1 ; RUN: llc -march=arm64 -aarch64-neon-syntax=apple < %s -asm-verbose=false -mcpu=cyclone | FileCheck %s 2 3 define signext i8 @test_vaddv_s8(<8 x i8> %a1) { 4 ; CHECK-LABEL: test_vaddv_s8: 5 ; CHECK: addv.8b b[[REGNUM:[0-9]+]], v0 6 ; CHECK-NEXT: smov.b w0, v[[REGNUM]][0] 7 ; CHECK-NEXT: ret 8 entry: 9 %vaddv.i = tail call i32 @llvm.aarch64.neon.saddv.i32.v8i8(<8 x i8> %a1) 10 %0 = trunc i32 %vaddv.i to i8 11 ret i8 %0 12 } 13 14 define <8 x i8> @test_vaddv_s8_used_by_laneop(<8 x i8> %a1, <8 x i8> %a2) { 15 ; CHECK-LABEL: test_vaddv_s8_used_by_laneop: 16 ; CHECK: addv.8b b[[REGNUM:[0-9]+]], v1 17 ; CHECK-NEXT: ins.b v0[3], v[[REGNUM]][0] 18 ; CHECK-NEXT: ret 19 entry: 20 %0 = tail call i32 @llvm.aarch64.neon.saddv.i32.v8i8(<8 x i8> %a2) 21 %1 = trunc i32 %0 to i8 22 %2 = insertelement <8 x i8> %a1, i8 %1, i32 3 23 ret <8 x i8> %2 24 } 25 26 define signext i16 @test_vaddv_s16(<4 x i16> %a1) { 27 ; CHECK-LABEL: test_vaddv_s16: 28 ; CHECK: addv.4h h[[REGNUM:[0-9]+]], v0 29 ; CHECK-NEXT: smov.h w0, v[[REGNUM]][0] 30 ; CHECK-NEXT: ret 31 entry: 32 %vaddv.i = tail call i32 @llvm.aarch64.neon.saddv.i32.v4i16(<4 x i16> %a1) 33 %0 = trunc i32 %vaddv.i to i16 34 ret i16 %0 35 } 36 37 define <4 x i16> @test_vaddv_s16_used_by_laneop(<4 x i16> %a1, <4 x i16> %a2) { 38 ; CHECK-LABEL: test_vaddv_s16_used_by_laneop: 39 ; CHECK: addv.4h h[[REGNUM:[0-9]+]], v1 40 ; CHECK-NEXT: ins.h v0[3], v[[REGNUM]][0] 41 ; CHECK-NEXT: ret 42 entry: 43 %0 = tail call i32 @llvm.aarch64.neon.saddv.i32.v4i16(<4 x i16> %a2) 44 %1 = trunc i32 %0 to i16 45 %2 = insertelement <4 x i16> %a1, i16 %1, i32 3 46 ret <4 x i16> %2 47 } 48 49 define i32 @test_vaddv_s32(<2 x i32> %a1) { 50 ; CHECK-LABEL: test_vaddv_s32: 51 ; 2 x i32 is not supported by the ISA, thus, this is a special case 52 ; CHECK: addp.2s v[[REGNUM:[0-9]+]], v0, v0 53 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 54 ; CHECK-NEXT: ret 55 entry: 56 %vaddv.i = tail call i32 @llvm.aarch64.neon.saddv.i32.v2i32(<2 x i32> %a1) 57 ret i32 %vaddv.i 58 } 59 60 define <2 x i32> @test_vaddv_s32_used_by_laneop(<2 x i32> %a1, <2 x i32> %a2) { 61 ; CHECK-LABEL: test_vaddv_s32_used_by_laneop: 62 ; CHECK: addp.2s v[[REGNUM:[0-9]+]], v1, v1 63 ; CHECK-NEXT: ins.s v0[1], v[[REGNUM]][0] 64 ; CHECK-NEXT: ret 65 entry: 66 %0 = tail call i32 @llvm.aarch64.neon.saddv.i32.v2i32(<2 x i32> %a2) 67 %1 = insertelement <2 x i32> %a1, i32 %0, i32 1 68 ret <2 x i32> %1 69 } 70 71 define i64 @test_vaddv_s64(<2 x i64> %a1) { 72 ; CHECK-LABEL: test_vaddv_s64: 73 ; CHECK: addp.2d [[REGNUM:d[0-9]+]], v0 74 ; CHECK-NEXT: fmov x0, [[REGNUM]] 75 ; CHECK-NEXT: ret 76 entry: 77 %vaddv.i = tail call i64 @llvm.aarch64.neon.saddv.i64.v2i64(<2 x i64> %a1) 78 ret i64 %vaddv.i 79 } 80 81 define <2 x i64> @test_vaddv_s64_used_by_laneop(<2 x i64> %a1, <2 x i64> %a2) { 82 ; CHECK-LABEL: test_vaddv_s64_used_by_laneop: 83 ; CHECK: addp.2d d[[REGNUM:[0-9]+]], v1 84 ; CHECK-NEXT: ins.d v0[1], v[[REGNUM]][0] 85 ; CHECK-NEXT: ret 86 entry: 87 %0 = tail call i64 @llvm.aarch64.neon.saddv.i64.v2i64(<2 x i64> %a2) 88 %1 = insertelement <2 x i64> %a1, i64 %0, i64 1 89 ret <2 x i64> %1 90 } 91 92 define zeroext i8 @test_vaddv_u8(<8 x i8> %a1) { 93 ; CHECK-LABEL: test_vaddv_u8: 94 ; CHECK: addv.8b b[[REGNUM:[0-9]+]], v0 95 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 96 ; CHECK-NEXT: ret 97 entry: 98 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v8i8(<8 x i8> %a1) 99 %0 = trunc i32 %vaddv.i to i8 100 ret i8 %0 101 } 102 103 define <8 x i8> @test_vaddv_u8_used_by_laneop(<8 x i8> %a1, <8 x i8> %a2) { 104 ; CHECK-LABEL: test_vaddv_u8_used_by_laneop: 105 ; CHECK: addv.8b b[[REGNUM:[0-9]+]], v1 106 ; CHECK-NEXT: ins.b v0[3], v[[REGNUM]][0] 107 ; CHECK-NEXT: ret 108 entry: 109 %0 = tail call i32 @llvm.aarch64.neon.uaddv.i32.v8i8(<8 x i8> %a2) 110 %1 = trunc i32 %0 to i8 111 %2 = insertelement <8 x i8> %a1, i8 %1, i32 3 112 ret <8 x i8> %2 113 } 114 115 define i32 @test_vaddv_u8_masked(<8 x i8> %a1) { 116 ; CHECK-LABEL: test_vaddv_u8_masked: 117 ; CHECK: addv.8b b[[REGNUM:[0-9]+]], v0 118 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 119 ; CHECK-NEXT: ret 120 entry: 121 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v8i8(<8 x i8> %a1) 122 %0 = and i32 %vaddv.i, 511 ; 0x1ff 123 ret i32 %0 124 } 125 126 define zeroext i16 @test_vaddv_u16(<4 x i16> %a1) { 127 ; CHECK-LABEL: test_vaddv_u16: 128 ; CHECK: addv.4h h[[REGNUM:[0-9]+]], v0 129 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 130 ; CHECK-NEXT: ret 131 entry: 132 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v4i16(<4 x i16> %a1) 133 %0 = trunc i32 %vaddv.i to i16 134 ret i16 %0 135 } 136 137 define <4 x i16> @test_vaddv_u16_used_by_laneop(<4 x i16> %a1, <4 x i16> %a2) { 138 ; CHECK-LABEL: test_vaddv_u16_used_by_laneop: 139 ; CHECK: addv.4h h[[REGNUM:[0-9]+]], v1 140 ; CHECK-NEXT: ins.h v0[3], v[[REGNUM]][0] 141 ; CHECK-NEXT: ret 142 entry: 143 %0 = tail call i32 @llvm.aarch64.neon.uaddv.i32.v4i16(<4 x i16> %a2) 144 %1 = trunc i32 %0 to i16 145 %2 = insertelement <4 x i16> %a1, i16 %1, i32 3 146 ret <4 x i16> %2 147 } 148 149 define i32 @test_vaddv_u16_masked(<4 x i16> %a1) { 150 ; CHECK-LABEL: test_vaddv_u16_masked: 151 ; CHECK: addv.4h h[[REGNUM:[0-9]+]], v0 152 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 153 ; CHECK-NEXT: ret 154 entry: 155 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v4i16(<4 x i16> %a1) 156 %0 = and i32 %vaddv.i, 3276799 ; 0x31ffff 157 ret i32 %0 158 } 159 160 define i32 @test_vaddv_u32(<2 x i32> %a1) { 161 ; CHECK-LABEL: test_vaddv_u32: 162 ; 2 x i32 is not supported by the ISA, thus, this is a special case 163 ; CHECK: addp.2s v[[REGNUM:[0-9]+]], v0, v0 164 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 165 ; CHECK-NEXT: ret 166 entry: 167 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v2i32(<2 x i32> %a1) 168 ret i32 %vaddv.i 169 } 170 171 define <2 x i32> @test_vaddv_u32_used_by_laneop(<2 x i32> %a1, <2 x i32> %a2) { 172 ; CHECK-LABEL: test_vaddv_u32_used_by_laneop: 173 ; CHECK: addp.2s v[[REGNUM:[0-9]+]], v1, v1 174 ; CHECK-NEXT: ins.s v0[1], v[[REGNUM]][0] 175 ; CHECK-NEXT: ret 176 entry: 177 %0 = tail call i32 @llvm.aarch64.neon.uaddv.i32.v2i32(<2 x i32> %a2) 178 %1 = insertelement <2 x i32> %a1, i32 %0, i32 1 179 ret <2 x i32> %1 180 } 181 182 define float @test_vaddv_f32(<2 x float> %a1) { 183 ; CHECK-LABEL: test_vaddv_f32: 184 ; CHECK: faddp.2s s0, v0 185 ; CHECK-NEXT: ret 186 entry: 187 %vaddv.i = tail call float @llvm.aarch64.neon.faddv.f32.v2f32(<2 x float> %a1) 188 ret float %vaddv.i 189 } 190 191 define float @test_vaddv_v4f32(<4 x float> %a1) { 192 ; CHECK-LABEL: test_vaddv_v4f32: 193 ; CHECK: faddp.4s [[REGNUM:v[0-9]+]], v0, v0 194 ; CHECK: faddp.2s s0, [[REGNUM]] 195 ; CHECK-NEXT: ret 196 entry: 197 %vaddv.i = tail call float @llvm.aarch64.neon.faddv.f32.v4f32(<4 x float> %a1) 198 ret float %vaddv.i 199 } 200 201 define double @test_vaddv_f64(<2 x double> %a1) { 202 ; CHECK-LABEL: test_vaddv_f64: 203 ; CHECK: faddp.2d d0, v0 204 ; CHECK-NEXT: ret 205 entry: 206 %vaddv.i = tail call double @llvm.aarch64.neon.faddv.f64.v2f64(<2 x double> %a1) 207 ret double %vaddv.i 208 } 209 210 define i64 @test_vaddv_u64(<2 x i64> %a1) { 211 ; CHECK-LABEL: test_vaddv_u64: 212 ; CHECK: addp.2d [[REGNUM:d[0-9]+]], v0 213 ; CHECK-NEXT: fmov x0, [[REGNUM]] 214 ; CHECK-NEXT: ret 215 entry: 216 %vaddv.i = tail call i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64> %a1) 217 ret i64 %vaddv.i 218 } 219 220 define <2 x i64> @test_vaddv_u64_used_by_laneop(<2 x i64> %a1, <2 x i64> %a2) { 221 ; CHECK-LABEL: test_vaddv_u64_used_by_laneop: 222 ; CHECK: addp.2d d[[REGNUM:[0-9]+]], v1 223 ; CHECK-NEXT: ins.d v0[1], v[[REGNUM]][0] 224 ; CHECK-NEXT: ret 225 entry: 226 %0 = tail call i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64> %a2) 227 %1 = insertelement <2 x i64> %a1, i64 %0, i64 1 228 ret <2 x i64> %1 229 } 230 231 define <1 x i64> @test_vaddv_u64_to_vec(<2 x i64> %a1) { 232 ; CHECK-LABEL: test_vaddv_u64_to_vec: 233 ; CHECK: addp.2d d0, v0 234 ; CHECK-NOT: fmov 235 ; CHECK-NOT: ins 236 ; CHECK: ret 237 entry: 238 %vaddv.i = tail call i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64> %a1) 239 %vec = insertelement <1 x i64> undef, i64 %vaddv.i, i32 0 240 ret <1 x i64> %vec 241 } 242 243 define signext i8 @test_vaddvq_s8(<16 x i8> %a1) { 244 ; CHECK-LABEL: test_vaddvq_s8: 245 ; CHECK: addv.16b b[[REGNUM:[0-9]+]], v0 246 ; CHECK-NEXT: smov.b