x i32> <i32 0, i32 3>
  %strided.vec1 = shufflevector <6 x i16> %wide.vec, <6 x i16> poison, <2 x i32> <i32 1, i32 4>
  %strided.vec2 = shufflevector <6 x i16> %wide.vec, <6 x i16> poison, <2 x i32> <i32 2, i32 5>

  store <2 x i16> %strided.vec0, <2 x i16>* %out.vec0, align 32
  store <2 x i16> %strided.vec1, <2 x i16>* %out.vec1, align 32
  store <2 x i16> %strided.vec2, <2 x i16>* %out.vec2, align 32

  ret void
}

define void @vf4(<12 x i16>* %in.vec, <4 x i16>* %out.vec0, <4 x i16>* %out.vec1, <4 x i16>* %out.vec2) nounwind {
; AVX2-SLOW-LABEL: vf4:
; AVX2-SLOW:       # %bb.0:
; AVX2-SLOW-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-SLOW-NEXT:    vmovdqa 16(%rdi), %xmm1
; AVX2-SLOW-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2,3,4,5,6,7]
; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[0,1,6,7,12,13,2,3,u,u,u,u,u,u,u,u]
; AVX2-SLOW-NEXT:    vpblendd {{.*#+}} xmm3 = xmm0[0],xmm1[1],xmm0[2,3]
; AVX2-SLOW-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[2,3,8,9,14,15,4,5,u,u,u,u,u,u,u,u]
; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,3,2,3,4,5,6,7]
; AVX2-SLOW-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,1,2,3]
; AVX2-SLOW-NEXT:    vpshuflw {{.*#+}} xmm0 = xmm0[2,1,2,3,4,5,6,7]
; AVX2-SLOW-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; AVX2-SLOW-NEXT:    vmovq %xmm2, (%rsi)
; AVX2-SLOW-NEXT:    vmovq %xmm3, (%rdx)
; AVX2-SLOW-NEXT:    vmovq %xmm0, (%rcx)
; AVX2-SLOW-NEXT:    retq
;
; AVX2-FAST-LABEL: vf4:
; AVX2-FAST:       # %bb.0:
; AVX2-FAST-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-FAST-NEXT:    vmovdqa 16(%rdi), %xmm1
; AVX2-FAST-NEXT:    vpblendw {{.*#+}} xmm2 = xmm0[0],xmm1[1],xmm0[2,3,4,5,6,7]
; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm2 = xmm2[0,1,6,7,12,13,2,3,u,u,u,u,u,u,u,u]
; AVX2-FAST-NEXT:    vpblendd {{.*#+}} xmm3 = xmm0[0],xmm1[1],xmm0[2,3]
; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[2,3,8,9,14,15,4,5,u,u,u,u,u,u,u,u]
; AVX2-FAST-NEXT:    vpshuflw {{.*#+}} xmm1 = xmm1[0,3,2,3,4,5,6,7]
; AVX2-FAST-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,10,11,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-FAST-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
; AVX2-FAST-NEXT:    vmovq %xmm2, (%rsi)
; AVX2-FAST-NEXT:    vmovq %xmm3, (%rdx)
; AVX2-FAST-NEXT:    vmovq %xmm0, (%rcx)
; AVX2-FAST-NEXT:    retq
  %wide.vec = load <12 x i16>, <12 x i16>* %in.vec, align 32

  %strided.vec0 = shufflevector <12 x i16> %wide.vec, <12 x i16> poison, <4 x i32> <i32 0, i32 3, i32 6, i32 9>
  %strided.vec1 = shufflevector <12 x i16> %wide.vec, <12 x i16> poison, <4 x i32> <i32 1, i32 4, i32 7, i32 10>
  %strided.vec2 = shufflevector <12 x i16> %wide.vec, <12 x i16> poison, <4 x i32> <i32 2, i32 5, i32 8, i32 11>

