Home | History | Annotate | Download | only in X86 
      1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
      2 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+xop | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,XOP
      3 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,AVX1
      4 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,INT256,AVX2
      5 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,INT256,AVX512,AVX512F
      6 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,INT256,AVX512,AVX512DQ
      7 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,INT256,AVX512,AVX512BW
      8 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512vbmi | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,INT256,AVX512,VBMI
      9 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,INT256,AVX512VL
     10 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefixes=AVX,AVXNOVLBW,INT256,AVX512VL,AVX512VLDQ
     11 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefixes=AVX,INT256,AVX512VL,AVX512VLBW
     12 ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw,+avx512vl,+avx512vbmi | FileCheck %s --check-prefixes=AVX,INT256,AVX512VL,VLVBMI
     13 
     14 define <4 x i64> @var_shuffle_v4i64(<4 x i64> %v, <4 x i64> %indices) nounwind {
     15 ; XOP-LABEL: var_shuffle_v4i64:
     16 ; XOP:       # %bb.0:
     17 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
     18 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
     19 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm3
     20 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm1
     21 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
     22 ; XOP-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
     23 ; XOP-NEXT:    vpermil2pd $0, %ymm1, %ymm2, %ymm0, %ymm0
     24 ; XOP-NEXT:    retq
     25 ;
     26 ; AVX1-LABEL: var_shuffle_v4i64:
     27 ; AVX1:       # %bb.0:
     28 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [2,2,2,2]
     29 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
     30 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
     31 ; AVX1-NEXT:    vpaddq %xmm4, %xmm4, %xmm4
     32 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
     33 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
     34 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm1, %xmm2
     35 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
     36 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
     37 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
     38 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
     39 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
     40 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
     41 ; AVX1-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
     42 ; AVX1-NEXT:    retq
     43 ;
     44 ; AVX2-LABEL: var_shuffle_v4i64:
     45 ; AVX2:       # %bb.0:
     46 ; AVX2-NEXT:    vpaddq %ymm1, %ymm1, %ymm1
     47 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
     48 ; AVX2-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm2
     49 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm0[2,3,2,3]
     50 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
     51 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
     52 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
     53 ; AVX2-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
     54 ; AVX2-NEXT:    retq
     55 ;
     56 ; AVX512-LABEL: var_shuffle_v4i64:
     57 ; AVX512:       # %bb.0:
     58 ; AVX512-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
     59 ; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
     60 ; AVX512-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
     61 ; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
     62 ; AVX512-NEXT:    retq
     63 ;
     64 ; AVX512VL-LABEL: var_shuffle_v4i64:
     65 ; AVX512VL:       # %bb.0:
     66 ; AVX512VL-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
     67 ; AVX512VL-NEXT:    retq
     68   %index0 = extractelement <4 x i64> %indices, i32 0
     69   %index1 = extractelement <4 x i64> %indices, i32 1
     70   %index2 = extractelement <4 x i64> %indices, i32 2
     71   %index3 = extractelement <4 x i64> %indices, i32 3
     72   %v0 = extractelement <4 x i64> %v, i64 %index0
     73   %v1 = extractelement <4 x i64> %v, i64 %index1
     74   %v2 = extractelement <4 x i64> %v, i64 %index2
     75   %v3 = extractelement <4 x i64> %v, i64 %index3
     76   %ret0 = insertelement <4 x i64> undef, i64 %v0, i32 0
     77   %ret1 = insertelement <4 x i64> %ret0, i64 %v1, i32 1
     78   %ret2 = insertelement <4 x i64> %ret1, i64 %v2, i32 2
     79   %ret3 = insertelement <4 x i64> %ret2, i64 %v3, i32 3
     80   ret <4 x i64> %ret3
     81 }
     82 
     83 define <8 x i32> @var_shuffle_v8i32(<8 x i32> %v, <8 x i32> %indices) nounwind {
     84 ; XOP-LABEL: var_shuffle_v8i32:
     85 ; XOP:       # %bb.0:
     86 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
     87 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
     88 ; XOP-NEXT:    vpermil2ps $0, %ymm1, %ymm2, %ymm0, %ymm0
     89 ; XOP-NEXT:    retq
     90 ;
     91 ; AVX1-LABEL: var_shuffle_v8i32:
     92 ; AVX1:       # %bb.0:
     93 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,3,3,3,3,3,3,3]
     94 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
     95 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
     96 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm4, %xmm3
     97 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm1, %xmm2
     98 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
     99 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
    100 ; AVX1-NEXT:    vpermilps %ymm1, %ymm3, %ymm3
    101 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    102 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
    103 ; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
    104 ; AVX1-NEXT:    retq
    105 ;
    106 ; INT256-LABEL: var_shuffle_v8i32:
    107 ; INT256:       # %bb.0:
    108 ; INT256-NEXT:    vpermps %ymm0, %ymm1, %ymm0
    109 ; INT256-NEXT:    retq
    110   %index0 = extractelement <8 x i32> %indices, i32 0
    111   %index1 = extractelement <8 x i32> %indices, i32 1
    112   %index2 = extractelement <8 x i32> %indices, i32 2
    113   %index3 = extractelement <8 x i32> %indices, i32 3
    114   %index4 = extractelement <8 x i32> %indices, i32 4
    115   %index5 = extractelement <8 x i32> %indices, i32 5
    116   %index6 = extractelement <8 x i32> %indices, i32 6
    117   %index7 = extractelement <8 x i32> %indices, i32 7
    118   %v0 = extractelement <8 x i32> %v, i32 %index0
    119   %v1 = extractelement <8 x i32> %v, i32 %index1
    120   %v2 = extractelement <8 x i32> %v, i32 %index2
    121   %v3 = extractelement <8 x i32> %v, i32 %index3
    122   %v4 = extractelement <8 x i32> %v, i32 %index4
    123   %v5 = extractelement <8 x i32> %v, i32 %index5
    124   %v6 = extractelement <8 x i32> %v, i32 %index6
    125   %v7 = extractelement <8 x i32> %v, i32 %index7
    126   %ret0 = insertelement <8 x i32> undef, i32 %v0, i32 0
    127   %ret1 = insertelement <8 x i32> %ret0, i32 %v1, i32 1
    128   %ret2 = insertelement <8 x i32> %ret1, i32 %v2, i32 2
    129   %ret3 = insertelement <8 x i32> %ret2, i32 %v3, i32 3
    130   %ret4 = insertelement <8 x i32> %ret3, i32 %v4, i32 4
    131   %ret5 = insertelement <8 x i32> %ret4, i32 %v5, i32 5
    132   %ret6 = insertelement <8 x i32> %ret5, i32 %v6, i32 6
    133   %ret7 = insertelement <8 x i32> %ret6, i32 %v7, i32 7
    134   ret <8 x i32> %ret7
    135 }
    136 
    137 define <16 x i16> @var_shuffle_v16i16(<16 x i16> %v, <16 x i16> %indices) nounwind {
    138 ; XOP-LABEL: var_shuffle_v16i16:
    139 ; XOP:       # %bb.0:
    140 ; XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [256,256,256,256,256,256,256,256]
    141 ; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [514,514,514,514,514,514,514,514]
    142 ; XOP-NEXT:    vpmacsww %xmm2, %xmm3, %xmm1, %xmm4
    143 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm1
    144 ; XOP-NEXT:    vpmacsww %xmm2, %xmm3, %xmm1, %xmm1
