1; RUN: llc < %s -march=arm -mattr=+neon | FileCheck %s 2 3define <8 x i8> @vshls8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 4;CHECK: vshls8: 5;CHECK: vshl.s8 6 %tmp1 = load <8 x i8>* %A 7 %tmp2 = load <8 x i8>* %B 8 %tmp3 = call <8 x i8> @llvm.arm.neon.vshifts.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2) 9 ret <8 x i8> %tmp3 10} 11 12define <4 x i16> @vshls16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 13;CHECK: vshls16: 14;CHECK: vshl.s16 15 %tmp1 = load <4 x i16>* %A 16 %tmp2 = load <4 x i16>* %B 17 %tmp3 = call <4 x i16> @llvm.arm.neon.vshifts.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2) 18 ret <4 x i16> %tmp3 19} 20 21define <2 x i32> @vshls32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 22;CHECK: vshls32: 23;CHECK: vshl.s32 24 %tmp1 = load <2 x i32>* %A 25 %tmp2 = load <2 x i32>* %B 26 %tmp3 = call <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2) 27 ret <2 x i32> %tmp3 28} 29 30define <1 x i64> @vshls64(<1 x i64>* %A, <1 x i64>* %B) nounwind { 31;CHECK: vshls64: 32;CHECK: vshl.s64 33 %tmp1 = load <1 x i64>* %A 34 %tmp2 = load <1 x i64>* %B 35 %tmp3 = call <1 x i64> @llvm.arm.neon.vshifts.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2) 36 ret <1 x i64> %tmp3 37} 38 39define <8 x i8> @vshlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 40;CHECK: vshlu8: 41;CHECK: vshl.u8 42 %tmp1 = load <8 x i8>* %A 43 %tmp2 = load <8 x i8>* %B 44 %tmp3 = call <8 x i8> @llvm.arm.neon.vshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2) 45 ret <8 x i8> %tmp3 46} 47 48define <4 x i16> @vshlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 49;CHECK: vshlu16: 50;CHECK: vshl.u16 51 %tmp1 = load <4 x i16>* %A 52 %tmp2 = load <4 x i16>* %B 53 %tmp3 = call <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2) 54 ret <4 x i16> %tmp3 55} 56 57define <2 x i32> @vshlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 58;CHECK: vshlu32: 59;CHECK: vshl.u32 60 %tmp1 = load <2 x i32>* %A 61 %tmp2 = load <2 x i32>* %B 62 %tmp3 = call <2 x i32> @llvm.arm.neon.vshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2) 63 ret <2 x i32> %tmp3 64} 65 66define <1 x i64> @vshlu64(<1 x i64>* %A, <1 x i64>* %B) nounwind { 67;CHECK: vshlu64: 68;CHECK: vshl.u64 69 %tmp1 = load <1 x i64>* %A 70 %tmp2 = load <1 x i64>* %B 71 %tmp3 = call <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2) 72 ret <1 x i64> %tmp3 73} 74 75define <16 x i8> @vshlQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind { 76;CHECK: vshlQs8: 77;CHECK: vshl.s8 78 %tmp1 = load <16 x i8>* %A 79 %tmp2 = load <16 x i8>* %B 80 %tmp3 = call <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2) 81 ret <16 x i8> %tmp3 82} 83 84define <8 x i16> @vshlQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 85;CHECK: vshlQs16: 86;CHECK: vshl.s16 87 %tmp1 = load <8 x i16>* %A 88 %tmp2 = load <8 x i16>* %B 89 %tmp3 = call <8 x i16> @llvm.arm.neon.vshifts.