  store <4 x i16> %strided.vec0, <4 x i16>* %out.vec0, align 32
  store <4 x i16> %strided.vec1, <4 x i16>* %out.vec1, align 32
  store <4 x i16> %strided.vec2, <4 x i16>* %out.vec2, align 32

  ret void
}

define void @vf8(<24 x i16>* %in.vec, <8 x i16>* %out.vec0, <8 x i16>* %out.vec1, <8 x i16>* %out.vec2) nounwind {
; AVX2-LABEL: vf8:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovdqa (%rdi), %xmm0
; AVX2-NEXT:    vmovdqa 16(%rdi), %xmm1
; AVX2-NEXT:    vmovdqa 32(%rdi), %xmm2
; AVX2-NEXT:    vpblendw {{.*#+}} xmm3 = xmm0[0,1],xmm2[2],xmm0[3,4],xmm2[5],xmm0[6,7]
; AVX2-NEXT:    vpblendw {{.*#+}} xmm3 = xmm3[0],xmm1[1],xmm3[2,3],xmm1[4],xmm3[5,6],xmm1[7]
; AVX2-NEXT:    vpshufb {{.*#+}} xmm3 = xmm3[0,1,6,7,12,13,2,3,8,9,14,15,4,5,10,11]
; AVX2-NEXT:    vpblendw {{.*#+}} xmm4 = xmm2[0],xmm0[1,2],xmm2[3],xmm0[4,5],xmm2[6],xmm0[7]
; AVX2-NEXT:    vpblendw {{.*#+}} xmm4 = xmm4[0,1],xmm1[2],xmm4[3,4],xmm1[5],xmm4[6,7]
; AVX2-NEXT:    vpshufb {{.*#+}} xmm4 = xmm4[2,3,8,9,14,15,4,5,10,11,0,1,6,7,12,13]
; AVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2,3],xmm2[4],xmm0[5,6],xmm2[7]
; AVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm1[0],xmm0[1,2],xmm1[3],xmm0[4,5],xmm1[6],xmm0[7]
; AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[4,5,10,11,0,1,6,7,12,13,2,3,8,9,14,15]
; AVX2-NEXT:    vmovdqa %xmm3, (%rsi)
; AVX2-NEXT:    vmovdqa %xmm4, (%rdx)
; AVX2-NEXT:    vmovdqa %xmm0, (%rcx)
; AVX2-NEXT:    retq
  %wide.vec = load <24 x i16>, <24 x i16>* %in.vec, align 32

  %strided.vec0 = shufflevector <24 x i16> %wide.vec, <24 x i16> poison, <8 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21>
  %strided.vec1 = shufflevector <24 x i16> %wide.vec, <24 x i16> poison, <8 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22>
  %strided.vec2 = shufflevector <24 x i16> %wide.vec, <24 x i16> poison, <8 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23>

  store <8 x i16> %strided.vec0, <8 x i16>* %out.vec0, align 32
  store <8 x i16> %strided.vec1, <8 x i16>* %out.vec1, align 32
  store <8 x i16> %strided.vec2, <8 x i16>* %out.vec2, align 32