    145 ; XOP-NEXT:    vextractf128 $1, %ymm0, %xmm2
    146 ; XOP-NEXT:    vpperm %xmm1, %xmm2, %xmm0, %xmm1
    147 ; XOP-NEXT:    vpperm %xmm4, %xmm2, %xmm0, %xmm0
    148 ; XOP-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
    149 ; XOP-NEXT:    retq
    150 ;
    151 ; AVX1-LABEL: var_shuffle_v16i16:
    152 ; AVX1:       # %bb.0:
    153 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [514,514,514,514,514,514,514,514]
    154 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm3
    155 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [256,256,256,256,256,256,256,256]
    156 ; AVX1-NEXT:    vpaddw %xmm4, %xmm3, %xmm3
    157 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
    158 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
    159 ; AVX1-NEXT:    vpaddw %xmm4, %xmm1, %xmm1
    160 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
    161 ; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm1, %xmm4
    162 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
    163 ; AVX1-NEXT:    vpshufb %xmm1, %xmm5, %xmm6
    164 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
    165 ; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm1, %xmm1
    166 ; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm3, %xmm2
    167 ; AVX1-NEXT:    vpshufb %xmm3, %xmm5, %xmm4
    168 ; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
    169 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
    170 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
    171 ; AVX1-NEXT:    retq
    172 ;
    173 ; AVX2-LABEL: var_shuffle_v16i16:
    174 ; AVX2:       # %bb.0:
    175 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
    176 ; AVX2-NEXT:    vpaddw {{.*}}(%rip), %ymm1, %ymm1
    177 ; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
    178 ; AVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
    179 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
    180 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
    181 ; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    182 ; AVX2-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    183 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
    184 ; AVX2-NEXT:    retq
    185 ;
    186 ; AVX512-LABEL: var_shuffle_v16i16:
    187 ; AVX512:       # %bb.0:
    188 ; AVX512-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
    189 ; AVX512-NEXT:    vpaddw {{.*}}(%rip), %ymm1, %ymm1
    190 ; AVX512-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
    191 ; AVX512-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
    192 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm3
    193 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
    194 ; AVX512-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    195 ; AVX512-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    196 ; AVX512-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
    197 ; AVX512-NEXT:    retq
    198 ;
    199 ; AVX512VLDQ-LABEL: var_shuffle_v16i16:
    200 ; AVX512VLDQ:       # %bb.0:
    201 ; AVX512VLDQ-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
    202 ; AVX512VLDQ-NEXT:    vpaddw {{.*}}(%rip), %ymm1, %ymm1
    203 ; AVX512VLDQ-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
    204 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
    205 ; AVX512VLDQ-NEXT:    vextracti128 $1, %ymm0, %xmm3
    206 ; AVX512VLDQ-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
    207 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    208 ; AVX512VLDQ-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    209 ; AVX512VLDQ-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
    210 ; AVX512VLDQ-NEXT:    retq
    211 ;
    212 ; AVX512VLBW-LABEL: var_shuffle_v16i16:
    213 ; AVX512VLBW:       # %bb.0:
    214 ; AVX512VLBW-NEXT:    vpermw %ymm0, %ymm1, %ymm0
    215 ; AVX512VLBW-NEXT:    retq
    216 ;
    217 ; VLVBMI-LABEL: var_shuffle_v16i16:
    218 ; VLVBMI:       # %bb.0:
    219 ; VLVBMI-NEXT:    vpermw %ymm0, %ymm1, %ymm0
    220 ; VLVBMI-NEXT:    retq
    221   %index0 = extractelement <16 x i16> %indices, i32 0
    222   %index1 = extractelement <16 x i16> %indices, i32 1
    223   %index2 = extractelement <16 x i16> %indices, i32 2
    224   %index3 = extractelement <16 x i16> %indices, i32 3
    225   %index4 = extractelement <16 x i16> %indices, i32 4
    226   %index5 = extractelement <16 x i16> %indices, i32 5
    227   %index6 = extractelement <16 x i16> %indices, i32 6
    228   %index7 = extractelement <16 x i16> %indices, i32 7
    229   %index8 = extractelement <16 x i16> %indices, i32 8
    230   %index9 = extractelement <16 x i16> %indices, i32 9
    231   %index10 = extractelement <16 x i16> %indices, i32 10
    232   %index11 = extractelement <16 x i16> %indices, i32 11
    233   %index12 = extractelement <16 x i16> %indices, i32 12
    234   %index13 = extractelement <16 x i16> %indices, i32 13
    235   %index14 = extractelement <16 x i16> %indices, i32 14
    236   %index15 = extractelement <16 x i16> %indices, i32 15
    237   %v0 = extractelement <16 x i16> %v, i16 %index0
    238   %v1 = extractelement <16 x i16> %v, i16 %index1
    239   %v2 = extractelement <16 x i16> %v, i16 %index2
    240   %v3 = extractelement <16 x i16> %v, i16 %index3
    241   %v4 = extractelement <16 x i16> %v, i16 %index4
    242   %v5 = extractelement <16 x i16> %v, i16 %index5
    243   %v6 = extractelement <16 x i16> %v, i16 %index6
    244   %v7 = extractelement <16 x i16> %v, i16 %index7
    245   %v8 = extractelement <16 x i16> %v, i16 %index8
    246   %v9 = extractelement <16 x i16> %v, i16 %index9
    247   %v10 = extractelement <16 x i16> %v, i16 %index10
    248   %v11 = extractelement <16 x i16> %v, i16 %index11
    249   %v12 = extractelement <16 x i16> %v, i16 %index12
    250   %v13 = extractelement <16 x i16> %v, i16 %index13
    251   %v14 = extractelement <16 x i16> %v, i16 %index14
    252   %v15 = extractelement <16 x i16> %v, i16 %index15
    253   %ret0 = insertelement <16 x i16> undef, i16 %v0, i32 0
    254   %ret1 = insertelement <16 x i16> %ret0, i16 %v1, i32 1
    255   %ret2 = insertelement <16 x i16> %ret1, i16 %v2, i32 2
    256   %ret3 = insertelement <16 x i16> %ret2, i16 %v3, i32 3
    257   %ret4 = insertelement <16 x i16> %ret3, i16 %v4, i32 4
    258   %ret5 = insertelement <16 x i16> %ret4, i16 %v5, i32 5
    259   %ret6 = insertelement <16 x i16> %ret5, i16 %v6, i32 6
    260   %ret7 = insertelement <16 x i16> %ret6, i16 %v7, i32 7
    261   %ret8 = insertelement <16 x i16> %ret7, i16 %v8, i32 8
    262   %ret9 = insertelement <16 x i16> %ret8, i16 %v9, i32 9
    263   %ret10 = insertelement <16 x i16> %ret9, i16 %v10, i32 10
    264   %ret11 = insertelement <16 x i16> %ret10, i16 %v11, i32 11
    265   %ret12 = insertelement <16 x i16> %ret11, i16 %v12, i32 12
    266   %ret13 = insertelement <16 x i16> %ret12, i16 %v13, i32 13
    267   %ret14 = insertelement <16 x i16> %ret13, i16 %v14, i32 14
    268   %ret15 = insertelement <16 x i16> %ret14, i16 %v15, i32 15
    269   ret <16 x i16> %ret15
    270 }
    271 
    272 define <32 x i8> @var_shuffle_v32i8(<32 x i8> %v, <32 x i8> %indices) nounwind {
    273 ; XOP-LABEL: var_shuffle_v32i8:
    274 ; XOP:       # %bb.0:
    275 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm2
    276 ; XOP-NEXT:    vextractf128 $1, %ymm0, %xmm3
    277 ; XOP-NEXT:    vpperm %xmm2, %xmm3, %xmm0, %xmm2
    278 ; XOP-NEXT:    vpperm %xmm1, %xmm3, %xmm0, %xmm0
    279 ; XOP-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
    280 ; XOP-NEXT:    retq
    281 ;
    282 ; AVX1-LABEL: var_shuffle_v32i8:
    283 ; AVX1:       # %bb.0:
    284 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
    285 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
    286 ; AVX1-NEXT:    vpcmpgtb %xmm3, %xmm2, %xmm4
    287 ; AVX1-NEXT:    vextractf128 $1, %ymm0, %xmm5
    288 ; AVX1-NEXT:    vpshufb %xmm2, %xmm5, %xmm6
    289 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
    290 ; AVX1-NEXT:    vpblendvb %xmm4, %xmm6, %xmm2, %xmm2
    291 ; AVX1-NEXT:    vpcmpgtb %xmm3, %xmm1, %xmm3
    292 ; AVX1-NEXT:    vpshufb %xmm1, %xmm5, %xmm4
    293 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