w0, v[[REGNUM]][0] 247 ; CHECK-NEXT: ret 248 entry: 249 %vaddv.i = tail call i32 @llvm.aarch64.neon.saddv.i32.v16i8(<16 x i8> %a1) 250 %0 = trunc i32 %vaddv.i to i8 251 ret i8 %0 252 } 253 254 define <16 x i8> @test_vaddvq_s8_used_by_laneop(<16 x i8> %a1, <16 x i8> %a2) { 255 ; CHECK-LABEL: test_vaddvq_s8_used_by_laneop: 256 ; CHECK: addv.16b b[[REGNUM:[0-9]+]], v1 257 ; CHECK-NEXT: ins.b v0[3], v[[REGNUM]][0] 258 ; CHECK-NEXT: ret 259 entry: 260 %0 = tail call i32 @llvm.aarch64.neon.saddv.i32.v16i8(<16 x i8> %a2) 261 %1 = trunc i32 %0 to i8 262 %2 = insertelement <16 x i8> %a1, i8 %1, i32 3 263 ret <16 x i8> %2 264 } 265 266 define signext i16 @test_vaddvq_s16(<8 x i16> %a1) { 267 ; CHECK-LABEL: test_vaddvq_s16: 268 ; CHECK: addv.8h h[[REGNUM:[0-9]+]], v0 269 ; CHECK-NEXT: smov.h w0, v[[REGNUM]][0] 270 ; CHECK-NEXT: ret 271 entry: 272 %vaddv.i = tail call i32 @llvm.aarch64.neon.saddv.i32.v8i16(<8 x i16> %a1) 273 %0 = trunc i32 %vaddv.i to i16 274 ret i16 %0 275 } 276 277 define <8 x i16> @test_vaddvq_s16_used_by_laneop(<8 x i16> %a1, <8 x i16> %a2) { 278 ; CHECK-LABEL: test_vaddvq_s16_used_by_laneop: 279 ; CHECK: addv.8h h[[REGNUM:[0-9]+]], v1 280 ; CHECK-NEXT: ins.h v0[3], v[[REGNUM]][0] 281 ; CHECK-NEXT: ret 282 entry: 283 %0 = tail call i32 @llvm.aarch64.neon.saddv.i32.v8i16(<8 x i16> %a2) 284 %1 = trunc i32 %0 to i16 285 %2 = insertelement <8 x i16> %a1, i16 %1, i32 3 286 ret <8 x i16> %2 287 } 288 289 define i32 @test_vaddvq_s32(<4 x i32> %a1) { 290 ; CHECK-LABEL: test_vaddvq_s32: 291 ; CHECK: addv.4s [[REGNUM:s[0-9]+]], v0 292 ; CHECK-NEXT: fmov w0, [[REGNUM]] 293 ; CHECK-NEXT: ret 294 entry: 295 %vaddv.i = tail call i32 @llvm.aarch64.neon.saddv.i32.v4i32(<4 x i32> %a1) 296 ret i32 %vaddv.i 297 } 298 299 define <4 x i32> @test_vaddvq_s32_used_by_laneop(<4 x i32> %a1, <4 x i32> %a2) { 300 ; CHECK-LABEL: test_vaddvq_s32_used_by_laneop: 301 ; CHECK: addv.4s s[[REGNUM:[0-9]+]], v1 302 ; CHECK-NEXT: ins.s v0[3], v[[REGNUM]][0] 303 ; CHECK-NEXT: ret 304 entry: 305 %0 = tail call i32 @llvm.aarch64.neon.saddv.i32.v4i32(<4 x i32> %a2) 306 %1 = insertelement <4 x i32> %a1, i32 %0, i32 3 307 ret <4 x i32> %1 308 } 309 310 define zeroext i8 @test_vaddvq_u8(<16 x i8> %a1) { 311 ; CHECK-LABEL: test_vaddvq_u8: 312 ; CHECK: addv.16b b[[REGNUM:[0-9]+]], v0 313 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 314 ; CHECK-NEXT: ret 315 entry: 316 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v16i8(<16 x i8> %a1) 317 %0 = trunc i32 %vaddv.i to i8 318 ret i8 %0 319 } 320 321 define <16 x i8> @test_vaddvq_u8_used_by_laneop(<16 x i8> %a1, <16 x i8> %a2) { 322 ; CHECK-LABEL: test_vaddvq_u8_used_by_laneop: 323 ; CHECK: addv.16b b[[REGNUM:[0-9]+]], v1 324 ; CHECK-NEXT: ins.b v0[3], v[[REGNUM]][0] 325 ; CHECK-NEXT: ret 326 entry: 327 %0 = tail call i32 @llvm.aarch64.neon.uaddv.i32.v16i8(<16 