1 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512f | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512F 2 ; RUN: llc < %s -mcpu=x86-64 -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW 3 4 target triple = "x86_64-unknown-unknown" 5 6 define <8 x double> @shuffle_v8f64_00000000(<8 x double> %a, <8 x double> %b) { 7 ; ALL-LABEL: shuffle_v8f64_00000000: 8 ; ALL: # BB#0: 9 ; ALL-NEXT: vbroadcastsd %xmm0, %zmm0 10 ; ALL-NEXT: retq 11 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 12 ret <8 x double> %shuffle 13 } 14 15 define <8 x double> @shuffle_v8f64_00000010(<8 x double> %a, <8 x double> %b) { 16 ; ALL-LABEL: shuffle_v8f64_00000010: 17 ; ALL: # BB#0: 18 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1 19 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,0] 20 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 21 ; ALL-NEXT: retq 22 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0> 23 ret <8 x double> %shuffle 24 } 25 26 define <8 x double> @shuffle_v8f64_00000200(<8 x double> %a, <8 x double> %b) { 27 ; ALL-LABEL: shuffle_v8f64_00000200: 28 ; ALL: # BB#0: 29 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1 30 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,0] 31 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 32 ; ALL-NEXT: retq 33 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0> 34 ret <8 x double> %shuffle 35 } 36 37 define <8 x double> @shuffle_v8f64_00003000(<8 x double> %a, <8 x double> %b) { 38 ; ALL-LABEL: shuffle_v8f64_00003000: 39 ; ALL: # BB#0: 40 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1 41 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,0,0,0] 42 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 43 ; ALL-NEXT: retq 44 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0> 45 ret <8 x double> %shuffle 46 } 47 48 define <8 x double> @shuffle_v8f64_00040000(<8 x double> %a, <8 x double> %b) { 49 ; ALL-LABEL: shuffle_v8f64_00040000: 50 ; ALL: # BB#0: 51 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 52 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1 53 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0 54 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3] 55 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 56 ; ALL-NEXT: retq 57 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 4, i32 0, i32 0, i32 0, i32 0> 58 ret <8 x double> %shuffle 59 } 60 61 define <8 x double> @shuffle_v8f64_00500000(<8 x double> %a, <8 x double> %b) { 62 ; ALL-LABEL: shuffle_v8f64_00500000: 63 ; ALL: # BB#0: 64 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 65 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm0[0],ymm1[1],ymm0[2,3] 66 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,1,0] 67 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0 68 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 69 ; ALL-NEXT: retq 70 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 5, i32 0, i32 0, i32 0, i32 0, i32 0> 71 ret <8 x double> %shuffle 72 } 73 74 define <8 x double> @shuffle_v8f64_06000000(<8 x double> %a, <8 x double> %b) { 75 ; ALL-LABEL: shuffle_v8f64_06000000: 76 ; ALL: # BB#0: 77 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 78 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm0[0,1],ymm1[2],ymm0[3] 79 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,2,0,0] 80 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0 81 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 82 ; ALL-NEXT: retq 83 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 6, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 84 ret <8 x double> %shuffle 85 } 86 87 define <8 x double> @shuffle_v8f64_70000000(<8 x double> %a, <8 x double> %b) { 88 ; ALL-LABEL: shuffle_v8f64_70000000: 89 ; ALL: # BB#0: 90 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 91 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3] 92 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[3,0,0,0] 93 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0 94 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 95 ; ALL-NEXT: retq 96 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 97 ret <8 x double> %shuffle 98 } 99 100 define <8 x double> @shuffle_v8f64_01014545(<8 x double> %a, <8 x double> %b) { 101 ; ALL-LABEL: shuffle_v8f64_01014545: 102 ; ALL: # BB#0: 103 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 104 ; ALL-NEXT: vinsertf128 $1, %xmm1, %ymm1, %ymm1 105 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 106 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0 107 ; ALL-NEXT: retq 108 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5> 109 ret <8 x double> %shuffle 110 } 111 112 define <8 x double> @shuffle_v8f64_00112233(<8 x double> %a, <8 x double> %b) { 113 ; ALL-LABEL: shuffle_v8f64_00112233: 114 ; ALL: # BB#0: 115 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,1,1] 116 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,3,3] 117 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 118 ; ALL-NEXT: retq 119 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3> 120 ret <8 x double> %shuffle 121 } 122 123 define <8 x double> @shuffle_v8f64_00001111(<8 x double> %a, <8 x double> %b) { 124 ; ALL-LABEL: shuffle_v8f64_00001111: 125 ; ALL: # BB#0: 126 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1 127 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,1,1,1] 128 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 129 ; ALL-NEXT: retq 130 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1> 131 ret <8 x double> %shuffle 132 } 133 134 define <8 x double> @shuffle_v8f64_81a3c5e7(<8 x double> %a, <8 x double> %b) { 135 ; ALL-LABEL: shuffle_v8f64_81a3c5e7: 136 ; ALL: # BB#0: 137 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2 138 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3 139 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3] 140 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2],ymm0[3] 141 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 142 ; ALL-NEXT: retq 143 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7> 144 ret <8 x double> %shuffle 145 } 146 147 define <8 x double> @shuffle_v8f64_08080808(<8 x double> %a, <8 x double> %b) { 148 ; ALL-LABEL: shuffle_v8f64_08080808: 149 ; ALL: # BB#0: 150 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 151 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1 152 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3] 153 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm0, %zmm0 154 ; ALL-NEXT: retq 155 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8> 156 ret <8 x double> %shuffle 157 } 158 159 define <8 x double> @shuffle_v8f64_08084c4c(<8 x double> %a, <8 x double> %b) { 160 ; ALL-LABEL: shuffle_v8f64_08084c4c: 161 ; ALL: # BB#0: 162 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2 163 ; ALL-NEXT: vinsertf128 $1, %xmm2, %ymm2, %ymm2 164 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3 165 ; ALL-NEXT: vbroadcastsd %xmm3, %ymm3 166 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm2[0],ymm3[1],ymm2[2],ymm3[3] 167 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 168 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1 169 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3] 170 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 171 ; ALL-NEXT: retq 172 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 4, i32 12, i32 4, i32 12> 173 ret <8 x double> %shuffle 174 } 175 176 define <8 x double> @shuffle_v8f64_8823cc67(<8 x double> %a, <8 x double> %b) { 177 ; ALL-LABEL: shuffle_v8f64_8823cc67: 178 ; ALL: # BB#0: 179 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2 180 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3 181 ; ALL-NEXT: vbroadcastsd %xmm3, %ymm3 182 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3] 183 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm1 184 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3] 185 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 186 ; ALL-NEXT: retq 187 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 8, i32 8, i32 2, i32 3, i32 12, i32 12, i32 6, i32 7> 188 ret <8 x double> %shuffle 189 } 190 191 define <8 x double> @shuffle_v8f64_9832dc76(<8 x double> %a, <8 x double> %b) { 192 ; ALL-LABEL: shuffle_v8f64_9832dc76: 193 ; ALL: # BB#0: 194 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm1[0,1],ymm0[2,3] 195 ; ALL-NEXT: vpermilpd {{.