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2) 90 ret <8 x i16> %tmp3 91} 92 93define <4 x i32> @vshlQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 94;CHECK: vshlQs32: 95;CHECK: vshl.s32 96 %tmp1 = load <4 x i32>* %A 97 %tmp2 = load <4 x i32>* %B 98 %tmp3 = call <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2) 99 ret <4 x i32> %tmp3 100} 101 102define <2 x i64> @vshlQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind { 103;CHECK: vshlQs64: 104;CHECK: vshl.s64 105 %tmp1 = load <2 x i64>* %A 106 %tmp2 = load <2 x i64>* %B 107 %tmp3 = call <2 x i64> @llvm.arm.neon.vshifts.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2) 108 ret <2 x i64> %tmp3 109} 110 111define <16 x i8> @vshlQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind { 112;CHECK: vshlQu8: 113;CHECK: vshl.u8 114 %tmp1 = load <16 x i8>* %A 115 %tmp2 = load <16 x i8>* %B 116 %tmp3 = call <16 x i8> @llvm.arm.neon.vshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2) 117 ret <16 x i8> %tmp3 118} 119 120define <8 x i16> @vshlQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 121;CHECK: vshlQu16: 122;CHECK: vshl.u16 123 %tmp1 = load <8 x i16>* %A 124 %tmp2 = load <8 x i16>* %B 125 %tmp3 = call <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2) 126 ret <8 x i16> %tmp3 127} 128 129define <4 x i32> @vshlQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 130;CHECK: vshlQu32: 131;CHECK: vshl.u32 132 %tmp1 = load <4 x i32>* %A 133 %tmp2 = load <4 x i32>* %B 134 %tmp3 = call <4 x i32> @llvm.arm.neon.vshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2) 135 ret <4 x i32> %tmp3 136} 137 138define <2 x i64> @vshlQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind { 139;CHECK: vshlQu64: 140;CHECK: vshl.u64 141 %tmp1 = load <2 x i64>* %A 142 %tmp2 = load <2 x i64>* %B 143 %tmp3 = call <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2) 144 ret <2 x i64> %tmp3 145} 146 147; For left shifts by immediates, the signedness is irrelevant. 148; Test a mix of both signed and unsigned intrinsics. 149 150define <8 x i8> @vshli8(<8 x i8>* %A) nounwind { 151;CHECK: vshli8: 152;CHECK: vshl.i8 153 %tmp1 = load <8 x i8>* %A 154 %tmp2 = call <8 x i8> @llvm.arm.neon.vshifts.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >) 155 ret <8 x i8> %tmp2 156} 157 158define <4 x i16> @vshli16(<4 x i16>* %A) nounwind { 159;CHECK: vshli16: 160;CHECK: vshl.i16 161 %tmp1 = load <4 x i16>* %A 162 %tmp2 = call <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 15, i16 15, i16 15, i16 15 >) 163 ret <4 x i16> %tmp2 164} 165 166define <2 x i32> @vshli32(<2 x i32>* %A) nounwind { 167;CHECK: vshli32: 168;CHECK: vshl.i32 169 %tmp1 = load <2 x i32>* %A 170 %tmp2 = call <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 31, i32 31 >) 171 ret <2 x i32> %tmp2 172} 173 174define <1 x i64> @vshli64(<1 x i64>* %A) nounwind { 175;CHECK: vshli64: 176;CHECK: vshl.i64 177 %tmp1 = load <1 x i64>* %A 178 %tmp2 = call <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 63 >) 179 ret <1 x i64> %tmp2 180} 181 182define <16 x i8> @vshlQi8(<16 x i8>* %A) nounwind { 183;CHECK: vshlQi8: 184;CHECK: vshl.i8 185 %tmp1 = load <16 x i8>* %A 186 %tmp2 = call <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7, i8 7 >) 187 ret <16 x i8> %tmp2 188} 189 190define <8 x i16> @vshlQi16(<8 x i16>* %A) nounwind { 191;CHECK: vshlQi16: 192;CHECK: vshl.i16 193 %tmp1 = load <8 x i16>* %A 194 %tmp2 = call <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15, i16 