x i8> %a2) 328 %1 = trunc i32 %0 to i8 329 %2 = insertelement <16 x i8> %a1, i8 %1, i32 3 330 ret <16 x i8> %2 331 } 332 333 define zeroext i16 @test_vaddvq_u16(<8 x i16> %a1) { 334 ; CHECK-LABEL: test_vaddvq_u16: 335 ; CHECK: addv.8h h[[REGNUM:[0-9]+]], v0 336 ; CHECK-NEXT: fmov w0, s[[REGNUM]] 337 ; CHECK-NEXT: ret 338 entry: 339 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v8i16(<8 x i16> %a1) 340 %0 = trunc i32 %vaddv.i to i16 341 ret i16 %0 342 } 343 344 define <8 x i16> @test_vaddvq_u16_used_by_laneop(<8 x i16> %a1, <8 x i16> %a2) { 345 ; CHECK-LABEL: test_vaddvq_u16_used_by_laneop: 346 ; CHECK: addv.8h h[[REGNUM:[0-9]+]], v1 347 ; CHECK-NEXT: ins.h v0[3], v[[REGNUM]][0] 348 ; CHECK-NEXT: ret 349 entry: 350 %0 = tail call i32 @llvm.aarch64.neon.uaddv.i32.v8i16(<8 x i16> %a2) 351 %1 = trunc i32 %0 to i16 352 %2 = insertelement <8 x i16> %a1, i16 %1, i32 3 353 ret <8 x i16> %2 354 } 355 356 define i32 @test_vaddvq_u32(<4 x i32> %a1) { 357 ; CHECK-LABEL: test_vaddvq_u32: 358 ; CHECK: addv.4s [[REGNUM:s[0-9]+]], v0 359 ; CHECK-NEXT: fmov [[FMOVRES:w[0-9]+]], [[REGNUM]] 360 ; CHECK-NEXT: ret 361 entry: 362 %vaddv.i = tail call i32 @llvm.aarch64.neon.uaddv.i32.v4i32(<4 x i32> %a1) 363 ret i32 %vaddv.i 364 } 365 366 define <4 x i32> @test_vaddvq_u32_used_by_laneop(<4 x i32> %a1, <4 x i32> %a2) { 367 ; CHECK-LABEL: test_vaddvq_u32_used_by_laneop: 368 ; CHECK: addv.4s s[[REGNUM:[0-9]+]], v1 369 ; CHECK-NEXT: ins.s v0[3], v[[REGNUM]][0] 370 ; CHECK-NEXT: ret 371 entry: 372 %0 = tail call i32 @llvm.aarch64.neon.uaddv.i32.v4i32(<4 x i32> %a2) 373 %1 = insertelement <4 x i32> %a1, i32 %0, i32 3 374 ret <4 x i32> %1 375 } 376 377 declare i32 @llvm.aarch64.neon.uaddv.i32.v4i32(<4 x i32>) 378 379 declare i32 @llvm.aarch64.neon.uaddv.i32.v8i16(<8 x i16>) 380 381 declare i32 @llvm.aarch64.neon.uaddv.i32.v16i8(<16 x i8>) 382 383 declare i32 @llvm.aarch64.neon.saddv.i32.v4i32(<4 x i32>) 384 385 declare i32 @llvm.aarch64.neon.saddv.i32.v8i16(<8 x i16>) 386 387 declare i32 @llvm.aarch64.neon.saddv.i32.v16i8(<16 x i8>) 388 389 declare i64 @llvm.aarch64.neon.uaddv.i64.v2i64(<2 x i64>) 390 391 declare i32 @llvm.aarch64.neon.uaddv.i32.v2i32(<2 x i32>) 392 393 declare i32 @llvm.aarch64.neon.uaddv.i32.v4i16(<4 x i16>) 394 395 declare i32 @llvm.aarch64.neon.uaddv.i32.v8i8(<8 x i8>) 396 397 declare i32 @llvm.aarch64.neon.saddv.i32.v2i32(<2 x i32>) 398 399 declare i64 @llvm.aarch64.neon.saddv.i64.v2i64(<2 x i64>) 400 401 declare i32 @llvm.aarch64.neon.saddv.i32.v4i16(<4 x i16>) 402 403 declare i32 @llvm.aarch64.neon.saddv.i32.v8i8(<8 x i8>) 404 405 declare float @llvm.aarch64.neon.faddv.f32.v2f32(<2 x float> %a1) 406 declare float @llvm.aarch64.neon.faddv.f32.v4f32(<4 x float> %a1) 407 declare double @llvm.aarch64.neon.faddv.f64.v2f64(<2 x double> %a1) 408