*#+}} ymm2 = ymm2[1,0,3,2] 196 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 197 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm1 198 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3] 199 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2] 200 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0 201 ; ALL-NEXT: retq 202 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6> 203 ret <8 x double> %shuffle 204 } 205 206 define <8 x double> @shuffle_v8f64_9810dc54(<8 x double> %a, <8 x double> %b) { 207 ; ALL-LABEL: shuffle_v8f64_9810dc54: 208 ; ALL: # BB#0: 209 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm2 210 ; ALL-NEXT: vpermilpd {{.*#+}} ymm2 = ymm2[1,0,3,2] 211 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm1 212 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 213 ; ALL-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 214 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2] 215 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0 216 ; ALL-NEXT: retq 217 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 9, i32 8, i32 1, i32 0, i32 13, i32 12, i32 5, i32 4> 218 ret <8 x double> %shuffle 219 } 220 221 define <8 x double> @shuffle_v8f64_08194c5d(<8 x double> %a, <8 x double> %b) { 222 ; ALL-LABEL: shuffle_v8f64_08194c5d: 223 ; ALL: # BB#0: 224 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2 225 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm2[0,0,2,1] 226 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3 227 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm3[0,1,1,3] 228 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3] 229 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,2,1] 230 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3] 231 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3] 232 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 233 ; ALL-NEXT: retq 234 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13> 235 ret <8 x double> %shuffle 236 } 237 238 define <8 x double> @shuffle_v8f64_2a3b6e7f(<8 x double> %a, <8 x double> %b) { 239 ; ALL-LABEL: shuffle_v8f64_2a3b6e7f: 240 ; ALL: # BB#0: 241 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2 242 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm2[0,2,2,3] 243 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm3 244 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm3[2,1,3,3] 245 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3] 246 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,2,2,3] 247 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,1,3,3] 248 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3] 249 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 250 ; ALL-NEXT: retq 251 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15> 252 ret <8 x double> %shuffle 253 } 254 255 define <8 x double> @shuffle_v8f64_08192a3b(<8 x double> %a, <8 x double> %b) { 256 ; ALL-LABEL: shuffle_v8f64_08192a3b: 257 ; ALL: # BB#0: 258 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm1[0,2,2,3] 259 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm0[2,1,3,3] 260 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3] 261 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[0,0,2,1] 262 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3] 263 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3] 264 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 265 ; ALL-NEXT: retq 266 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11> 267 ret <8 x double> %shuffle 268 } 269 270 define <8 x double> @shuffle_v8f64_08991abb(<8 x double> %a, <8 x double> %b) { 271 ; ALL-LABEL: shuffle_v8f64_08991abb: 272 ; ALL: # BB#0: 273 ; ALL-NEXT: vpermpd {{.*#+}} ymm2 = ymm1[0,0,1,1] 274 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm0[0],ymm2[1,2,3] 275 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0],ymm0[1],ymm1[2,3] 276 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,2,3,3] 277 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0 278 ; ALL-NEXT: retq 279 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 9, i32 1, i32 10, i32 11, i32 11> 280 ret <8 x double> %shuffle 281 } 282 283 define <8 x double> @shuffle_v8f64_091b2d3f(<8 x double> %a, <8 x double> %b) { 284 ; ALL-LABEL: shuffle_v8f64_091b2d3f: 285 ; ALL: # BB#0: 286 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2 287 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm0[2,1,3,3] 288 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1],ymm3[2],ymm2[3] 289 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,3] 290 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3] 291 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 292 ; ALL-NEXT: retq 293 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 9, i32 1, i32 11, i32 2, i32 13, i32 3, i32 15> 294 ret <8 x double> %shuffle 295 } 296 297 define <8 x double> @shuffle_v8f64_09ab1def(<8 x double> %a, <8 x double> %b) { 298 ; ALL-LABEL: shuffle_v8f64_09ab1def: 299 ; ALL: # BB#0: 300 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm2 301 ; ALL-NEXT: vpermilpd {{.*#+}} ymm3 = ymm0[1,0,2,2] 302 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm3[0],ymm2[1,2,3] 303 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm1[1,2,3] 304 ; ALL-NEXT: vinsertf64x4 $1, %ymm2, %zmm0, %zmm0 305 ; ALL-NEXT: retq 306 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 1, i32 13, i32 14, i32 15> 307 ret <8 x double> %shuffle 308 } 309 310 define <8 x double> @shuffle_v8f64_00014445(<8 x double> %a, <8 x double> %b) { 311 ; ALL-LABEL: shuffle_v8f64_00014445: 312 ; ALL: # BB#0: 313 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,0,1] 314 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 315 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,0,1] 316 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 317 ; ALL-NEXT: retq 318 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 4, i32 4, i32 4, i32 5> 319 ret <8 x double> %shuffle 320 } 321 322 define <8 x double> @shuffle_v8f64_00204464(<8 x double> %a, <8 x double> %b) { 323 ; ALL-LABEL: shuffle_v8f64_00204464: 324 ; ALL: # BB#0: 325 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,2,0] 326 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 327 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,2,0] 328 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 329 ; ALL-NEXT: retq 330 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 4, i32 6, i32 4> 331 ret <8 x double> %shuffle 332 } 333 334 define <8 x double> @shuffle_v8f64_03004744(<8 x double> %a, <8 x double> %b) { 335 ; ALL-LABEL: shuffle_v8f64_03004744: 336 ; ALL: # BB#0: 337 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,3,0,0] 338 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 339 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,3,0,0] 340 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 341 ; ALL-NEXT: retq 342 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 7, i32 4, i32 4> 343 ret <8 x double> %shuffle 344 } 345 346 define <8 x double> @shuffle_v8f64_10005444(<8 x double> %a, <8 x double> %b) { 347 ; ALL-LABEL: shuffle_v8f64_10005444: 348 ; ALL: # BB#0: 349 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[1,0,0,0] 350 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 351 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,0,0,0] 352 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 353 ; ALL-NEXT: retq 354 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 5, i32 4, i32 4, i32 4> 355 ret <8 x double> %shuffle 356 } 357 358 define <8 x double> @shuffle_v8f64_22006644(<8 x double> %a, <8 x double> %b) { 359 ; ALL-LABEL: shuffle_v8f64_22006644: 360 ; ALL: # BB#0: 361 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[2,2,0,0] 362 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 363 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,0,0] 