15 >) 195 ret <8 x i16> %tmp2 196} 197 198define <4 x i32> @vshlQi32(<4 x i32>* %A) nounwind { 199;CHECK: vshlQi32: 200;CHECK: vshl.i32 201 %tmp1 = load <4 x i32>* %A 202 %tmp2 = call <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 31, i32 31, i32 31, i32 31 >) 203 ret <4 x i32> %tmp2 204} 205 206define <2 x i64> @vshlQi64(<2 x i64>* %A) nounwind { 207;CHECK: vshlQi64: 208;CHECK: vshl.i64 209 %tmp1 = load <2 x i64>* %A 210 %tmp2 = call <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 63, i64 63 >) 211 ret <2 x i64> %tmp2 212} 213 214; Right shift by immediate: 215 216define <8 x i8> @vshrs8(<8 x i8>* %A) nounwind { 217;CHECK: vshrs8: 218;CHECK: vshr.s8 219 %tmp1 = load <8 x i8>* %A 220 %tmp2 = call <8 x i8> @llvm.arm.neon.vshifts.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 221 ret <8 x i8> %tmp2 222} 223 224define <4 x i16> @vshrs16(<4 x i16>* %A) nounwind { 225;CHECK: vshrs16: 226;CHECK: vshr.s16 227 %tmp1 = load <4 x i16>* %A 228 %tmp2 = call <4 x i16> @llvm.arm.neon.vshifts.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >) 229 ret <4 x i16> %tmp2 230} 231 232define <2 x i32> @vshrs32(<2 x i32>* %A) nounwind { 233;CHECK: vshrs32: 234;CHECK: vshr.s32 235 %tmp1 = load <2 x i32>* %A 236 %tmp2 = call <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >) 237 ret <2 x i32> %tmp2 238} 239 240define <1 x i64> @vshrs64(<1 x i64>* %A) nounwind { 241;CHECK: vshrs64: 242;CHECK: vshr.s64 243 %tmp1 = load <1 x i64>* %A 244 %tmp2 = call <1 x i64> @llvm.arm.neon.vshifts.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >) 245 ret <1 x i64> %tmp2 246} 247 248define <8 x i8> @vshru8(<8 x i8>* %A) nounwind { 249;CHECK: vshru8: 250;CHECK: vshr.u8 251 %tmp1 = load <8 x i8>* %A 252 %tmp2 = call <8 x i8> @llvm.arm.neon.vshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 253 ret <8 x i8> %tmp2 254} 255 256define <4 x i16> @vshru16(<4 x i16>* %A) nounwind { 257;CHECK: vshru16: 258;CHECK: vshr.u16 259 %tmp1 = load <4 x i16>* %A 260 %tmp2 = call <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >) 261 ret <4 x i16> %tmp2 262} 263 264define <2 x i32> @vshru32(<2 x i32>* %A) nounwind { 265;CHECK: vshru32: 266;CHECK: vshr.u32 267 %tmp1 = load <2 x i32>* %A 268 %tmp2 = call <2 x i32> @llvm.arm.neon.vshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >) 269 ret <2 x i32> %tmp2 270} 271 272define <1 x i64> @vshru64(<1 x i64>* %A) nounwind { 273;CHECK: vshru64: 274;CHECK: vshr.u64 275 %tmp1 = load <1 x i64>* %A 276 %tmp2 = call <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >) 277 ret <1 x i64> %tmp2 278} 279 280define <16 x i8> @vshrQs8(<16 x i8>* %A) nounwind { 281;CHECK: vshrQs8: 282;CHECK: vshr.s8 283 %tmp1 = load <16 x i8>* %A 284 %tmp2 = call <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 285 ret <16 x i8> %tmp2 286} 287 288define <8 x i16> @vshrQs16(<8 x i16>* %A) nounwind { 289;CHECK: vshrQs16: 290;CHECK: vshr.s16 291 %tmp1 = load <8 x i16>* %A 292 %tmp2 = call <8 x i16> @llvm.arm.neon.vshifts.