  ret void
}

define void @vf16(<48 x i16>* %in.vec, <16 x i16>* %out.vec0, <16 x i16>* %out.vec1, <16 x i16>* %out.vec2) nounwind {
; AVX2-LABEL: vf16:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovdqa (%rdi), %ymm0
; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm1
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm2 = <255,255,u,u,0,0,255,255,u,u,0,0,255,255,u,u,0,0,255,255,u,u,0,0,255,255,u,u,0,0,255,255>
; AVX2-NEXT:    vpblendvb %ymm2, %ymm0, %ymm1, %ymm2
; AVX2-NEXT:    vpermq {{.*#+}} ymm3 = ymm2[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2,3],ymm3[4],ymm2[5,6],ymm3[7],ymm2[8],ymm3[9],ymm2[10,11],ymm3[12],ymm2[13,14],ymm3[15]
; AVX2-NEXT:    vpshufb {{.*#+}} ymm2 = ymm2[0,1,6,7,12,13,2,3,4,5,14,15,8,9,10,11,16,17,22,23,28,29,18,19,20,21,30,31,24,25,26,27]
; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm3
; AVX2-NEXT:    vmovdqa 64(%rdi), %xmm4
; AVX2-NEXT:    vpblendw {{.*#+}} xmm5 = xmm4[0,1],xmm3[2],xmm4[3,4],xmm3[5],xmm4[6,7]
; AVX2-NEXT:    vpshufb {{.*#+}} xmm5 = xmm5[u,u,u,u,u,u,2,3,8,9,14,15,4,5,10,11]
; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm5
; AVX2-NEXT:    vpblendw {{.*#+}} ymm5 = ymm2[0,1,2],ymm5[3,4,5,6,7],ymm2[8,9,10],ymm5[11,12,13,14,15]
; AVX2-NEXT:    vpshufhw {{.*#+}} xmm2 = xmm2[0,1,2,3,6,5,4,7]
; AVX2-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm5[4,5,6,7]
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm5 = <255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255>
; AVX2-NEXT:    vpblendvb %ymm5, %ymm1, %ymm0, %ymm5
; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm5[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm5 = ymm5[0,1],ymm6[2],ymm5[3,4],ymm6[5],ymm5[6,7,8,9],ymm6[10],ymm5[11,12],ymm6[13],ymm5[14,15]
; AVX2-NEXT:    vpshufb {{.*#+}} ymm5 = ymm5[2,3,8,9,14,15,4,5,12,13,10,11,0,1,6,7,18,19,24,25,30,31,20,21,28,29,26,27,16,17,22,23]
; AVX2-NEXT:    vpblendw {{.*#+}} xmm6 = xmm3[0,1],xmm4[2],xmm3[3,4],xmm4[5],xmm3[6,7]
; AVX2-NEXT:    vpshufb {{.*#+}} xmm6 = xmm6[u,u,u,u,u,u,4,5,10,11,0,1,6,7,12,13]
; AVX2-NEXT:    vinserti128 $1, %xmm6, %ymm0, %ymm6
; AVX2-NEXT:    vpblendw {{.*#+}} ymm6 = ymm5[0,1,2],ymm6[3,4,5,6,7],ymm5[8,9,10],ymm6[11,12,13,14,15]
; AVX2-NEXT:    vpshufhw {{.*#+}} xmm5 = xmm5[0,1,2,3,5,6,7,4]
; AVX2-NEXT:    vpblendd {{.*#+}} ymm5 = ymm5[0,1,2,3],ymm6[4,5,6,7]
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm6 = <u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u>
; AVX2-NEXT:    vpblendvb %ymm6, %ymm1, %ymm0, %ymm0
; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm1
; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm1[0],ymm0[1,2],ymm1[3],ymm0[4,5],ymm1[6],ymm0[7],ymm1[8],ymm0[9,10],ymm1[11],ymm0[12,13],ymm1[14],ymm0[15]
; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[4,5,10,11,0,1,6,7,12,13,2,3,8,9,14,15,20,21,26,27,u,u,u,u,u,u,u,u,u,u,u,u]
; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm4[0],xmm3[1],xmm4[2,3],xmm3[4],xmm4[5,6],xmm3[7]
; AVX2-NEXT:    vpshufb {{.*#+}} xmm1 = xmm1[u,u,u,u,0,1,6,7,12,13,2,3,8,9,14,15]
; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3,4],ymm1[5,6,7]
; AVX2-NEXT:    vmovdqa %ymm2, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm5, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, (%rcx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %wide.vec = load <48 x i16>, <48 x i16>* %in.vec, align 32