364 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 365 ; ALL-NEXT: retq 366 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 6, i32 4, i32 4> 367 ret <8 x double> %shuffle 368 } 369 370 define <8 x double> @shuffle_v8f64_33307774(<8 x double> %a, <8 x double> %b) { 371 ; ALL-LABEL: shuffle_v8f64_33307774: 372 ; ALL: # BB#0: 373 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,3,3,0] 374 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 375 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,3,3,0] 376 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 377 ; ALL-NEXT: retq 378 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 7, i32 7, i32 4> 379 ret <8 x double> %shuffle 380 } 381 382 define <8 x double> @shuffle_v8f64_32107654(<8 x double> %a, <8 x double> %b) { 383 ; ALL-LABEL: shuffle_v8f64_32107654: 384 ; ALL: # BB#0: 385 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,2,1,0] 386 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 387 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,2,1,0] 388 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 389 ; ALL-NEXT: retq 390 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4> 391 ret <8 x double> %shuffle 392 } 393 394 define <8 x double> @shuffle_v8f64_00234467(<8 x double> %a, <8 x double> %b) { 395 ; ALL-LABEL: shuffle_v8f64_00234467: 396 ; ALL: # BB#0: 397 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,3] 398 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 399 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[0,0,2,3] 400 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 401 ; ALL-NEXT: retq 402 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 4, i32 4, i32 6, i32 7> 403 ret <8 x double> %shuffle 404 } 405 406 define <8 x double> @shuffle_v8f64_00224466(<8 x double> %a, <8 x double> %b) { 407 ; ALL-LABEL: shuffle_v8f64_00224466: 408 ; ALL: # BB#0: 409 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm0[0,0,2,2] 410 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 411 ; ALL-NEXT: vmovddup {{.*#+}} ymm0 = ymm0[0,0,2,2] 412 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 413 ; ALL-NEXT: retq 414 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6> 415 ret <8 x double> %shuffle 416 } 417 418 define <8 x double> @shuffle_v8f64_10325476(<8 x double> %a, <8 x double> %b) { 419 ; ALL-LABEL: shuffle_v8f64_10325476: 420 ; ALL: # BB#0: 421 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2] 422 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 423 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,3,2] 424 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 425 ; ALL-NEXT: retq 426 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6> 427 ret <8 x double> %shuffle 428 } 429 430 define <8 x double> @shuffle_v8f64_11335577(<8 x double> %a, <8 x double> %b) { 431 ; ALL-LABEL: shuffle_v8f64_11335577: 432 ; ALL: # BB#0: 433 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3] 434 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 435 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,1,3,3] 436 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 437 ; ALL-NEXT: retq 438 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7> 439 ret <8 x double> %shuffle 440 } 441 442 define <8 x double> @shuffle_v8f64_10235467(<8 x double> %a, <8 x double> %b) { 443 ; ALL-LABEL: shuffle_v8f64_10235467: 444 ; ALL: # BB#0: 445 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,2,3] 446 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 447 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,3] 448 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 449 ; ALL-NEXT: retq 450 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7> 451 ret <8 x double> %shuffle 452 } 453 454 define <8 x double> @shuffle_v8f64_10225466(<8 x double> %a, <8 x double> %b) { 455 ; ALL-LABEL: shuffle_v8f64_10225466: 456 ; ALL: # BB#0: 457 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,2,2] 458 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 459 ; ALL-NEXT: vpermilpd {{.*#+}} ymm0 = ymm0[1,0,2,2] 460 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 461 ; ALL-NEXT: retq 462 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 2, i32 5, i32 4, i32 6, i32 6> 463 ret <8 x double> %shuffle 464 } 465 466 define <8 x double> @shuffle_v8f64_00015444(<8 x double> %a, <8 x double> %b) { 467 ; ALL-LABEL: shuffle_v8f64_00015444: 468 ; ALL: # BB#0: 469 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,0,1] 470 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 471 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[1,0,0,0] 472 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 473 ; ALL-NEXT: retq 474 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 5, i32 4, i32 4, i32 4> 475 ret <8 x double> %shuffle 476 } 477 478 define <8 x double> @shuffle_v8f64_00204644(<8 x double> %a, <8 x double> %b) { 479 ; ALL-LABEL: shuffle_v8f64_00204644: 480 ; ALL: # BB#0: 481 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,0,2,0] 482 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 483 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,2,0,0] 484 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 485 ; ALL-NEXT: retq 486 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 6, i32 4, i32 4> 487 ret <8 x double> %shuffle 488 } 489 490 define <8 x double> @shuffle_v8f64_03004474(<8 x double> %a, <8 x double> %b) { 491 ; ALL-LABEL: shuffle_v8f64_03004474: 492 ; ALL: # BB#0: 493 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[0,3,0,0] 494 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 495 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,3,0] 496 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 497 ; ALL-NEXT: retq 498 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 4, i32 7, i32 4> 499 ret <8 x double> %shuffle 500 } 501 502 define <8 x double> @shuffle_v8f64_10004444(<8 x double> %a, <8 x double> %b) { 503 ; ALL-LABEL: shuffle_v8f64_10004444: 504 ; ALL: # BB#0: 505 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[1,0,0,0] 506 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 507 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0 508 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 509 ; ALL-NEXT: retq 510 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 4, i32 4, i32 4, i32 4> 511 ret <8 x double> %shuffle 512 } 513 514 define <8 x double> @shuffle_v8f64_22006446(<8 x double> %a, <8 x double> %b) { 515 ; ALL-LABEL: shuffle_v8f64_22006446: 516 ; ALL: # BB#0: 517 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[2,2,0,0] 518 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 519 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,0,0,2] 520 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 521 ; ALL-NEXT: retq 522 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 4, i32 4, i32 6> 523 ret <8 x double> %shuffle 524 } 525 526 define <8 x double> @shuffle_v8f64_33307474(<8 x double> %a, <8 x double> %b) { 527 ; ALL-LABEL: shuffle_v8f64_33307474: 528 ; ALL: # BB#0: 529 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,3,3,0] 530 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 531 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[3,0,3,0] 532 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 533 ; ALL-NEXT: retq 534 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 4, i32 7, i32 4> 535 ret <8 x double> %shuffle 536 } 537 538 define <8 x double> @shuffle_v8f64_32104567(<8 x double> %a, <8 x double> %b) { 539 ; ALL-LABEL: shuffle_v8f64_32104567: 540 ; ALL: # BB#0: 541 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm0[3,2,1,0] 542 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 543 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 544 ; ALL-NEXT: retq 545 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7> 546 ret <8 x double> %shuffle 547 } 548 549 define <8 x double> @shuffle_v8f64_00236744(<8 x double> %a, <8 x double> %b) { 550 ; ALL-LABEL: shuffle_v8f64_00236744: 551 ; ALL: # BB#0: 552 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[0,0,2,3] 553 