    294 ; AVX1-NEXT:    vpblendvb %xmm3, %xmm4, %xmm0, %xmm0
    295 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
    296 ; AVX1-NEXT:    retq
    297 ;
    298 ; AVX2-LABEL: var_shuffle_v32i8:
    299 ; AVX2:       # %bb.0:
    300 ; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
    301 ; AVX2-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
    302 ; AVX2-NEXT:    vextracti128 $1, %ymm0, %xmm3
    303 ; AVX2-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
    304 ; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    305 ; AVX2-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    306 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
    307 ; AVX2-NEXT:    retq
    308 ;
    309 ; AVX512-LABEL: var_shuffle_v32i8:
    310 ; AVX512:       # %bb.0:
    311 ; AVX512-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
    312 ; AVX512-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
    313 ; AVX512-NEXT:    vextracti128 $1, %ymm0, %xmm3
    314 ; AVX512-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
    315 ; AVX512-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    316 ; AVX512-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    317 ; AVX512-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
    318 ; AVX512-NEXT:    retq
    319 ;
    320 ; AVX512VLDQ-LABEL: var_shuffle_v32i8:
    321 ; AVX512VLDQ:       # %bb.0:
    322 ; AVX512VLDQ-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
    323 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
    324 ; AVX512VLDQ-NEXT:    vextracti128 $1, %ymm0, %xmm3
    325 ; AVX512VLDQ-NEXT:    vpblendd {{.*#+}} ymm0 = ymm3[0,1,2,3],ymm0[4,5,6,7]
    326 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    327 ; AVX512VLDQ-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    328 ; AVX512VLDQ-NEXT:    vpblendvb %ymm1, %ymm0, %ymm2, %ymm0
    329 ; AVX512VLDQ-NEXT:    retq
    330 ;
    331 ; AVX512VLBW-LABEL: var_shuffle_v32i8:
    332 ; AVX512VLBW:       # %bb.0:
    333 ; AVX512VLBW-NEXT:    vextracti128 $1, %ymm0, %xmm2
    334 ; AVX512VLBW-NEXT:    vpblendd {{.*#+}} ymm2 = ymm2[0,1,2,3],ymm0[4,5,6,7]
    335 ; AVX512VLBW-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
    336 ; AVX512VLBW-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    337 ; AVX512VLBW-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %k1
    338 ; AVX512VLBW-NEXT:    vpshufb %ymm1, %ymm2, %ymm0 {%k1}
    339 ; AVX512VLBW-NEXT:    retq
    340 ;
    341 ; VLVBMI-LABEL: var_shuffle_v32i8:
    342 ; VLVBMI:       # %bb.0:
    343 ; VLVBMI-NEXT:    vpermb %ymm0, %ymm1, %ymm0
    344 ; VLVBMI-NEXT:    retq
    345   %index0 = extractelement <32 x i8> %indices, i32 0
    346   %index1 = extractelement <32 x i8> %indices, i32 1
    347   %index2 = extractelement <32 x i8> %indices, i32 2
    348   %index3 = extractelement <32 x i8> %indices, i32 3
    349   %index4 = extractelement <32 x i8> %indices, i32 4
    350   %index5 = extractelement <32 x i8> %indices, i32 5
    351   %index6 = extractelement <32 x i8> %indices, i32 6
    352   %index7 = extractelement <32 x i8> %indices, i32 7
    353   %index8 = extractelement <32 x i8> %indices, i32 8
    354   %index9 = extractelement <32 x i8> %indices, i32 9
    355   %index10 = extractelement <32 x i8> %indices, i32 10
    356   %index11 = extractelement <32 x i8> %indices, i32 11
    357   %index12 = extractelement <32 x i8> %indices, i32 12
    358   %index13 = extractelement <32 x i8> %indices, i32 13
    359   %index14 = extractelement <32 x i8> %indices, i32 14
    360   %index15 = extractelement <32 x i8> %indices, i32 15
    361   %index16 = extractelement <32 x i8> %indices, i32 16
    362   %index17 = extractelement <32 x i8> %indices, i32 17
    363   %index18 = extractelement <32 x i8> %indices, i32 18
    364   %index19 = extractelement <32 x i8> %indices, i32 19
    365   %index20 = extractelement <32 x i8> %indices, i32 20
    366   %index21 = extractelement <32 x i8> %indices, i32 21
    367   %index22 = extractelement <32 x i8> %indices, i32 22
    368   %index23 = extractelement <32 x i8> %indices, i32 23
    369   %index24 = extractelement <32 x i8> %indices, i32 24
    370   %index25 = extractelement <32 x i8> %indices, i32 25
    371   %index26 = extractelement <32 x i8> %indices, i32 26
    372   %index27 = extractelement <32 x i8> %indices, i32 27
    373   %index28 = extractelement <32 x i8> %indices, i32 28
    374   %index29 = extractelement <32 x i8> %indices, i32 29
    375   %index30 = extractelement <32 x i8> %indices, i32 30
    376   %index31 = extractelement <32 x i8> %indices, i32 31
    377   %v0 = extractelement <32 x i8> %v, i8 %index0
    378   %v1 = extractelement <32 x i8> %v, i8 %index1
    379   %v2 = extractelement <32 x i8> %v, i8 %index2
    380   %v3 = extractelement <32 x i8> %v, i8 %index3
    381   %v4 = extractelement <32 x i8> %v, i8 %index4
    382   %v5 = extractelement <32 x i8> %v, i8 %index5
    383   %v6 = extractelement <32 x i8> %v, i8 %index6
    384   %v7 = extractelement <32 x i8> %v, i8 %index7
    385   %v8 = extractelement <32 x i8> %v, i8 %index8
    386   %v9 = extractelement <32 x i8> %v, i8 %index9
    387   %v10 = extractelement <32 x i8> %v, i8 %index10
    388   %v11 = extractelement <32 x i8> %v, i8 %index11
    389   %v12 = extractelement <32 x i8> %v, i8 %index12
    390   %v13 = extractelement <32 x i8> %v, i8 %index13
    391   %v14 = extractelement <32 x i8> %v, i8 %index14
    392   %v15 = extractelement <32 x i8> %v, i8 %index15
    393   %v16 = extractelement <32 x i8> %v, i8 %index16
    394   %v17 = extractelement <32 x i8> %v, i8 %index17
    395   %v18 = extractelement <32 x i8> %v, i8 %index18
    396   %v19 = extractelement <32 x i8> %v, i8 %index19
    397   %v20 = extractelement <32 x i8> %v, i8 %index20
    398   %v21 = extractelement <32 x i8> %v, i8 %index21
    399   %v22 = extractelement <32 x i8> %v, i8 %index22
    400   %v23 = extractelement <32 x i8> %v, i8 %index23
    401   %v24 = extractelement <32 x i8> %v, i8 %index24
    402   %v25 = extractelement <32 x i8> %v, i8 %index25
    403   %v26 = extractelement <32 x i8> %v, i8 %index26
    404   %v27 = extractelement <32 x i8> %v, i8 %index27
    405   %v28 = extractelement <32 x i8> %v, i8 %index28
    406   %v29 = extractelement <32 x i8> %v, i8 %index29
    407   %v30 = extractelement <32 x i8> %v, i8 %index30
    408   %v31 = extractelement <32 x i8> %v, i8 %index31
    409   %ret0 = insertelement <32 x i8> undef, i8 %v0, i32 0
    410   %ret1 = insertelement <32 x i8> %ret0, i8 %v1, i32 1
    411   %ret2 = insertelement <32 x i8> %ret1, i8 %v2, i32 2
    412   %ret3 = insertelement <32 x i8> %ret2, i8 %v3, i32 3
    413   %ret4 = insertelement <32 x i8> %ret3, i8 %v4, i32 4
    414   %ret5 = insertelement <32 x i8> %ret4, i8 %v5, i32 5
    415   %ret6 = insertelement <32 x i8> %ret5, i8 %v6, i32 6
    416   %ret7 = insertelement <32 x i8> %ret6, i8 %v7, i32 7
    417   %ret8 = insertelement <32 x i8> %ret7, i8 %v8, i32 8
    418   %ret9 = insertelement <32 x i8> %ret8, i8 %v9, i32 9
    419   %ret10 = insertelement <32 x i8> %ret9, i8 %v10, i32 10
    420   %ret11 = insertelement <32 x i8> %ret10, i8 %v11, i32 11
    421   %ret12 = insertelement <32 x i8> %ret11, i8 %v12, i32 12
    422   %ret13 = insertelement <32 x i8> %ret12, i8 %v13, i32 13
    423   %ret14 = insertelement <32 x i8> %ret13, i8 %v14, i32 14
    424   %ret15 = insertelement <32 x i8> %ret14, i8 %v15, i32 15
    425   %ret16 = insertelement <32 x i8> %ret15, i8 %v16, i32 16
    426   %ret17 = insertelement <32 x i8> %ret16, i8 %v17, i32 17
    427   %ret18 = insertelement <32 x i8> %ret17, i8 %v18, i32 18
    428   %ret19 = insertelement <32 x i8> %ret18, i8 %v19, i32 19
    429   %ret20 = insertelement <32 x i8> %ret19, i8 %v20, i32 20
    430   %ret21 = insertelement <32 x i8> %ret20, i8 %v21, i32 21
    431   %ret22 = insertelement <32 x i8> %ret21, i8 %v22, i32 22
    432   %ret23 = insertelement <32 x i8> %ret22, i8 %v23, i32 23
    433   %ret24 = insertelement <32 x i8> %ret23, i8 %v24, i32 24
    434   %ret25 = insertelement <32 x i8> %ret24, i8 %v25, i32 25
    435   %ret26 = insertelement <32 x i8> %ret25, i8 %v26, i32 26