  %strided.vec0 = shufflevector <48 x i16> %wide.vec, <48 x i16> poison, <16 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45>
  %strided.vec1 = shufflevector <48 x i16> %wide.vec, <48 x i16> poison, <16 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46>
  %strided.vec2 = shufflevector <48 x i16> %wide.vec, <48 x i16> poison, <16 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47>

  store <16 x i16> %strided.vec0, <16 x i16>* %out.vec0, align 32
  store <16 x i16> %strided.vec1, <16 x i16>* %out.vec1, align 32
  store <16 x i16> %strided.vec2, <16 x i16>* %out.vec2, align 32

  ret void
}

define void @vf32(<96 x i16>* %in.vec, <32 x i16>* %out.vec0, <32 x i16>* %out.vec1, <32 x i16>* %out.vec2) nounwind {
; AVX2-LABEL: vf32:
; AVX2:       # %bb.0:
; AVX2-NEXT:    vmovdqa (%rdi), %ymm11
; AVX2-NEXT:    vmovdqa 32(%rdi), %ymm15
; AVX2-NEXT:    vmovdqa 96(%rdi), %ymm3
; AVX2-NEXT:    vmovdqa 128(%rdi), %ymm4
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm9 = <255,255,u,u,0,0,255,255,u,u,0,0,255,255,u,u,0,0,255,255,u,u,0,0,255,255,u,u,0,0,255,255>
; AVX2-NEXT:    vpblendvb %ymm9, %ymm3, %ymm4, %ymm5
; AVX2-NEXT:    vpermq {{.*#+}} ymm6 = ymm5[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm5 = ymm5[0],ymm6[1],ymm5[2,3],ymm6[4],ymm5[5,6],ymm6[7],ymm5[8],ymm6[9],ymm5[10,11],ymm6[12],ymm5[13,14],ymm6[15]
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm7 = [0,1,6,7,12,13,2,3,4,5,14,15,8,9,10,11,16,17,22,23,28,29,18,19,20,21,30,31,24,25,26,27]
; AVX2-NEXT:    vpshufb %ymm7, %ymm5, %ymm8
; AVX2-NEXT:    vmovdqa 176(%rdi), %xmm5
; AVX2-NEXT:    vmovdqa 160(%rdi), %xmm6
; AVX2-NEXT:    vpblendw {{.*#+}} xmm2 = xmm6[0,1],xmm5[2],xmm6[3,4],xmm5[5],xmm6[6,7]
; AVX2-NEXT:    vmovdqa {{.*#+}} xmm10 = [4,5,14,15,0,1,2,3,8,9,14,15,4,5,10,11]
; AVX2-NEXT:    vpshufb %xmm10, %xmm2, %xmm2
; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
; AVX2-NEXT:    vpblendw {{.*#+}} ymm2 = ymm8[0,1,2],ymm2[3,4,5,6,7],ymm8[8,9,10],ymm2[11,12,13,14,15]
; AVX2-NEXT:    vpshufhw {{.*#+}} xmm8 = xmm8[0,1,2,3,6,5,4,7]
; AVX2-NEXT:    vpblendd {{.*#+}} ymm8 = ymm8[0,1,2,3],ymm2[4,5,6,7]
; AVX2-NEXT:    vpblendvb %ymm9, %ymm11, %ymm15, %ymm2
; AVX2-NEXT:    vpermq {{.*#+}} ymm9 = ymm2[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm2 = ymm2[0],ymm9[1],ymm2[2,3],ymm9[4],ymm2[5,6],ymm9[7],ymm2[8],ymm9[9],ymm2[10,11],ymm9[12],ymm2[13,14],ymm9[15]
; AVX2-NEXT:    vpshufb %ymm7, %ymm2, %ymm9
; AVX2-NEXT:    vmovdqa 80(%rdi), %xmm7
; AVX2-NEXT:    vmovdqa 64(%rdi), %xmm2
; AVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm2[0,1],xmm7[2],xmm2[3,4],xmm7[5],xmm2[6,7]
; AVX2-NEXT:    vpshufb %xmm10, %xmm0, %xmm0
; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm9[0,1,2],ymm0[3,4,5,6,7],ymm9[8,9,10],ymm0[11,12,13,14,15]
; AVX2-NEXT:    vpshufhw {{.*#+}} xmm9 = xmm9[0,1,2,3,6,5,4,7]
; AVX2-NEXT:    vpblendd {{.*#+}} ymm9 = ymm9[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm13 = <255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255>
; AVX2-NEXT:    vpblendvb %ymm13, %ymm4, %ymm3, %ymm10
; AVX2-NEXT:    vpermq {{.*#+}} ymm12 = ymm10[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm10 = ymm10[0,1],ymm12[2],ymm10[3,4],ymm12[5],ymm10[6,7,8,9],ymm12[10],ymm10[11,12],ymm12[13],ymm10[14,15]