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 554 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,3,0,0] 555 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 556 ; ALL-NEXT: retq 557 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 6, i32 7, i32 4, i32 4> 558 ret <8 x double> %shuffle 559 } 560 561 define <8 x double> @shuffle_v8f64_00226644(<8 x double> %a, <8 x double> %b) { 562 ; ALL-LABEL: shuffle_v8f64_00226644: 563 ; ALL: # BB#0: 564 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm0[0,0,2,2] 565 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 566 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,0,0] 567 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 568 ; ALL-NEXT: retq 569 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 6, i32 6, i32 4, i32 4> 570 ret <8 x double> %shuffle 571 } 572 573 define <8 x double> @shuffle_v8f64_10324567(<8 x double> %a, <8 x double> %b) { 574 ; ALL-LABEL: shuffle_v8f64_10324567: 575 ; ALL: # BB#0: 576 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2] 577 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 578 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 579 ; ALL-NEXT: retq 580 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7> 581 ret <8 x double> %shuffle 582 } 583 584 define <8 x double> @shuffle_v8f64_11334567(<8 x double> %a, <8 x double> %b) { 585 ; ALL-LABEL: shuffle_v8f64_11334567: 586 ; ALL: # BB#0: 587 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3] 588 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 589 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 590 ; ALL-NEXT: retq 591 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 4, i32 5, i32 6, i32 7> 592 ret <8 x double> %shuffle 593 } 594 595 define <8 x double> @shuffle_v8f64_01235467(<8 x double> %a, <8 x double> %b) { 596 ; ALL-LABEL: shuffle_v8f64_01235467: 597 ; ALL: # BB#0: 598 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 599 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,3] 600 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0 601 ; ALL-NEXT: retq 602 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7> 603 ret <8 x double> %shuffle 604 } 605 606 define <8 x double> @shuffle_v8f64_01235466(<8 x double> %a, <8 x double> %b) { 607 ; ALL-LABEL: shuffle_v8f64_01235466: 608 ; ALL: # BB#0: 609 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 610 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2] 611 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0 612 ; ALL-NEXT: retq 613 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 6> 614 ret <8 x double> %shuffle 615 } 616 617 define <8 x double> @shuffle_v8f64_002u6u44(<8 x double> %a, <8 x double> %b) { 618 ; ALL-LABEL: shuffle_v8f64_002u6u44: 619 ; ALL: # BB#0: 620 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm0[0,0,2,2] 621 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 622 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,1,0,0] 623 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 624 ; ALL-NEXT: retq 625 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 undef, i32 6, i32 undef, i32 4, i32 4> 626 ret <8 x double> %shuffle 627 } 628 629 define <8 x double> @shuffle_v8f64_00uu66uu(<8 x double> %a, <8 x double> %b) { 630 ; ALL-LABEL: shuffle_v8f64_00uu66uu: 631 ; ALL: # BB#0: 632 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm1 633 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 634 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[2,2,2,3] 635 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 636 ; ALL-NEXT: retq 637 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 0, i32 undef, i32 undef, i32 6, i32 6, i32 undef, i32 undef> 638 ret <8 x double> %shuffle 639 } 640 641 define <8 x double> @shuffle_v8f64_103245uu(<8 x double> %a, <8 x double> %b) { 642 ; ALL-LABEL: shuffle_v8f64_103245uu: 643 ; ALL: # BB#0: 644 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,0,3,2] 645 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 646 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 647 ; ALL-NEXT: retq 648 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 undef, i32 undef> 649 ret <8 x double> %shuffle 650 } 651 652 define <8 x double> @shuffle_v8f64_1133uu67(<8 x double> %a, <8 x double> %b) { 653 ; ALL-LABEL: shuffle_v8f64_1133uu67: 654 ; ALL: # BB#0: 655 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm0[1,1,3,3] 656 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm0 657 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm1, %zmm0 658 ; ALL-NEXT: retq 659 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 undef, i32 6, i32 7> 660 ret <8 x double> %shuffle 661 } 662 663 define <8 x double> @shuffle_v8f64_0uu354uu(<8 x double> %a, <8 x double> %b) { 664 ; ALL-LABEL: shuffle_v8f64_0uu354uu: 665 ; ALL: # BB#0: 666 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 667 ; ALL-NEXT: vpermilpd {{.*#+}} ymm1 = ymm1[1,0,2,2] 668 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0 669 ; ALL-NEXT: retq 670 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 0, i32 undef, i32 undef, i32 3, i32 5, i32 4, i32 undef, i32 undef> 671 ret <8 x double> %shuffle 672 } 673 674 define <8 x double> @shuffle_v8f64_uuu3uu66(<8 x double> %a, <8 x double> %b) { 675 ; ALL-LABEL: shuffle_v8f64_uuu3uu66: 676 ; ALL: # BB#0: 677 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm1 678 ; ALL-NEXT: vmovddup {{.*#+}} ymm1 = ymm1[0,0,2,2] 679 ; ALL-NEXT: vinsertf64x4 $1, %ymm1, %zmm0, %zmm0 680 ; ALL-NEXT: retq 681 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 3, i32 undef, i32 undef, i32 6, i32 6> 682 ret <8 x double> %shuffle 683 } 684 685 define <8 x double> @shuffle_v8f64_c348cda0(<8 x double> %a, <8 x double> %b) { 686 ; ALL-LABEL: shuffle_v8f64_c348cda0: 687 ; ALL: # BB#0: 688 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2 689 ; ALL-NEXT: vperm2f128 {{.*#+}} ymm2 = ymm0[2,3],ymm2[0,1] 690 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm3 691 ; ALL-NEXT: vbroadcastsd %xmm1, %ymm4 692 ; ALL-NEXT: vblendpd {{.*#+}} ymm4 = ymm3[0,1,2],ymm4[3] 693 ; ALL-NEXT: vblendpd {{.*#+}} ymm2 = ymm4[0],ymm2[1,2],ymm4[3] 694 ; ALL-NEXT: vblendpd {{.*#+}} ymm1 = ymm3[0,1],ymm1[2],ymm3[3] 695 ; ALL-NEXT: vbroadcastsd %xmm0, %ymm0 696 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3] 697 ; ALL-NEXT: vinsertf64x4 $1, %ymm0, %zmm2, %zmm0 698 ; ALL-NEXT: retq 699 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 12, i32 3, i32 4, i32 8, i32 12, i32 13, i32 10, i32 0> 700 ret <8 x double> %shuffle 701 } 702 703 define <8 x double> @shuffle_v8f64_f511235a(<8 x double> %a, <8 x double> %b) { 704 ; ALL-LABEL: shuffle_v8f64_f511235a: 705 ; ALL: # BB#0: 706 ; ALL-NEXT: vextractf64x4 $1, %zmm0, %ymm2 707 ; ALL-NEXT: vblendpd {{.*#+}} ymm3 = ymm0[0],ymm2[1],ymm0[2,3] 708 ; ALL-NEXT: vpermpd {{.*#+}} ymm3 = ymm3[2,3,1,3] 709 ; ALL-NEXT: vmovddup {{.*#+}} ymm4 = ymm1[0,0,2,2] 710 ; ALL-NEXT: vblendpd {{.*#+}} ymm3 = ymm3[0,1,2],ymm4[3] 711 ; ALL-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,1,1,1] 712 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm0[0],ymm2[1],ymm0[2,3] 713 ; ALL-NEXT: vextractf64x4 $1, %zmm1, %ymm1 714 ; ALL-NEXT: vpermpd {{.*#+}} ymm1 = ymm1[3,1,2,3] 715 ; ALL-NEXT: vblendpd {{.