    436   %ret27 = insertelement <32 x i8> %ret26, i8 %v27, i32 27
    437   %ret28 = insertelement <32 x i8> %ret27, i8 %v28, i32 28
    438   %ret29 = insertelement <32 x i8> %ret28, i8 %v29, i32 29
    439   %ret30 = insertelement <32 x i8> %ret29, i8 %v30, i32 30
    440   %ret31 = insertelement <32 x i8> %ret30, i8 %v31, i32 31
    441   ret <32 x i8> %ret31
    442 }
    443 
    444 define <4 x double> @var_shuffle_v4f64(<4 x double> %v, <4 x i64> %indices) nounwind {
    445 ; XOP-LABEL: var_shuffle_v4f64:
    446 ; XOP:       # %bb.0:
    447 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
    448 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    449 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm3
    450 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm1
    451 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
    452 ; XOP-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
    453 ; XOP-NEXT:    vpermil2pd $0, %ymm1, %ymm2, %ymm0, %ymm0
    454 ; XOP-NEXT:    retq
    455 ;
    456 ; AVX1-LABEL: var_shuffle_v4f64:
    457 ; AVX1:       # %bb.0:
    458 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [2,2,2,2]
    459 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
    460 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
    461 ; AVX1-NEXT:    vpaddq %xmm4, %xmm4, %xmm4
    462 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
    463 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
    464 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm1, %xmm2
    465 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
    466 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
    467 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
    468 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
    469 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    470 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
    471 ; AVX1-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
    472 ; AVX1-NEXT:    retq
    473 ;
    474 ; AVX2-LABEL: var_shuffle_v4f64:
    475 ; AVX2:       # %bb.0:
    476 ; AVX2-NEXT:    vpaddq %ymm1, %ymm1, %ymm1
    477 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
    478 ; AVX2-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm2
    479 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm0[2,3,2,3]
    480 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
    481 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
    482 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
    483 ; AVX2-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
    484 ; AVX2-NEXT:    retq
    485 ;
    486 ; AVX512-LABEL: var_shuffle_v4f64:
    487 ; AVX512:       # %bb.0:
    488 ; AVX512-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
    489 ; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 def $zmm0
    490 ; AVX512-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
    491 ; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
    492 ; AVX512-NEXT:    retq
    493 ;
    494 ; AVX512VL-LABEL: var_shuffle_v4f64:
    495 ; AVX512VL:       # %bb.0:
    496 ; AVX512VL-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
    497 ; AVX512VL-NEXT:    retq
    498   %index0 = extractelement <4 x i64> %indices, i32 0
    499   %index1 = extractelement <4 x i64> %indices, i32 1
    500   %index2 = extractelement <4 x i64> %indices, i32 2
    501   %index3 = extractelement <4 x i64> %indices, i32 3
    502   %v0 = extractelement <4 x double> %v, i64 %index0
    503   %v1 = extractelement <4 x double> %v, i64 %index1
    504   %v2 = extractelement <4 x double> %v, i64 %index2
    505   %v3 = extractelement <4 x double> %v, i64 %index3
    506   %ret0 = insertelement <4 x double> undef, double %v0, i32 0
    507   %ret1 = insertelement <4 x double> %ret0, double %v1, i32 1
    508   %ret2 = insertelement <4 x double> %ret1, double %v2, i32 2
    509   %ret3 = insertelement <4 x double> %ret2, double %v3, i32 3
    510   ret <4 x double> %ret3
    511 }
    512 
    513 define <8 x float> @var_shuffle_v8f32(<8 x float> %v, <8 x i32> %indices) nounwind {
    514 ; XOP-LABEL: var_shuffle_v8f32:
    515 ; XOP:       # %bb.0:
    516 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
    517 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    518 ; XOP-NEXT:    vpermil2ps $0, %ymm1, %ymm2, %ymm0, %ymm0
    519 ; XOP-NEXT:    retq
    520 ;
    521 ; AVX1-LABEL: var_shuffle_v8f32:
    522 ; AVX1:       # %bb.0:
    523 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,3,3,3,3,3,3,3]
    524 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
    525 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
    526 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm4, %xmm3
    527 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm1, %xmm2
    528 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
    529 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
    530 ; AVX1-NEXT:    vpermilps %ymm1, %ymm3, %ymm3
    531 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    532 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
    533 ; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
    534 ; AVX1-NEXT:    retq
    535 ;
    536 ; INT256-LABEL: var_shuffle_v8f32:
    537 ; INT256:       # %bb.0:
    538 ; INT256-NEXT:    vpermps %ymm0, %ymm1, %ymm0
    539 ; INT256-NEXT:    retq
    540   %index0 = extractelement <8 x i32> %indices, i32 0
    541   %index1 = extractelement <8 x i32> %indices, i32 1
    542   %index2 = extractelement <8 x i32> %indices, i32 2
    543   %index3 = extractelement <8 x i32> %indices, i32 3
    544   %index4 = extractelement <8 x i32> %indices, i32 4
    545   %index5 = extractelement <8 x i32> %indices, i32 5
    546   %index6 = extractelement <8 x i32> %indices, i32 6
    547   %index7 = extractelement <8 x i32> %indices, i32 7
    548   %v0 = extractelement <8 x float> %v, i32 %index0
    549   %v1 = extractelement <8 x float> %v, i32 %index1
    550   %v2 = extractelement <8 x float> %v, i32 %index2
    551   %v3 = extractelement <8 x float> %v, i32 %index3
    552   %v4 = extractelement <8 x float> %v, i32 %index4
    553   %v5 = extractelement <8 x float> %v, i32 %index5
    554   %v6 = extractelement <8 x float> %v, i32 %index6
    555   %v7 = extractelement <8 x float> %v, i32 %index7
    556   %ret0 = insertelement <8 x float> undef, float %v0, i32 0
    557   %ret1 = insertelement <8 x float> %ret0, float %v1, i32 1
    558   %ret2 = insertelement <8 x float> %ret1, float %v2, i32 2
    559   %ret3 = insertelement <8 x float> %ret2, float %v3, i32 3
    560   %ret4 = insertelement <8 x float> %ret3, float %v4, i32 4
    561   %ret5 = insertelement <8 x float> %ret4, float %v5, i32 5
    562   %ret6 = insertelement <8 x float> %ret5, float %v6, i32 6
    563   %ret7 = insertelement <8 x float> %ret6, float %v7, i32 7
    564   ret <8 x float> %ret7
    565 }
    566 
    567 ;
    568 ; PR35820 - Unequal source/destination vector sizes
    569 ;
    570 
    571 define <4 x i64> @var_shuffle_v4i64_from_v2i64(<2 x i64> %v, <4 x i64> %indices) nounwind {
    572 ; XOP-LABEL: var_shuffle_v4i64_from_v2i64:
    573 ; XOP:       # %bb.0:
    574 ; XOP-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    575 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
    576 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    577 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm3
    578 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm1
    579 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
    580 ; XOP-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
    581 ; XOP-NEXT:    vpermil2pd $0, %ymm1, %ymm2, %ymm0, %ymm0
    582 ; XOP-NEXT:    retq
    583 ;
    584 ; AVX1-LABEL: var_shuffle_v4i64_from_v2i64:
    585 ; AVX1:       # %bb.0:
    586 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    587 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [2,2,2,2]
    588 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
    589 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
    590 ; AVX1-NEXT:    vpaddq %xmm4, %xmm4, %xmm4
    591 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
    592 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