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm12 = [2,3,8,9,14,15,4,5,12,13,10,11,0,1,6,7,18,19,24,25,30,31,20,21,28,29,26,27,16,17,22,23]
; AVX2-NEXT:    vpshufb %ymm12, %ymm10, %ymm10
; AVX2-NEXT:    vpblendw {{.*#+}} xmm0 = xmm5[0,1],xmm6[2],xmm5[3,4],xmm6[5],xmm5[6,7]
; AVX2-NEXT:    vmovdqa {{.*#+}} xmm14 = [4,5,4,5,4,5,4,5,10,11,0,1,6,7,12,13]
; AVX2-NEXT:    vpshufb %xmm14, %xmm0, %xmm0
; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm10[0,1,2],ymm0[3,4,5,6,7],ymm10[8,9,10],ymm0[11,12,13,14,15]
; AVX2-NEXT:    vpshufhw {{.*#+}} xmm10 = xmm10[0,1,2,3,5,6,7,4]
; AVX2-NEXT:    vpblendd {{.*#+}} ymm10 = ymm10[0,1,2,3],ymm0[4,5,6,7]
; AVX2-NEXT:    vpblendvb %ymm13, %ymm15, %ymm11, %ymm0
; AVX2-NEXT:    vpermq {{.*#+}} ymm13 = ymm0[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm0 = ymm0[0,1],ymm13[2],ymm0[3,4],ymm13[5],ymm0[6,7,8,9],ymm13[10],ymm0[11,12],ymm13[13],ymm0[14,15]
; AVX2-NEXT:    vpshufb %ymm12, %ymm0, %ymm0
; AVX2-NEXT:    vpblendw {{.*#+}} xmm1 = xmm7[0,1],xmm2[2],xmm7[3,4],xmm2[5],xmm7[6,7]
; AVX2-NEXT:    vpshufb %xmm14, %xmm1, %xmm1
; AVX2-NEXT:    vinserti128 $1, %xmm1, %ymm0, %ymm1
; AVX2-NEXT:    vpblendw {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3,4,5,6,7],ymm0[8,9,10],ymm1[11,12,13,14,15]
; AVX2-NEXT:    vpshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,6,7,4]
; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm1 = <u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u,255,255,0,0,u,u>
; AVX2-NEXT:    vpblendvb %ymm1, %ymm4, %ymm3, %ymm3
; AVX2-NEXT:    vpermq {{.*#+}} ymm4 = ymm3[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm3 = ymm4[0],ymm3[1,2],ymm4[3],ymm3[4,5],ymm4[6],ymm3[7],ymm4[8],ymm3[9,10],ymm4[11],ymm3[12,13],ymm4[14],ymm3[15]
; AVX2-NEXT:    vmovdqa {{.*#+}} ymm4 = [4,5,10,11,0,1,6,7,12,13,2,3,8,9,14,15,20,21,26,27,16,17,22,23,28,29,18,19,24,25,30,31]
; AVX2-NEXT:    vpshufb %ymm4, %ymm3, %ymm3
; AVX2-NEXT:    vpblendw {{.*#+}} xmm5 = xmm6[0],xmm5[1],xmm6[2,3],xmm5[4],xmm6[5,6],xmm5[7]
; AVX2-NEXT:    vmovdqa {{.*#+}} xmm6 = [0,1,2,3,0,1,6,7,12,13,2,3,8,9,14,15]
; AVX2-NEXT:    vpshufb %xmm6, %xmm5, %xmm5
; AVX2-NEXT:    vinserti128 $1, %xmm5, %ymm0, %ymm5
; AVX2-NEXT:    vpblendd {{.*#+}} ymm3 = ymm3[0,1,2,3,4],ymm5[5,6,7]
; AVX2-NEXT:    vpblendvb %ymm1, %ymm15, %ymm11, %ymm1
; AVX2-NEXT:    vpermq {{.*#+}} ymm5 = ymm1[2,3,0,1]
; AVX2-NEXT:    vpblendw {{.*#+}} ymm1 = ymm5[0],ymm1[1,2],ymm5[3],ymm1[4,5],ymm5[6],ymm1[7],ymm5[8],ymm1[9,10],ymm5[11],ymm1[12,13],ymm5[14],ymm1[15]
; AVX2-NEXT:    vpshufb %ymm4, %ymm1, %ymm1
; AVX2-NEXT:    vpblendw {{.*#+}} xmm2 = xmm2[0],xmm7[1],xmm2[2,3],xmm7[4],xmm2[5,6],xmm7[7]
; AVX2-NEXT:    vpshufb %xmm6, %xmm2, %xmm2
; AVX2-NEXT:    vinserti128 $1, %xmm2, %ymm0, %ymm2
; AVX2-NEXT:    vpblendd {{.*#+}} ymm1 = ymm1[0,1,2,3,4],ymm2[5,6,7]
; AVX2-NEXT:    vmovdqa %ymm9, (%rsi)
; AVX2-NEXT:    vmovdqa %ymm8, 32(%rsi)
; AVX2-NEXT:    vmovdqa %ymm10, 32(%rdx)
; AVX2-NEXT:    vmovdqa %ymm0, (%rdx)
; AVX2-NEXT:    vmovdqa %ymm3, 32(%rcx)
; AVX2-NEXT:    vmovdqa %ymm1, (%rcx)
; AVX2-NEXT:    vzeroupper
; AVX2-NEXT:    retq
  %wide.vec = load <96 x i16>, <96 x i16>* %in.vec, align 32