*#+}} ymm0 = ymm1[0],ymm0[1,2,3] 716 ; ALL-NEXT: vinsertf64x4 $1, %ymm3, %zmm0, %zmm0 717 ; ALL-NEXT: retq 718 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32> <i32 15, i32 5, i32 1, i32 1, i32 2, i32 3, i32 5, i32 10> 719 ret <8 x double> %shuffle 720 } 721 722 define <8 x i64> @shuffle_v8i64_00000000(<8 x i64> %a, <8 x i64> %b) { 723 ; ALL-LABEL: shuffle_v8i64_00000000: 724 ; ALL: # BB#0: 725 ; ALL-NEXT: vpbroadcastq %xmm0, %zmm0 726 ; ALL-NEXT: retq 727 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 728 ret <8 x i64> %shuffle 729 } 730 731 define <8 x i64> @shuffle_v8i64_00000010(<8 x i64> %a, <8 x i64> %b) { 732 ; ALL-LABEL: shuffle_v8i64_00000010: 733 ; ALL: # BB#0: 734 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1 735 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,1,0] 736 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 737 ; ALL-NEXT: retq 738 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 1, i32 0> 739 ret <8 x i64> %shuffle 740 } 741 742 define <8 x i64> @shuffle_v8i64_00000200(<8 x i64> %a, <8 x i64> %b) { 743 ; ALL-LABEL: shuffle_v8i64_00000200: 744 ; ALL: # BB#0: 745 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1 746 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,0,0] 747 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 748 ; ALL-NEXT: retq 749 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 0, i32 2, i32 0, i32 0> 750 ret <8 x i64> %shuffle 751 } 752 753 define <8 x i64> @shuffle_v8i64_00003000(<8 x i64> %a, <8 x i64> %b) { 754 ; ALL-LABEL: shuffle_v8i64_00003000: 755 ; ALL: # BB#0: 756 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1 757 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,0,0,0] 758 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 759 ; ALL-NEXT: retq 760 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 3, i32 0, i32 0, i32 0> 761 ret <8 x i64> %shuffle 762 } 763 764 define <8 x i64> @shuffle_v8i64_00040000(<8 x i64> %a, <8 x i64> %b) { 765 ; ALL-LABEL: shuffle_v8i64_00040000: 766 ; ALL: # BB#0: 767 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 768 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1 769 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0 770 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3,4,5],ymm1[6,7] 771 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 772 ; ALL-NEXT: retq 773 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 4, i32 0, i32 0, i32 0, i32 0> 774 ret <8 x i64> %shuffle 775 } 776 777 define <8 x i64> @shuffle_v8i64_00500000(<8 x i64> %a, <8 x i64> %b) { 778 ; ALL-LABEL: shuffle_v8i64_00500000: 779 ; ALL: # BB#0: 780 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 781 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1],ymm1[2,3],ymm0[4,5,6,7] 782 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,1,0] 783 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0 784 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 785 ; ALL-NEXT: retq 786 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 5, i32 0, i32 0, i32 0, i32 0, i32 0> 787 ret <8 x i64> %shuffle 788 } 789 790 define <8 x i64> @shuffle_v8i64_06000000(<8 x i64> %a, <8 x i64> %b) { 791 ; ALL-LABEL: shuffle_v8i64_06000000: 792 ; ALL: # BB#0: 793 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 794 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3],ymm1[4,5],ymm0[6,7] 795 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,0,0] 796 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0 797 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 798 ; ALL-NEXT: retq 799 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 6, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 800 ret <8 x i64> %shuffle 801 } 802 803 define <8 x i64> @shuffle_v8i64_70000000(<8 x i64> %a, <8 x i64> %b) { 804 ; ALL-LABEL: shuffle_v8i64_70000000: 805 ; ALL: # BB#0: 806 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 807 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm0[0,1,2,3,4,5],ymm1[6,7] 808 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[3,0,0,0] 809 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0 810 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 811 ; ALL-NEXT: retq 812 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 7, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0, i32 0> 813 ret <8 x i64> %shuffle 814 } 815 816 define <8 x i64> @shuffle_v8i64_01014545(<8 x i64> %a, <8 x i64> %b) { 817 ; ALL-LABEL: shuffle_v8i64_01014545: 818 ; ALL: # BB#0: 819 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 820 ; ALL-NEXT: vinserti128 $1, %xmm1, %ymm1, %ymm1 821 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0 822 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 823 ; ALL-NEXT: retq 824 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 0, i32 1, i32 4, i32 5, i32 4, i32 5> 825 ret <8 x i64> %shuffle 826 } 827 828 define <8 x i64> @shuffle_v8i64_00112233(<8 x i64> %a, <8 x i64> %b) { 829 ; ALL-LABEL: shuffle_v8i64_00112233: 830 ; ALL: # BB#0: 831 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,1,1] 832 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,3,3] 833 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 834 ; ALL-NEXT: retq 835 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 1, i32 1, i32 2, i32 2, i32 3, i32 3> 836 ret <8 x i64> %shuffle 837 } 838 839 define <8 x i64> @shuffle_v8i64_00001111(<8 x i64> %a, <8 x i64> %b) { 840 ; ALL-LABEL: shuffle_v8i64_00001111: 841 ; ALL: # BB#0: 842 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1 843 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,1,1,1] 844 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 845 ; ALL-NEXT: retq 846 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 0, i32 1, i32 1, i32 1, i32 1> 847 ret <8 x i64> %shuffle 848 } 849 850 define <8 x i64> @shuffle_v8i64_81a3c5e7(<8 x i64> %a, <8 x i64> %b) { 851 ; ALL-LABEL: shuffle_v8i64_81a3c5e7: 852 ; ALL: # BB#0: 853 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2 854 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3 855 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7] 856 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5],ymm0[6,7] 857 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 858 ; ALL-NEXT: retq 859 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 8, i32 1, i32 10, i32 3, i32 12, i32 5, i32 14, i32 7> 860 ret <8 x i64> %shuffle 861 } 862 863 define <8 x i64> @shuffle_v8i64_08080808(<8 x i64> %a, <8 x i64> %b) { 864 ; ALL-LABEL: shuffle_v8i64_08080808: 865 ; ALL: # BB#0: 866 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0 867 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1 868 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7] 869 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm0, %zmm0 870 ; ALL-NEXT: retq 871 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 0, i32 8, i32 0, i32 8> 872 ret <8 x i64> %shuffle 873 } 874 875 define <8 x i64> @shuffle_v8i64_08084c4c(<8 x i64> %a, <8 x i64> %b) { 876 ; ALL-LABEL: shuffle_v8i64_08084c4c: 877 ; ALL: # BB#0: 878 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2 879 ; ALL-NEXT: vinserti128 $1, %xmm2, %ymm2, %ymm2 880 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3 881 ; ALL-NEXT: vpbroadcastq %xmm3, %ymm3 882 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm2[0,1],ymm3[2,3],ymm2[4,5],ymm3[6,7] 883 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm0 884 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1 885 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7] 886 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 887 ; ALL-NEXT: retq 888 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 0, i32 8, i32 4, i32 12, i32 4, i32 12> 889 ret <8 x i64> %shuffle 890 } 891 892 define <8 x i64> @shuffle_v8i64_8823cc67(<8 x i64> %a, <8 x i64> %b) { 893 ; ALL-LABEL: shuffle_v8i64_8823cc67: 894 ; ALL: # BB#0: 895 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm2 896 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm3 897 ; ALL-NEXT: vpbroadcastq %xmm3, %ymm3 898 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1,2,3],ymm2[4,5,6,7] 899 ; ALL-NEXT: vpbroadcastq %xmm1, %ymm1 900 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7] 901 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 902 ; ALL-NEXT: retq 903 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 8, i32 8, i32 2, i32 3, i32 12, i32 12, i32 6, i32 7> 904 ret <8 x i64> %shuffle 905 } 906 907 define <8 x i64> @shuffle_v8i64_9832dc76(<8 x i64> %a, <8 x i64> %b) { 908 ; ALL-LABEL: shuffle_v8i64_9832dc76: 909 ; ALL: # BB#0: 910 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm1[0,1,2,3],ymm0[4,5,6,7] 911 ; ALL-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5] 912 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 913 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm1 914 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7] 915 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5] 916 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0 917 ; ALL-NEXT: retq 918 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 9, i32 8, i32 3, i32 2, i32 13, i32 12, i32 7, i32 6> 919 ret <8 x i64> %shuffle 920 } 921 922 define <8 x i64> @shuffle_v8i64_9810dc54(<8 x i64> %a, <8 x i64> %b) { 923 ; ALL-LABEL: shuffle_v8i64_9810dc54: 924 ; ALL: # BB#0: 925 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm2 926 ; ALL-NEXT: vpshufd {{.