    593 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm1, %xmm2
    594 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
    595 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
    596 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
    597 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
    598 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    599 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
    600 ; AVX1-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
    601 ; AVX1-NEXT:    retq
    602 ;
    603 ; AVX2-LABEL: var_shuffle_v4i64_from_v2i64:
    604 ; AVX2:       # %bb.0:
    605 ; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    606 ; AVX2-NEXT:    vpaddq %ymm1, %ymm1, %ymm1
    607 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
    608 ; AVX2-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm2
    609 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm0[2,3,2,3]
    610 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
    611 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
    612 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
    613 ; AVX2-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
    614 ; AVX2-NEXT:    retq
    615 ;
    616 ; AVX512-LABEL: var_shuffle_v4i64_from_v2i64:
    617 ; AVX512:       # %bb.0:
    618 ; AVX512-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
    619 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
    620 ; AVX512-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
    621 ; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
    622 ; AVX512-NEXT:    retq
    623 ;
    624 ; AVX512VL-LABEL: var_shuffle_v4i64_from_v2i64:
    625 ; AVX512VL:       # %bb.0:
    626 ; AVX512VL-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    627 ; AVX512VL-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
    628 ; AVX512VL-NEXT:    retq
    629   %index0 = extractelement <4 x i64> %indices, i32 0
    630   %index1 = extractelement <4 x i64> %indices, i32 1
    631   %index2 = extractelement <4 x i64> %indices, i32 2
    632   %index3 = extractelement <4 x i64> %indices, i32 3
    633   %v0 = extractelement <2 x i64> %v, i64 %index0
    634   %v1 = extractelement <2 x i64> %v, i64 %index1
    635   %v2 = extractelement <2 x i64> %v, i64 %index2
    636   %v3 = extractelement <2 x i64> %v, i64 %index3
    637   %ret0 = insertelement <4 x i64> undef, i64 %v0, i32 0
    638   %ret1 = insertelement <4 x i64> %ret0, i64 %v1, i32 1
    639   %ret2 = insertelement <4 x i64> %ret1, i64 %v2, i32 2
    640   %ret3 = insertelement <4 x i64> %ret2, i64 %v3, i32 3
    641   ret <4 x i64> %ret3
    642 }
    643 
    644 define <8 x i32> @var_shuffle_v8i32_from_v4i32(<4 x i32> %v, <8 x i32> %indices) unnamed_addr nounwind {
    645 ; XOP-LABEL: var_shuffle_v8i32_from_v4i32:
    646 ; XOP:       # %bb.0: # %entry
    647 ; XOP-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    648 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
    649 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    650 ; XOP-NEXT:    vpermil2ps $0, %ymm1, %ymm2, %ymm0, %ymm0
    651 ; XOP-NEXT:    retq
    652 ;
    653 ; AVX1-LABEL: var_shuffle_v8i32_from_v4i32:
    654 ; AVX1:       # %bb.0: # %entry
    655 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    656 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,3,3,3,3,3,3,3]
    657 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
    658 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
    659 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm4, %xmm3
    660 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm1, %xmm2
    661 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
    662 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
    663 ; AVX1-NEXT:    vpermilps %ymm1, %ymm3, %ymm3
    664 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
    665 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
    666 ; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
    667 ; AVX1-NEXT:    retq
    668 ;
    669 ; INT256-LABEL: var_shuffle_v8i32_from_v4i32:
    670 ; INT256:       # %bb.0: # %entry
    671 ; INT256-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    672 ; INT256-NEXT:    vpermps %ymm0, %ymm1, %ymm0
    673 ; INT256-NEXT:    retq
    674 entry:
    675   %tmp1 = extractelement <8 x i32> %indices, i32 0
    676   %vecext2.8 = extractelement <4 x i32> %v, i32 %tmp1
    677   %tmp2 = extractelement <8 x i32> %indices, i32 1
    678   %vecext2.9 = extractelement <4 x i32> %v, i32 %tmp2
    679   %tmp3 = extractelement <8 x i32> %indices, i32 2
    680   %vecext2.10 = extractelement <4 x i32> %v, i32 %tmp3
    681   %tmp4 = extractelement <8 x i32> %indices, i32 3
    682   %vecext2.11 = extractelement <4 x i32> %v, i32 %tmp4
    683   %tmp5 = extractelement <8 x i32> %indices, i32 4
    684   %vecext2.12 = extractelement <4 x i32> %v, i32 %tmp5
    685   %tmp6 = extractelement <8 x i32> %indices, i32 5
    686   %vecext2.13 = extractelement <4 x i32> %v, i32 %tmp6
    687   %tmp7 = extractelement <8 x i32> %indices, i32 6
    688   %vecext2.14 = extractelement <4 x i32> %v, i32 %tmp7
    689   %tmp8 = extractelement <8 x i32> %indices, i32 7
    690   %vecext2.15 = extractelement <4 x i32> %v, i32 %tmp8
    691   %tmp9 = insertelement <8 x i32> undef, i32 %vecext2.8, i32 0
    692   %tmp10 = insertelement <8 x i32> %tmp9, i32 %vecext2.9, i32 1
    693   %tmp11 = insertelement <8 x i32> %tmp10, i32 %vecext2.10, i32 2
    694   %tmp12 = insertelement <8 x i32> %tmp11, i32 %vecext2.11, i32 3
    695   %tmp13 = insertelement <8 x i32> %tmp12, i32 %vecext2.12, i32 4
    696   %tmp14 = insertelement <8 x i32> %tmp13, i32 %vecext2.13, i32 5
    697   %tmp15 = insertelement <8 x i32> %tmp14, i32 %vecext2.14, i32 6
    698   %tmp16 = insertelement <8 x i32> %tmp15, i32 %vecext2.15, i32 7
    699   ret <8 x i32> %tmp16
    700 }
    701 
    702 define <16 x i16> @var_shuffle_v16i16_from_v8i16(<8 x i16> %v, <16 x i16> %indices) nounwind {
    703 ; XOP-LABEL: var_shuffle_v16i16_from_v8i16:
    704 ; XOP:       # %bb.0:
    705 ; XOP-NEXT:    vmovdqa {{.*#+}} xmm2 = [256,256,256,256,256,256,256,256]
    706 ; XOP-NEXT:    vmovdqa {{.*#+}} xmm3 = [514,514,514,514,514,514,514,514]
    707 ; XOP-NEXT:    vpmacsww %xmm2, %xmm3, %xmm1, %xmm4
    708 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm1
    709 ; XOP-NEXT:    vpmacsww %xmm2, %xmm3, %xmm1, %xmm1
    710 ; XOP-NEXT:    vpperm %xmm1, %xmm0, %xmm0, %xmm1
    711 ; XOP-NEXT:    vpperm %xmm4, %xmm0, %xmm0, %xmm0
    712 ; XOP-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
    713 ; XOP-NEXT:    retq
    714 ;
    715 ; AVX1-LABEL: var_shuffle_v16i16_from_v8i16:
    716 ; AVX1:       # %bb.0:
    717 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [514,514,514,514,514,514,514,514]
    718 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm3
    719 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm4 = [256,256,256,256,256,256,256,256]
    720 ; AVX1-NEXT:    vpaddw %xmm4, %xmm3, %xmm3
    721 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm1
    722 ; AVX1-NEXT:    vpmullw %xmm2, %xmm1, %xmm1
    723 ; AVX1-NEXT:    vpaddw %xmm4, %xmm1, %xmm1
    724 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
    725 ; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm1, %xmm4
    726 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm5
    727 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm1
    728 ; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm1, %xmm1
    729 ; AVX1-NEXT:    vpcmpgtb %xmm2, %xmm3, %xmm2
    730 ; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm4
    731 ; AVX1-NEXT:    vpshufb %xmm3, %xmm0, %xmm0
    732 ; AVX1-NEXT:    vpblendvb %xmm2, %xmm4, %xmm0, %xmm0
    733 ; AVX1-NEXT:    vinsertf128 $1, %xmm1, %ymm0, %ymm0
    734 ; AVX1-NEXT:    retq
    735 ;
    736 ; AVX2-LABEL: var_shuffle_v16i16_from_v8i16:
    737 ; AVX2:       # %bb.0:
    738 ; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    739 ; AVX2-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
    740 ; AVX2-NEXT:    vpaddw {{.*}}(%rip), %ymm1, %ymm1
    741 ; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