  %strided.vec0 = shufflevector <96 x i16> %wide.vec, <96 x i16> poison, <32 x i32> <i32 0, i32 3, i32 6, i32 9, i32 12, i32 15, i32 18, i32 21, i32 24, i32 27, i32 30, i32 33, i32 36, i32 39, i32 42, i32 45, i32 48, i32 51, i32 54, i32 57, i32 60, i32 63, i32 66, i32 69, i32 72, i32 75, i32 78, i32 81, i32 84, i32 87, i32 90, i32 93>
  %strided.vec1 = shufflevector <96 x i16> %wide.vec, <96 x i16> poison, <32 x i32> <i32 1, i32 4, i32 7, i32 10, i32 13, i32 16, i32 19, i32 22, i32 25, i32 28, i32 31, i32 34, i32 37, i32 40, i32 43, i32 46, i32 49, i32 52, i32 55, i32 58, i32 61, i32 64, i32 67, i32 70, i32 73, i32 76, i32 79, i32 82, i32 85, i32 88, i32 91, i32 94>
  %strided.vec2 = shufflevector <96 x i16> %wide.vec, <96 x i16> poison, <32 x i32> <i32 2, i32 5, i32 8, i32 11, i32 14, i32 17, i32 20, i32 23, i32 26, i32 29, i32 32, i32 35, i32 38, i32 41, i32 44, i32 47, i32 50, i32 53, i32 56, i32 59, i32 62, i32 65, i32 68, i32 71, i32 74, i32 77, i32 80, i32 83, i32 86, i32 89, i32 92, i32 95>

  store <32 x i16> %strided.vec0, <32 x i16>* %out.vec0, align 32
  store <32 x i16> %strided.vec1, <32 x i16>* %out.vec1, align 32
  store <32 x i16> %strided.vec2, <32 x i16>* %out.vec2, align 32

  ret void
}