*#+}} ymm2 = ymm2[2,3,0,1,6,7,4,5] 927 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm1 928 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 929 ; ALL-NEXT: vinserti128 $1, %xmm0, %ymm1, %ymm0 930 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5] 931 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0 932 ; ALL-NEXT: retq 933 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 9, i32 8, i32 1, i32 0, i32 13, i32 12, i32 5, i32 4> 934 ret <8 x i64> %shuffle 935 } 936 937 define <8 x i64> @shuffle_v8i64_08194c5d(<8 x i64> %a, <8 x i64> %b) { 938 ; ALL-LABEL: shuffle_v8i64_08194c5d: 939 ; ALL: # BB#0: 940 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2 941 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,0,2,1] 942 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm3 943 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm3[0,1,1,3] 944 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7] 945 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1] 946 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3] 947 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7] 948 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 949 ; ALL-NEXT: retq 950 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 4, i32 12, i32 5, i32 13> 951 ret <8 x i64> %shuffle 952 } 953 954 define <8 x i64> @shuffle_v8i64_2a3b6e7f(<8 x i64> %a, <8 x i64> %b) { 955 ; ALL-LABEL: shuffle_v8i64_2a3b6e7f: 956 ; ALL: # BB#0: 957 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2 958 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm2[0,2,2,3] 959 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm3 960 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm3[2,1,3,3] 961 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7] 962 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,2,2,3] 963 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,3,3] 964 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7] 965 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 966 ; ALL-NEXT: retq 967 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 2, i32 10, i32 3, i32 11, i32 6, i32 14, i32 7, i32 15> 968 ret <8 x i64> %shuffle 969 } 970 971 define <8 x i64> @shuffle_v8i64_08192a3b(<8 x i64> %a, <8 x i64> %b) { 972 ; ALL-LABEL: shuffle_v8i64_08192a3b: 973 ; ALL: # BB#0: 974 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm1[0,2,2,3] 975 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,1,3,3] 976 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7] 977 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[0,0,2,1] 978 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3] 979 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7] 980 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 981 ; ALL-NEXT: retq 982 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 1, i32 9, i32 2, i32 10, i32 3, i32 11> 983 ret <8 x i64> %shuffle 984 } 985 986 define <8 x i64> @shuffle_v8i64_08991abb(<8 x i64> %a, <8 x i64> %b) { 987 ; ALL-LABEL: shuffle_v8i64_08991abb: 988 ; ALL: # BB#0: 989 ; ALL-NEXT: vpermq {{.*#+}} ymm2 = ymm1[0,0,1,1] 990 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm0[0,1],ymm2[2,3,4,5,6,7] 991 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1],ymm0[2,3],ymm1[4,5,6,7] 992 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,2,3,3] 993 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm2, %zmm0 994 ; ALL-NEXT: retq 995 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 8, i32 9, i32 9, i32 1, i32 10, i32 11, i32 11> 996 ret <8 x i64> %shuffle 997 } 998 999 define <8 x i64> @shuffle_v8i64_091b2d3f(<8 x i64> %a, <8 x i64> %b) { 1000 ; ALL-LABEL: shuffle_v8i64_091b2d3f: 1001 ; ALL: # BB#0: 1002 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2 1003 ; ALL-NEXT: vpermq {{.*#+}} ymm3 = ymm0[2,1,3,3] 1004 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3],ymm3[4,5],ymm2[6,7] 1005 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,1,1,3] 1006 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3],ymm0[4,5],ymm1[6,7] 1007 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 1008 ; ALL-NEXT: retq 1009 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 9, i32 1, i32 11, i32 2, i32 13, i32 3, i32 15> 1010 ret <8 x i64> %shuffle 1011 } 1012 1013 define <8 x i64> @shuffle_v8i64_09ab1def(<8 x i64> %a, <8 x i64> %b) { 1014 ; ALL-LABEL: shuffle_v8i64_09ab1def: 1015 ; ALL: # BB#0: 1016 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2 1017 ; ALL-NEXT: vpshufd {{.*#+}} ymm3 = ymm0[2,3,2,3,6,7,6,7] 1018 ; ALL-NEXT: vpblendd {{.*#+}} ymm2 = ymm3[0,1],ymm2[2,3,4,5,6,7] 1019 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7] 1020 ; ALL-NEXT: vinserti64x4 $1, %ymm2, %zmm0, %zmm0 1021 ; ALL-NEXT: retq 1022 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 1, i32 13, i32 14, i32 15> 1023 ret <8 x i64> %shuffle 1024 } 1025 1026 define <8 x i64> @shuffle_v8i64_00014445(<8 x i64> %a, <8 x i64> %b) { 1027 ; ALL-LABEL: shuffle_v8i64_00014445: 1028 ; ALL: # BB#0: 1029 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,0,1] 1030 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1031 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,0,1] 1032 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1033 ; ALL-NEXT: retq 1034 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 4, i32 4, i32 4, i32 5> 1035 ret <8 x i64> %shuffle 1036 } 1037 1038 define <8 x i64> @shuffle_v8i64_00204464(<8 x i64> %a, <8 x i64> %b) { 1039 ; ALL-LABEL: shuffle_v8i64_00204464: 1040 ; ALL: # BB#0: 1041 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,0] 1042 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1043 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,2,0] 1044 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1045 ; ALL-NEXT: retq 1046 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 4, i32 6, i32 4> 1047 ret <8 x i64> %shuffle 1048 } 1049 1050 define <8 x i64> @shuffle_v8i64_03004744(<8 x i64> %a, <8 x i64> %b) { 1051 ; ALL-LABEL: shuffle_v8i64_03004744: 1052 ; ALL: # BB#0: 1053 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,3,0,0] 1054 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1055 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,3,0,0] 1056 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1057 ; ALL-NEXT: retq 1058 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 7, i32 4, i32 4> 1059 ret <8 x i64> %shuffle 1060 } 1061 1062 define <8 x i64> @shuffle_v8i64_10005444(<8 x i64> %a, <8 x i64> %b) { 1063 ; ALL-LABEL: shuffle_v8i64_10005444: 1064 ; ALL: # BB#0: 1065 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,0,0] 1066 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1067 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,0,0] 1068 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1069 ; ALL-NEXT: retq 1070 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 5, i32 4, i32 4, i32 4> 1071 ret <8 x i64> %shuffle 1072 } 1073 1074 define <8 x i64> @shuffle_v8i64_22006644(<8 x i64> %a, <8 x i64> %b) { 1075 ; ALL-LABEL: shuffle_v8i64_22006644: 1076 ; ALL: # BB#0: 1077 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,2,0,0] 1078 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1079 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,0,0] 1080 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1081 ; ALL-NEXT: retq 1082 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 6, i32 4, i32 4> 1083 ret <8 x i64> %shuffle 1084 } 1085 1086 define <8 x i64> @shuffle_v8i64_33307774(<8 x i64> %a, <8 x i64> %b) { 1087 ; ALL-LABEL: shuffle_v8i64_33307774: 1088 ; ALL: # BB#0: 1089 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,3,3,0] 1090 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1091 