    742 ; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
    743 ; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    744 ; AVX2-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    745 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
    746 ; AVX2-NEXT:    retq
    747 ;
    748 ; AVX512-LABEL: var_shuffle_v16i16_from_v8i16:
    749 ; AVX512:       # %bb.0:
    750 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    751 ; AVX512-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
    752 ; AVX512-NEXT:    vpaddw {{.*}}(%rip), %ymm1, %ymm1
    753 ; AVX512-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
    754 ; AVX512-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
    755 ; AVX512-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    756 ; AVX512-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    757 ; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
    758 ; AVX512-NEXT:    retq
    759 ;
    760 ; AVX512VLDQ-LABEL: var_shuffle_v16i16_from_v8i16:
    761 ; AVX512VLDQ:       # %bb.0:
    762 ; AVX512VLDQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    763 ; AVX512VLDQ-NEXT:    vpmullw {{.*}}(%rip), %ymm1, %ymm1
    764 ; AVX512VLDQ-NEXT:    vpaddw {{.*}}(%rip), %ymm1, %ymm1
    765 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
    766 ; AVX512VLDQ-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
    767 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    768 ; AVX512VLDQ-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    769 ; AVX512VLDQ-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
    770 ; AVX512VLDQ-NEXT:    retq
    771 ;
    772 ; AVX512VLBW-LABEL: var_shuffle_v16i16_from_v8i16:
    773 ; AVX512VLBW:       # %bb.0:
    774 ; AVX512VLBW-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    775 ; AVX512VLBW-NEXT:    vpermw %ymm0, %ymm1, %ymm0
    776 ; AVX512VLBW-NEXT:    retq
    777 ;
    778 ; VLVBMI-LABEL: var_shuffle_v16i16_from_v8i16:
    779 ; VLVBMI:       # %bb.0:
    780 ; VLVBMI-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    781 ; VLVBMI-NEXT:    vpermw %ymm0, %ymm1, %ymm0
    782 ; VLVBMI-NEXT:    retq
    783   %index0 = extractelement <16 x i16> %indices, i32 0
    784   %index1 = extractelement <16 x i16> %indices, i32 1
    785   %index2 = extractelement <16 x i16> %indices, i32 2
    786   %index3 = extractelement <16 x i16> %indices, i32 3
    787   %index4 = extractelement <16 x i16> %indices, i32 4
    788   %index5 = extractelement <16 x i16> %indices, i32 5
    789   %index6 = extractelement <16 x i16> %indices, i32 6
    790   %index7 = extractelement <16 x i16> %indices, i32 7
    791   %index8 = extractelement <16 x i16> %indices, i32 8
    792   %index9 = extractelement <16 x i16> %indices, i32 9
    793   %index10 = extractelement <16 x i16> %indices, i32 10
    794   %index11 = extractelement <16 x i16> %indices, i32 11
    795   %index12 = extractelement <16 x i16> %indices, i32 12
    796   %index13 = extractelement <16 x i16> %indices, i32 13
    797   %index14 = extractelement <16 x i16> %indices, i32 14
    798   %index15 = extractelement <16 x i16> %indices, i32 15
    799   %v0 = extractelement <8 x i16> %v, i16 %index0
    800   %v1 = extractelement <8 x i16> %v, i16 %index1
    801   %v2 = extractelement <8 x i16> %v, i16 %index2
    802   %v3 = extractelement <8 x i16> %v, i16 %index3
    803   %v4 = extractelement <8 x i16> %v, i16 %index4
    804   %v5 = extractelement <8 x i16> %v, i16 %index5
    805   %v6 = extractelement <8 x i16> %v, i16 %index6
    806   %v7 = extractelement <8 x i16> %v, i16 %index7
    807   %v8 = extractelement <8 x i16> %v, i16 %index8
    808   %v9 = extractelement <8 x i16> %v, i16 %index9
    809   %v10 = extractelement <8 x i16> %v, i16 %index10
    810   %v11 = extractelement <8 x i16> %v, i16 %index11
    811   %v12 = extractelement <8 x i16> %v, i16 %index12
    812   %v13 = extractelement <8 x i16> %v, i16 %index13
    813   %v14 = extractelement <8 x i16> %v, i16 %index14
    814   %v15 = extractelement <8 x i16> %v, i16 %index15
    815   %ret0 = insertelement <16 x i16> undef, i16 %v0, i32 0
    816   %ret1 = insertelement <16 x i16> %ret0, i16 %v1, i32 1
    817   %ret2 = insertelement <16 x i16> %ret1, i16 %v2, i32 2
    818   %ret3 = insertelement <16 x i16> %ret2, i16 %v3, i32 3
    819   %ret4 = insertelement <16 x i16> %ret3, i16 %v4, i32 4
    820   %ret5 = insertelement <16 x i16> %ret4, i16 %v5, i32 5
    821   %ret6 = insertelement <16 x i16> %ret5, i16 %v6, i32 6
    822   %ret7 = insertelement <16 x i16> %ret6, i16 %v7, i32 7
    823   %ret8 = insertelement <16 x i16> %ret7, i16 %v8, i32 8
    824   %ret9 = insertelement <16 x i16> %ret8, i16 %v9, i32 9
    825   %ret10 = insertelement <16 x i16> %ret9, i16 %v10, i32 10
    826   %ret11 = insertelement <16 x i16> %ret10, i16 %v11, i32 11
    827   %ret12 = insertelement <16 x i16> %ret11, i16 %v12, i32 12
    828   %ret13 = insertelement <16 x i16> %ret12, i16 %v13, i32 13
    829   %ret14 = insertelement <16 x i16> %ret13, i16 %v14, i32 14
    830   %ret15 = insertelement <16 x i16> %ret14, i16 %v15, i32 15
    831   ret <16 x i16> %ret15
    832 }
    833 
    834 define <32 x i8> @var_shuffle_v32i8_from_v16i8(<16 x i8> %v, <32 x i8> %indices) nounwind {
    835 ; XOP-LABEL: var_shuffle_v32i8_from_v16i8:
    836 ; XOP:       # %bb.0:
    837 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm2
    838 ; XOP-NEXT:    vpperm %xmm2, %xmm0, %xmm0, %xmm2
    839 ; XOP-NEXT:    vpperm %xmm1, %xmm0, %xmm0, %xmm0
    840 ; XOP-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
    841 ; XOP-NEXT:    retq
    842 ;
    843 ; AVX1-LABEL: var_shuffle_v32i8_from_v16i8:
    844 ; AVX1:       # %bb.0:
    845 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm2
    846 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15]
    847 ; AVX1-NEXT:    vpcmpgtb %xmm3, %xmm2, %xmm4
    848 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm5
    849 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm2
    850 ; AVX1-NEXT:    vpblendvb %xmm4, %xmm5, %xmm2, %xmm2
    851 ; AVX1-NEXT:    vpcmpgtb %xmm3, %xmm1, %xmm3
    852 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm4
    853 ; AVX1-NEXT:    vpshufb %xmm1, %xmm0, %xmm0
    854 ; AVX1-NEXT:    vpblendvb %xmm3, %xmm4, %xmm0, %xmm0
    855 ; AVX1-NEXT:    vinsertf128 $1, %xmm2, %ymm0, %ymm0
    856 ; AVX1-NEXT:    retq
    857 ;
    858 ; AVX2-LABEL: var_shuffle_v32i8_from_v16i8:
    859 ; AVX2:       # %bb.0:
    860 ; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    861 ; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
    862 ; AVX2-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
    863 ; AVX2-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    864 ; AVX2-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    865 ; AVX2-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
    866 ; AVX2-NEXT:    retq
    867 ;
    868 ; AVX512-LABEL: var_shuffle_v32i8_from_v16i8:
    869 ; AVX512:       # %bb.0:
    870 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    871 ; AVX512-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
    872 ; AVX512-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
    873 ; AVX512-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    874 ; AVX512-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    875 ; AVX512-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
    876 ; AVX512-NEXT:    retq
    877 ;
    878 ; AVX512VLDQ-LABEL: var_shuffle_v32i8_from_v16i8:
    879 ; AVX512VLDQ:       # %bb.0:
    880 ; AVX512VLDQ-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    881 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm0, %ymm2
    882 ; AVX512VLDQ-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm0
    883 ; AVX512VLDQ-NEXT:    vpshufb %ymm1, %ymm0, %ymm0
    884 ; AVX512VLDQ-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %ymm1
    885 ; AVX512VLDQ-NEXT:    vpblendvb %ymm1, %ymm2, %ymm0, %ymm0
    886 ; AVX512VLDQ-NEXT:    retq
    887 ;
    888 ; AVX512VLBW-LABEL: var_shuffle_v32i8_from_v16i8:
    889 ; AVX512VLBW:       # %bb.0:
    890 ; AVX512VLBW-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    891 ; AVX512VLBW-NEXT:    vinserti128 $1, %xmm0, %ymm0, %ymm2
    892 ; AVX512VLBW-NEXT:    vpshufb %ymm1, %ymm2, %ymm2
    893 ; AVX512VLBW-NEXT:    vpcmpgtb {{.*}}(%rip), %ymm1, %k1
    894 ; AVX512VLBW-NEXT:    vpshufb %ymm1, %ymm0, %ymm2 {%k1}
    895 ; AVX512VLBW-NEXT:    vmovdqa %ymm2, %ymm0
    896 ; AVX512VLBW-NEXT:    retq
    897 ;
    898 ; VLVBMI-LABEL: var_shuffle_v32i8_from_v16i8:
    899 ; VLVBMI:       # %bb.0:
    900 ; VLVBMI-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
    901 ; VLVBMI-NEXT:    vpermb %ymm0, %ymm1, %ymm0
    902 ; VLVBMI-NEXT:    retq
    903   %index0 = extractelement <32 x i8> %indices, i32 0
    904   %index1 = extractelement <32 x i8> %indices, i32 1
    905   %index2 = extractelement <32 x i8> %indices, i32 2
    906   %index3 = extractelement <32 x i8> %indices, i32 3
    907   %index4 = extractelement <32 x i8> %indices, i32 4
    908   %index5 = extractelement <32 x i8> %indices, i32 5
    909   %index6 = extractelement <32 x i8> %indices, i32 6
    910   %index7 = extractelement <32 x i8> %indices, i32 7
    911   %index8 = extractelement <32 x i8> %indices, i32 8
    912   %index9 = extractelement <32 x i8> %indices, i32 9
    913   %index10 = extractelement <32 x i8> %indices, i32 10
    914   %index11 = extractelement <32 x i8> %indices, i32 11
    915   %index12 = extractelement <32 x i8> %indices, i32 12
    916   %index13 = extractelement <32 x i8> %indices, i32 13
    917   %index14 = extractelement <32 x i8> %indices, i32 14
    918   %index15 = extractelement <32 x i8> %indices, i32 15
    919   %index16 = extractelement <32 x i8> %indices, i32 16
    920   %index17 = extractelement <32 x i8> %indices, i32 17
    921   %index18 = extractelement <32 x i8> %indices, i32 18
    922   %index19 = extractelement <32 x i8> %indices, i32 19
    923   %index20 = extractelement <32 x i8> %indices, i32 20
    924   %index21 = extractelement <32 x i8> %indices, i32 21
    925   %index22 = extractelement <32 x i8> %indices, i32 22
    926   %index23 = extractelement <32 x i8> %indices, i32 23
    927   %index24 = extractelement <32 x i8> %indices, i32 24
    928   %index25 = extractelement <32 x i8> %indices, i32 25
    929   %index26 = extractelement <32 x i8> %indices, i32 26
    930   %index27 = extractelement <32 x i8> %indices, i32 27
    931   %index28 = extractelement <32 x i8> %indices, i32 28
    932   %index29 = extractelement <32 x i8> %indices, i32 29
    933   %index30 = extractelement <32 x i8> %indices, i32 30
    934   %index31 = extractelement <32 x i8> %indices, i32 31
    935   %v0 = extractelement <16 x i8> %v, i8 %index0
    936   %v1 = extractelement <16 x i8> %v, i8 %index1
    937   %v2 = extractelement <16 x i8> %v, i8 %index2
    938   %v3 = extractelement <16 x i8> %v, i8 %index3
    939   %v4 = extractelement <16 x i8> %v, i8 %index4
    940   %v5 = extractelement <16 x i8> %v, i8 %index5
    941   %v6 = extractelement <16 x i8> %v, i8 %index6
    942   %v7 = extractelement <16 x i8> %v, i8 %index7
    943   %v8 = extractelement <16 x i8> %v, i8 %index8
    944   %v9 = extractelement <16 x i8> %v, i8 %index9
    945   %v10 = extractelement <16 x i8> %v, i8 %index10
    946   %v11 = extractelement <16 x i8> %v, i8 %index11
    947   %v12 = extractelement <16 x i8> %v, i8 %index12
    948   %v13 = extractelement <16 x i8> %v, i8 %index13
    949   %v14 = extractelement <16 x i8> %v, i8 %index14
    950   %v15 = extractelement <16 x i8> %v, i8 %index15
    951   %v16 = extractelement <16 x i8> %v, i8 %index16
    952   %v17 = extractelement <16 x i8> %v, i8 %index17
    953   %v18 = extractelement <16 x i8> %v, i8 %index18
    954   %v19 = extractelement <16 x i8> %v, i8 %index19
    955   %v20 = extractelement <16 x i8> %v, i8 %index20
    956   %v21 = extractelement <16 x i8> %v, i8 %index21
    957   %v22 = extractelement <16 x i8> %v, i8 %index22
    958   %v23 = extractelement <16 x i8> %v, i8 %index23
    959   %v24 = extractelement <16 x i8> %v, i8 %index24
    960   %v25 = extractelement <16 x i8> %v, i8 %index25
    961   %v26 = extractelement <16 x i8> %v, i8 %index26
    962   %v27 = extractelement <16 x i8> %v, i8 %index27
    963   %v28 = extractelement <16 x i8> %v, i8 %index28
    964   %v29 = extractelement <16 x i8> %v, i8 %index29
    965   %v30 = extractelement <16 x i8> %v, i8 %index30
    966   %v31 = extractelement <16 x i8> %v, i8 %index31
    967   %ret0 = insertelement <32 x i8> undef, i8 %v0, i32 0
    968   %ret1 = insertelement <32 x i8> %ret0, i8 %v1, i32 1
    969   %ret2 = insertelement <32 x i8> %ret1, i8 %v2, i32 2
    970   %ret3 = insertelement <32 x i8> %ret2, i8 %v3, i32 3
    971   %ret4 = insertelement <32 x i8> %ret3, i8 %v4, i32 4
    972   %ret5 = insertelement <32 x i8> %ret4, i8 %v5, i32 5
    973   %ret6 = insertelement <32 x i8> %ret5, i8 %v6, i32 6
    974   %ret7 = insertelement <32 x i8> %ret6, i8 %v7, i32 7
    975   %ret8 = insertelement <32 x i8> %ret7, i8 %v8, i32 8
    976   %ret9 = insertelement <32 x i8> %ret8, i8 %v9, i32 9
    977   %ret10 = insertelement <32 x i8> %ret9, i8 %v10, i32 10
    978   %ret11 = insertelement <32 x i8> %ret10, i8 %v11, i32 11
    979   %ret12 = insertelement <32 x i8> %ret11, i8 %v12, i32 12
    980   %ret13 = insertelement <32 x i8> %ret12, i8 %v13, i32 13
    981   %ret14 = insertelement <32 x i8> %ret13, i8 %v14, i32 14
    982   %ret15 = insertelement <32 x i8> %ret14, i8 %v15, i32 15
    983   %ret16 = insertelement <32 x i8> %ret15, i8 %v16, i32 16
    984   %ret17 = insertelement <32 x i8> %ret16, i8 %v17, i32 17
    985   %ret18 = insertelement <32 x i8> %ret17, i8 %v18, i32 18
    986   %ret19 = insertelement <32 x i8> %ret18, i8 %v19, i32 19
    987   %ret20 = insertelement <32 x i8> %ret19, i8 %v20, i32 20
    988   %ret21 = insertelement <32 x i8> %ret20, i8 %v21, i32 21
    989   %ret22 = insertelement <32 x i8> %ret21, i8 %v22, i32 22
    990   %ret23 = insertelement <32 x i8> %ret22, i8 %v23, i32 23
    991   %ret24 = insertelement <32 x i8> %ret23, i8 %v24, i32 24
    992   %ret25 = insertelement <32 x i8> %ret24, i8 %v25, i32 25
    993   %ret26 = insertelement <32 x i8> %ret25, i8 %v26, i32 26
    994   %ret27 = insertelement <32 x i8> %ret26, i8 %v27, i32 27
    995   %ret28 = insertelement <32 x i8> %ret27, i8 %v28, i32 28
    996   %ret29 = insertelement <32 x i8> %ret28, i8 %v29, i32 29
    997   %ret30 = insertelement <32 x i8> %ret29, i8 %v30, i32 30
    998   %ret31 = insertelement <32 x i8> %ret30, i8 %v31, i32 31
    999   ret <32 x i8> %ret31
   1000 }
   1001 
   1002 define <4 x double> @var_shuffle_v4f64_from_v2f64(<2 x double> %v, <4 x i64> %indices) nounwind {
   1003 ; XOP-LABEL: var_shuffle_v4f64_from_v2f64:
   1004 ; XOP:       # %bb.0:
   1005 ; XOP-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
   1006 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
   1007 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
   1008 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm3
   1009 ; XOP-NEXT:    vextractf128 $1, %ymm1, %xmm1
   1010 ; XOP-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
   1011 ; XOP-NEXT:    vinsertf128 $1, %xmm1, %ymm3, %ymm1
   1012 ; XOP-NEXT:    vpermil2pd $0, %ymm1, %ymm2, %ymm0, %ymm0
   1013 ; XOP-NEXT:    retq
   1014 ;
   1015 ; AVX1-LABEL: var_shuffle_v4f64_from_v2f64:
   1016 ; AVX1:       # %bb.0:
   1017 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
   1018 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [2,2,2,2]
   1019 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
   1020 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
   1021 ; AVX1-NEXT:    vpaddq %xmm4, %xmm4, %xmm4
   1022 ; AVX1-NEXT:    vpcmpgtq %xmm3, %xmm4, %xmm3
   1023 ; AVX1-NEXT:    vpaddq %xmm1, %xmm1, %xmm1
   1024 ; AVX1-NEXT:    vpcmpgtq %xmm2, %xmm1, %xmm2
   1025 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
   1026 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
   1027 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
   1028 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
   1029 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
   1030 ; AVX1-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