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,3,3,0] 1092 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1093 ; ALL-NEXT: retq 1094 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 7, i32 7, i32 4> 1095 ret <8 x i64> %shuffle 1096 } 1097 1098 define <8 x i64> @shuffle_v8i64_32107654(<8 x i64> %a, <8 x i64> %b) { 1099 ; ALL-LABEL: shuffle_v8i64_32107654: 1100 ; ALL: # BB#0: 1101 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,2,1,0] 1102 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1103 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,2,1,0] 1104 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1105 ; ALL-NEXT: retq 1106 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4> 1107 ret <8 x i64> %shuffle 1108 } 1109 1110 define <8 x i64> @shuffle_v8i64_00234467(<8 x i64> %a, <8 x i64> %b) { 1111 ; ALL-LABEL: shuffle_v8i64_00234467: 1112 ; ALL: # BB#0: 1113 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,3] 1114 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1115 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,2,3] 1116 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1117 ; ALL-NEXT: retq 1118 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 4, i32 4, i32 6, i32 7> 1119 ret <8 x i64> %shuffle 1120 } 1121 1122 define <8 x i64> @shuffle_v8i64_00224466(<8 x i64> %a, <8 x i64> %b) { 1123 ; ALL-LABEL: shuffle_v8i64_00224466: 1124 ; ALL: # BB#0: 1125 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5] 1126 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1127 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[0,1,0,1,4,5,4,5] 1128 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1129 ; ALL-NEXT: retq 1130 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 4, i32 4, i32 6, i32 6> 1131 ret <8 x i64> %shuffle 1132 } 1133 1134 define <8 x i64> @shuffle_v8i64_10325476(<8 x i64> %a, <8 x i64> %b) { 1135 ; ALL-LABEL: shuffle_v8i64_10325476: 1136 ; ALL: # BB#0: 1137 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5] 1138 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1139 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,0,1,6,7,4,5] 1140 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1141 ; ALL-NEXT: retq 1142 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 5, i32 4, i32 7, i32 6> 1143 ret <8 x i64> %shuffle 1144 } 1145 1146 define <8 x i64> @shuffle_v8i64_11335577(<8 x i64> %a, <8 x i64> %b) { 1147 ; ALL-LABEL: shuffle_v8i64_11335577: 1148 ; ALL: # BB#0: 1149 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7] 1150 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1151 ; ALL-NEXT: vpshufd {{.*#+}} ymm0 = ymm0[2,3,2,3,6,7,6,7] 1152 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1153 ; ALL-NEXT: retq 1154 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 5, i32 5, i32 7, i32 7> 1155 ret <8 x i64> %shuffle 1156 } 1157 1158 define <8 x i64> @shuffle_v8i64_10235467(<8 x i64> %a, <8 x i64> %b) { 1159 ; ALL-LABEL: shuffle_v8i64_10235467: 1160 ; ALL: # BB#0: 1161 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,2,3] 1162 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1163 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,2,3] 1164 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1165 ; ALL-NEXT: retq 1166 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7> 1167 ret <8 x i64> %shuffle 1168 } 1169 1170 define <8 x i64> @shuffle_v8i64_10225466(<8 x i64> %a, <8 x i64> %b) { 1171 ; ALL-LABEL: shuffle_v8i64_10225466: 1172 ; ALL: # BB#0: 1173 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,2,2] 1174 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1175 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,2,2] 1176 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1177 ; ALL-NEXT: retq 1178 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 2, i32 2, i32 5, i32 4, i32 6, i32 6> 1179 ret <8 x i64> %shuffle 1180 } 1181 1182 define <8 x i64> @shuffle_v8i64_00015444(<8 x i64> %a, <8 x i64> %b) { 1183 ; ALL-LABEL: shuffle_v8i64_00015444: 1184 ; ALL: # BB#0: 1185 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,0,1] 1186 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1187 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[1,0,0,0] 1188 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1189 ; ALL-NEXT: retq 1190 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 0, i32 1, i32 5, i32 4, i32 4, i32 4> 1191 ret <8 x i64> %shuffle 1192 } 1193 1194 define <8 x i64> @shuffle_v8i64_00204644(<8 x i64> %a, <8 x i64> %b) { 1195 ; ALL-LABEL: shuffle_v8i64_00204644: 1196 ; ALL: # BB#0: 1197 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,0] 1198 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1199 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,2,0,0] 1200 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1201 ; ALL-NEXT: retq 1202 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 0, i32 4, i32 6, i32 4, i32 4> 1203 ret <8 x i64> %shuffle 1204 } 1205 1206 define <8 x i64> @shuffle_v8i64_03004474(<8 x i64> %a, <8 x i64> %b) { 1207 ; ALL-LABEL: shuffle_v8i64_03004474: 1208 ; ALL: # BB#0: 1209 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,3,0,0] 1210 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1211 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[0,0,3,0] 1212 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1213 ; ALL-NEXT: retq 1214 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 3, i32 0, i32 0, i32 4, i32 4, i32 7, i32 4> 1215 ret <8 x i64> %shuffle 1216 } 1217 1218 define <8 x i64> @shuffle_v8i64_10004444(<8 x i64> %a, <8 x i64> %b) { 1219 ; ALL-LABEL: shuffle_v8i64_10004444: 1220 ; ALL: # BB#0: 1221 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[1,0,0,0] 1222 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1223 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm0 1224 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1225 ; ALL-NEXT: retq 1226 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 0, i32 0, i32 4, i32 4, i32 4, i32 4> 1227 ret <8 x i64> %shuffle 1228 } 1229 1230 define <8 x i64> @shuffle_v8i64_22006446(<8 x i64> %a, <8 x i64> %b) { 1231 ; ALL-LABEL: shuffle_v8i64_22006446: 1232 ; ALL: # BB#0: 1233 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[2,2,0,0] 1234 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1235 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,0,0,2] 1236 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1237 ; ALL-NEXT: retq 1238 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 2, i32 2, i32 0, i32 0, i32 6, i32 4, i32 4, i32 6> 1239 ret <8 x i64> %shuffle 1240 } 1241 1242 define <8 x i64> @shuffle_v8i64_33307474(<8 x i64> %a, <8 x i64> %b) { 1243 ; ALL-LABEL: shuffle_v8i64_33307474: 1244 ; ALL: # BB#0: 1245 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,3,3,0] 1246 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1247 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[3,0,3,0] 1248 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1249 ; ALL-NEXT: retq 1250 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 3, i32 3, i32 0, i32 7, i32 4, i32 7, i32 4> 1251 ret <8 x i64> %shuffle 1252 } 1253 1254 define <8 x i64> @shuffle_v8i64_32104567(<8 x i64> %a, <8 x i64> %b) { 1255 ; ALL-LABEL: shuffle_v8i64_32104567: 1256 ; ALL: # BB#0: 1257 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[3,2,1,0] 1258 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1259 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1260 ; ALL-NEXT: retq 1261 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 4, i32 5, i32 6, i32 7> 1262 ret <8 x i64> %shuffle 1263 } 1264 1265 define <8 x i64> @shuffle_v8i64_00236744(<8 x i64> %a, <8 x i64> %b) { 1266 ; ALL-LABEL: shuffle_v8i64_00236744: 1267 ; ALL: # BB#0: 1268 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm0[0,0,2,3] 1269 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1270 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,3,0,0] 1271 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1272 ; ALL-NEXT: retq 1273 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 3, i32 6, i32 