   1031 ; AVX1-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
   1032 ; AVX1-NEXT:    retq
   1033 ;
   1034 ; AVX2-LABEL: var_shuffle_v4f64_from_v2f64:
   1035 ; AVX2:       # %bb.0:
   1036 ; AVX2-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
   1037 ; AVX2-NEXT:    vpaddq %ymm1, %ymm1, %ymm1
   1038 ; AVX2-NEXT:    vpbroadcastq {{.*#+}} ymm2 = [2,2,2,2]
   1039 ; AVX2-NEXT:    vpcmpgtq %ymm2, %ymm1, %ymm2
   1040 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm3 = ymm0[2,3,2,3]
   1041 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm3, %ymm3
   1042 ; AVX2-NEXT:    vpermpd {{.*#+}} ymm0 = ymm0[0,1,0,1]
   1043 ; AVX2-NEXT:    vpermilpd %ymm1, %ymm0, %ymm0
   1044 ; AVX2-NEXT:    vblendvpd %ymm2, %ymm3, %ymm0, %ymm0
   1045 ; AVX2-NEXT:    retq
   1046 ;
   1047 ; AVX512-LABEL: var_shuffle_v4f64_from_v2f64:
   1048 ; AVX512:       # %bb.0:
   1049 ; AVX512-NEXT:    # kill: def $ymm1 killed $ymm1 def $zmm1
   1050 ; AVX512-NEXT:    # kill: def $xmm0 killed $xmm0 def $zmm0
   1051 ; AVX512-NEXT:    vpermpd %zmm0, %zmm1, %zmm0
   1052 ; AVX512-NEXT:    # kill: def $ymm0 killed $ymm0 killed $zmm0
   1053 ; AVX512-NEXT:    retq
   1054 ;
   1055 ; AVX512VL-LABEL: var_shuffle_v4f64_from_v2f64:
   1056 ; AVX512VL:       # %bb.0:
   1057 ; AVX512VL-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
   1058 ; AVX512VL-NEXT:    vpermpd %ymm0, %ymm1, %ymm0
   1059 ; AVX512VL-NEXT:    retq
   1060   %index0 = extractelement <4 x i64> %indices, i32 0
   1061   %index1 = extractelement <4 x i64> %indices, i32 1
   1062   %index2 = extractelement <4 x i64> %indices, i32 2
   1063   %index3 = extractelement <4 x i64> %indices, i32 3
   1064   %v0 = extractelement <2 x double> %v, i64 %index0
   1065   %v1 = extractelement <2 x double> %v, i64 %index1
   1066   %v2 = extractelement <2 x double> %v, i64 %index2
   1067   %v3 = extractelement <2 x double> %v, i64 %index3
   1068   %ret0 = insertelement <4 x double> undef, double %v0, i32 0
   1069   %ret1 = insertelement <4 x double> %ret0, double %v1, i32 1
   1070   %ret2 = insertelement <4 x double> %ret1, double %v2, i32 2
   1071   %ret3 = insertelement <4 x double> %ret2, double %v3, i32 3
   1072   ret <4 x double> %ret3
   1073 }
   1074 
   1075 define <8 x float> @var_shuffle_v8f32_from_v4f32(<4 x float> %v, <8 x i32> %indices) unnamed_addr nounwind {
   1076 ; XOP-LABEL: var_shuffle_v8f32_from_v4f32:
   1077 ; XOP:       # %bb.0: # %entry
   1078 ; XOP-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
   1079 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
   1080 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
   1081 ; XOP-NEXT:    vpermil2ps $0, %ymm1, %ymm2, %ymm0, %ymm0
   1082 ; XOP-NEXT:    retq
   1083 ;
   1084 ; AVX1-LABEL: var_shuffle_v8f32_from_v4f32:
   1085 ; AVX1:       # %bb.0: # %entry
   1086 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
   1087 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm2 = [3,3,3,3,3,3,3,3]
   1088 ; AVX1-NEXT:    vextractf128 $1, %ymm2, %xmm3
   1089 ; AVX1-NEXT:    vextractf128 $1, %ymm1, %xmm4
   1090 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm4, %xmm3
   1091 ; AVX1-NEXT:    vpcmpgtd %xmm2, %xmm1, %xmm2
   1092 ; AVX1-NEXT:    vinsertf128 $1, %xmm3, %ymm2, %ymm2
   1093 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm3 = ymm0[2,3,2,3]
   1094 ; AVX1-NEXT:    vpermilps %ymm1, %ymm3, %ymm3
   1095 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
   1096 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
   1097 ; AVX1-NEXT:    vblendvps %ymm2, %ymm3, %ymm0, %ymm0
   1098 ; AVX1-NEXT:    retq
   1099 ;
   1100 ; INT256-LABEL: var_shuffle_v8f32_from_v4f32:
   1101 ; INT256:       # %bb.0: # %entry
   1102 ; INT256-NEXT:    # kill: def $xmm0 killed $xmm0 def $ymm0
   1103 ; INT256-NEXT:    vpermps %ymm0, %ymm1, %ymm0
   1104 ; INT256-NEXT:    retq
   1105 entry:
   1106   %tmp1 = extractelement <8 x i32> %indices, i32 0
   1107   %vecext2.8 = extractelement <4 x float> %v, i32 %tmp1
   1108   %tmp2 = extractelement <8 x i32> %indices, i32 1
   1109   %vecext2.9 = extractelement <4 x float> %v, i32 %tmp2
   1110   %tmp3 = extractelement <8 x i32> %indices, i32 2
   1111   %vecext2.10 = extractelement <4 x float> %v, i32 %tmp3
   1112   %tmp4 = extractelement <8 x i32> %indices, i32 3
   1113   %vecext2.11 = extractelement <4 x float> %v, i32 %tmp4
   1114   %tmp5 = extractelement <8 x i32> %indices, i32 4
   1115   %vecext2.12 = extractelement <4 x float> %v, i32 %tmp5
   1116   %tmp6 = extractelement <8 x i32> %indices, i32 5
   1117   %vecext2.13 = extractelement <4 x float> %v, i32 %tmp6
   1118   %tmp7 = extractelement <8 x i32> %indices, i32 6
   1119   %vecext2.14 = extractelement <4 x float> %v, i32 %tmp7
   1120   %tmp8 = extractelement <8 x i32> %indices, i32 7
   1121   %vecext2.15 = extractelement <4 x float> %v, i32 %tmp8
   1122   %tmp9 = insertelement <8 x float> undef, float %vecext2.8, i32 0
   1123   %tmp10 = insertelement <8 x float> %tmp9, float %vecext2.9, i32 1
   1124   %tmp11 = insertelement <8 x float> %tmp10, float %vecext2.10, i32 2
   1125   %tmp12 = insertelement <8 x float> %tmp11, float %vecext2.11, i32 3
   1126   %tmp13 = insertelement <8 x float> %tmp12, float %vecext2.12, i32 4
   1127   %tmp14 = insertelement <8 x float> %tmp13, float %vecext2.13, i32 5
   1128   %tmp15 = insertelement <8 x float> %tmp14, float %vecext2.14, i32 6
   1129   %tmp16 = insertelement <8 x float> %tmp15, float %vecext2.15, i32 7
   1130   ret <8 x float> %tmp16
   1131 }
   1132 
   1133 define <4 x i32> @var_shuffle_v4i32_from_v8i32(<8 x i32> %v, <4 x i32> %indices) unnamed_addr nounwind {
   1134 ; XOP-LABEL: var_shuffle_v4i32_from_v8i32:
   1135 ; XOP:       # %bb.0: # %entry
   1136 ; XOP-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
   1137 ; XOP-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
   1138 ; XOP-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
   1139 ; XOP-NEXT:    vpermil2ps $0, %ymm1, %ymm2, %ymm0, %ymm0
   1140 ; XOP-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
   1141 ; XOP-NEXT:    vzeroupper
   1142 ; XOP-NEXT:    retq
   1143 ;
   1144 ; AVX1-LABEL: var_shuffle_v4i32_from_v8i32:
   1145 ; AVX1:       # %bb.0: # %entry
   1146 ; AVX1-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
   1147 ; AVX1-NEXT:    vperm2f128 {{.*#+}} ymm2 = ymm0[2,3,2,3]
   1148 ; AVX1-NEXT:    vpermilps %ymm1, %ymm2, %ymm2
   1149 ; AVX1-NEXT:    vinsertf128 $1, %xmm0, %ymm0, %ymm0
   1150 ; AVX1-NEXT:    vpermilps %ymm1, %ymm0, %ymm0
   1151 ; AVX1-NEXT:    vmovdqa {{.*#+}} ymm3 = [3,3,3,3,3,3,3,3]
   1152 ; AVX1-NEXT:    vextractf128 $1, %ymm3, %xmm4
   1153 ; AVX1-NEXT:    vpcmpgtd %xmm4, %xmm0, %xmm4
   1154 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm1, %xmm1
   1155 ; AVX1-NEXT:    vinsertf128 $1, %xmm4, %ymm1, %ymm1
   1156 ; AVX1-NEXT:    vblendvps %ymm1, %ymm2, %ymm0, %ymm0
   1157 ; AVX1-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
   1158 ; AVX1-NEXT:    vzeroupper
   1159 ; AVX1-NEXT:    retq
   1160 ;
   1161 ; INT256-LABEL: var_shuffle_v4i32_from_v8i32:
   1162 ; INT256:       # %bb.0: # %entry
   1163 ; INT256-NEXT:    # kill: def $xmm1 killed $xmm1 def $ymm1
   1164 ; INT256-NEXT:    vpermps %ymm0, %ymm1, %ymm0
   1165 ; INT256-NEXT:    # kill: def $xmm0 killed $xmm0 killed $ymm0
   1166 ; INT256-NEXT:    vzeroupper
   1167 ; INT256-NEXT:    retq
   1168 entry:
   1169   %tmp1 = extractelement <4 x i32> %indices, i32 0
   1170   %vecext2.8 = extractelement <8 x i32> %v, i32 %tmp1
   1171   %tmp2 = extractelement <4 x i32> %indices, i32 1
   1172   %vecext2.9 = extractelement <8 x i32> %v, i32 %tmp2
   1173   %tmp3 = extractelement <4 x i32> %indices, i32 2
   1174   %vecext2.10 = extractelement <8 x i32> %v, i32 %tmp3
   1175   %tmp4 = extractelement <4 x i32> %indices, i32 3
   1176   %vecext2.11 = extractelement <8 x i32> %v, i32 %tmp4
   1177   %tmp9 = insertelement <4 x i32> undef, i32 %vecext2.8, i32 0
   1178   %tmp10 = insertelement <4 x i32> %tmp9, i32 %vecext2.9, i32 1
   1179   %tmp11 = insertelement <4 x i32> %tmp10, i32 %vecext2.10, i32 2
   1180   %tmp12 = insertelement <4 x i32> %tmp11, i32 %vecext2.11, i32 3
   1181   ret <4 x i32> %tmp12
   1182 }
   1183