7, i32 4, i32 4> 1274 ret <8 x i64> %shuffle 1275 } 1276 1277 define <8 x i64> @shuffle_v8i64_00226644(<8 x i64> %a, <8 x i64> %b) { 1278 ; ALL-LABEL: shuffle_v8i64_00226644: 1279 ; ALL: # BB#0: 1280 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5] 1281 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1282 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,0,0] 1283 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1284 ; ALL-NEXT: retq 1285 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 2, i32 6, i32 6, i32 4, i32 4> 1286 ret <8 x i64> %shuffle 1287 } 1288 1289 define <8 x i64> @shuffle_v8i64_10324567(<8 x i64> %a, <8 x i64> %b) { 1290 ; ALL-LABEL: shuffle_v8i64_10324567: 1291 ; ALL: # BB#0: 1292 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5] 1293 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1294 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1295 ; ALL-NEXT: retq 1296 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 6, i32 7> 1297 ret <8 x i64> %shuffle 1298 } 1299 1300 define <8 x i64> @shuffle_v8i64_11334567(<8 x i64> %a, <8 x i64> %b) { 1301 ; ALL-LABEL: shuffle_v8i64_11334567: 1302 ; ALL: # BB#0: 1303 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7] 1304 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1305 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1306 ; ALL-NEXT: retq 1307 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 4, i32 5, i32 6, i32 7> 1308 ret <8 x i64> %shuffle 1309 } 1310 1311 define <8 x i64> @shuffle_v8i64_01235467(<8 x i64> %a, <8 x i64> %b) { 1312 ; ALL-LABEL: shuffle_v8i64_01235467: 1313 ; ALL: # BB#0: 1314 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 1315 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[1,0,2,3] 1316 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 1317 ; ALL-NEXT: retq 1318 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 7> 1319 ret <8 x i64> %shuffle 1320 } 1321 1322 define <8 x i64> @shuffle_v8i64_01235466(<8 x i64> %a, <8 x i64> %b) { 1323 ; ALL-LABEL: shuffle_v8i64_01235466: 1324 ; ALL: # BB#0: 1325 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 1326 ; ALL-NEXT: vpermq {{.*#+}} ymm1 = ymm1[1,0,2,2] 1327 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 1328 ; ALL-NEXT: retq 1329 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 5, i32 4, i32 6, i32 6> 1330 ret <8 x i64> %shuffle 1331 } 1332 1333 define <8 x i64> @shuffle_v8i64_002u6u44(<8 x i64> %a, <8 x i64> %b) { 1334 ; ALL-LABEL: shuffle_v8i64_002u6u44: 1335 ; ALL: # BB#0: 1336 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[0,1,0,1,4,5,4,5] 1337 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1338 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,1,0,0] 1339 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1340 ; ALL-NEXT: retq 1341 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 2, i32 undef, i32 6, i32 undef, i32 4, i32 4> 1342 ret <8 x i64> %shuffle 1343 } 1344 1345 define <8 x i64> @shuffle_v8i64_00uu66uu(<8 x i64> %a, <8 x i64> %b) { 1346 ; ALL-LABEL: shuffle_v8i64_00uu66uu: 1347 ; ALL: # BB#0: 1348 ; ALL-NEXT: vpbroadcastq %xmm0, %ymm1 1349 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1350 ; ALL-NEXT: vpermq {{.*#+}} ymm0 = ymm0[2,2,2,3] 1351 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1352 ; ALL-NEXT: retq 1353 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 0, i32 undef, i32 undef, i32 6, i32 6, i32 undef, i32 undef> 1354 ret <8 x i64> %shuffle 1355 } 1356 1357 define <8 x i64> @shuffle_v8i64_103245uu(<8 x i64> %a, <8 x i64> %b) { 1358 ; ALL-LABEL: shuffle_v8i64_103245uu: 1359 ; ALL: # BB#0: 1360 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,0,1,6,7,4,5] 1361 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1362 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1363 ; ALL-NEXT: retq 1364 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 0, i32 3, i32 2, i32 4, i32 5, i32 undef, i32 undef> 1365 ret <8 x i64> %shuffle 1366 } 1367 1368 define <8 x i64> @shuffle_v8i64_1133uu67(<8 x i64> %a, <8 x i64> %b) { 1369 ; ALL-LABEL: shuffle_v8i64_1133uu67: 1370 ; ALL: # BB#0: 1371 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm0[2,3,2,3,6,7,6,7] 1372 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1373 ; ALL-NEXT: vinserti64x4 $1, %ymm0, %zmm1, %zmm0 1374 ; ALL-NEXT: retq 1375 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 1, i32 1, i32 3, i32 3, i32 undef, i32 undef, i32 6, i32 7> 1376 ret <8 x i64> %shuffle 1377 } 1378 1379 define <8 x i64> @shuffle_v8i64_0uu354uu(<8 x i64> %a, <8 x i64> %b) { 1380 ; ALL-LABEL: shuffle_v8i64_0uu354uu: 1381 ; ALL: # BB#0: 1382 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 1383 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[2,3,0,1,6,7,4,5] 1384 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 1385 ; ALL-NEXT: retq 1386 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 0, i32 undef, i32 undef, i32 3, i32 5, i32 4, i32 undef, i32 undef> 1387 ret <8 x i64> %shuffle 1388 } 1389 1390 define <8 x i64> @shuffle_v8i64_uuu3uu66(<8 x i64> %a, <8 x i64> %b) { 1391 ; ALL-LABEL: shuffle_v8i64_uuu3uu66: 1392 ; ALL: # BB#0: 1393 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm1 1394 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,1,4,5,4,5] 1395 ; ALL-NEXT: vinserti64x4 $1, %ymm1, %zmm0, %zmm0 1396 ; ALL-NEXT: retq 1397 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 undef, i32 undef, i32 undef, i32 3, i32 undef, i32 undef, i32 6, i32 6> 1398 ret <8 x i64> %shuffle 1399 } 1400 1401 define <8 x i64> @shuffle_v8i64_6caa87e5(<8 x i64> %a, <8 x i64> %b) { 1402 ; ALL-LABEL: shuffle_v8i64_6caa87e5: 1403 ; ALL: # BB#0: 1404 ; ALL-NEXT: vextracti64x4 $1, %zmm0, %ymm0 1405 ; ALL-NEXT: vperm2i128 {{.*#+}} ymm0 = ymm0[2,3,0,1] 1406 ; ALL-NEXT: vextracti64x4 $1, %zmm1, %ymm2 1407 ; ALL-NEXT: vpblendd {{.*#+}} ymm3 = ymm1[0,1,2,3],ymm2[4,5],ymm1[6,7] 1408 ; ALL-NEXT: vpblendd {{.*#+}} ymm3 = ymm3[0,1],ymm0[2,3],ymm3[4,5],ymm0[6,7] 1409 ; ALL-NEXT: vpblendd {{.*#+}} ymm1 = ymm2[0,1,2,3],ymm1[4,5,6,7] 1410 ; ALL-NEXT: vpshufd {{.*#+}} ymm1 = ymm1[0,1,0,1,4,5,4,5] 1411 ; ALL-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1],ymm1[2,3,4,5,6,7] 1412 ; ALL-NEXT: vinserti64x4 $1, %ymm3, %zmm0, %zmm0 1413 ; ALL-NEXT: retq 1414 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32> <i32 6, i32 12, i32 10, i32 10, i32 8, i32 7, i32 14, i32 5> 1415 ret <8 x i64> %shuffle 1416 } 1417 1418 define <8 x double> @shuffle_v8f64_082a4c6e(<8 x double> %a, <8 x double> %b) { 1419 ; ALL-LABEL: shuffle_v8f64_082a4c6e: 1420 ; ALL: # BB#0: 1421 ; ALL-NEXT: vunpcklpd {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6] 1422 ; ALL-NEXT: retq 1423 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32><i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14> 1424 ret <8 x double> %shuffle 1425 } 1426 1427 define <8 x i64> @shuffle_v8i64_082a4c6e(<8 x i64> %a, <8 x i64> %b) { 1428 ; ALL-LABEL: shuffle_v8i64_082a4c6e: 1429 ; ALL: # BB#0: 1430 ; ALL-NEXT: vpunpcklqdq {{.*#+}} zmm0 = zmm0[0],zmm1[0],zmm0[2],zmm1[2],zmm0[4],zmm1[4],zmm0[6],zmm1[6] 1431 ; ALL-NEXT: retq 1432 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32><i32 0, i32 8, i32 2, i32 10, i32 4, i32 12, i32 6, i32 14> 1433 ret <8 x i64> %shuffle 1434 } 1435 1436 define <8 x double> @shuffle_v8f64_193b5d7f(<8 x double> %a, <8 x double> %b) { 1437 ; ALL-LABEL: shuffle_v8f64_193b5d7f: 1438 ; ALL: # BB#0: 1439 ; ALL-NEXT: vunpckhpd {{.*#+}} zmm0 = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7] 1440 ; ALL-NEXT: retq 1441 %shuffle = shufflevector <8 x double> %a, <8 x double> %b, <8 x i32><i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15> 1442 ret <8 x double> %shuffle 1443 } 1444 1445 define <8 x i64> @shuffle_v8i64_193b5d7f(<8 x i64> %a, <8 x i64> %b) { 1446 ; ALL-LABEL: shuffle_v8i64_193b5d7f: 1447 ; ALL: # BB#0: 1448 ; ALL-NEXT: vpunpckhqdq {{.*#+}} zmm0 = zmm0[1],zmm1[1],zmm0[3],zmm1[3],zmm0[5],zmm1[5],zmm0[7],zmm1[7] 1449 ; ALL-NEXT: retq 1450 %shuffle = shufflevector <8 x i64> %a, <8 x i64> %b, <8 x i32><i32 1, i32 9, i32 3, i32 11, i32 5, i32 13, i32 7, i32 15> 1451 ret <8 x i64> %shuffle 1452 } 1453