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >) 293 ret <8 x i16> %tmp2 294} 295 296define <4 x i32> @vshrQs32(<4 x i32>* %A) nounwind { 297;CHECK: vshrQs32: 298;CHECK: vshr.s32 299 %tmp1 = load <4 x i32>* %A 300 %tmp2 = call <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >) 301 ret <4 x i32> %tmp2 302} 303 304define <2 x i64> @vshrQs64(<2 x i64>* %A) nounwind { 305;CHECK: vshrQs64: 306;CHECK: vshr.s64 307 %tmp1 = load <2 x i64>* %A 308 %tmp2 = call <2 x i64> @llvm.arm.neon.vshifts.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >) 309 ret <2 x i64> %tmp2 310} 311 312define <16 x i8> @vshrQu8(<16 x i8>* %A) nounwind { 313;CHECK: vshrQu8: 314;CHECK: vshr.u8 315 %tmp1 = load <16 x i8>* %A 316 %tmp2 = call <16 x i8> @llvm.arm.neon.vshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 317 ret <16 x i8> %tmp2 318} 319 320define <8 x i16> @vshrQu16(<8 x i16>* %A) nounwind { 321;CHECK: vshrQu16: 322;CHECK: vshr.u16 323 %tmp1 = load <8 x i16>* %A 324 %tmp2 = call <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >) 325 ret <8 x i16> %tmp2 326} 327 328define <4 x i32> @vshrQu32(<4 x i32>* %A) nounwind { 329;CHECK: vshrQu32: 330;CHECK: vshr.u32 331 %tmp1 = load <4 x i32>* %A 332 %tmp2 = call <4 x i32> @llvm.arm.neon.vshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >) 333 ret <4 x i32> %tmp2 334} 335 336define <2 x i64> @vshrQu64(<2 x i64>* %A) nounwind { 337;CHECK: vshrQu64: 338;CHECK: vshr.u64 339 %tmp1 = load <2 x i64>* %A 340 %tmp2 = call <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >) 341 ret <2 x i64> %tmp2 342} 343 344declare <8 x i8> @llvm.arm.neon.vshifts.v8i8(<8 x i8>, <8 x i8>) nounwind readnone 345declare <4 x i16> @llvm.arm.neon.vshifts.v4i16(<4 x i16>, <4 x i16>) nounwind readnone 346declare <2 x i32> @llvm.arm.neon.vshifts.v2i32(<2 x i32>, <2 x i32>) nounwind readnone 347declare <1 x i64> @llvm.arm.neon.vshifts.v1i64(<1 x i64>, <1 x i64>) nounwind readnone 348 349declare <8 x i8> @llvm.arm.neon.vshiftu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone 350declare <4 x i16> @llvm.arm.neon.vshiftu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone 351declare <2 x i32> @llvm.arm.neon.vshiftu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone 352declare <1 x i64> @llvm.arm.neon.vshiftu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone 353 354declare <16 x i8> @llvm.arm.neon.vshifts.v16i8(<16 x i8>, <16 x i8>) nounwind readnone 355declare <8 x i16> @llvm.arm.neon.vshifts.v8i16(<8 x i16>, <8 x i16>) nounwind readnone 356declare <4 x i32> @llvm.arm.neon.vshifts.v4i32(<4 x i32>, <4 x i32>) nounwind readnone 357declare <2 x i64> @llvm.arm.neon.vshifts.v2i64(<2 x i64>, <2 x i64>) nounwind readnone 358 359declare <16 x i8> @llvm.arm.neon.vshiftu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone 360declare <8 x i16> @llvm.arm.neon.vshiftu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone 361declare <4 x i32> @llvm.arm.neon.vshiftu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone 362declare <2 x i64> @llvm.arm.neon.vshiftu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone 363 364define <8 x i8> @vrshls8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 365;CHECK: vrshls8: 366;CHECK: vrshl.s8 367 %tmp1 = load <8 x i8>* %A 368 %tmp2 = load <8 x i8>* %B 369 %tmp3 = call <8 x i8> @llvm.arm.neon.vrshifts.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2) 370 ret <8 x i8> %tmp3 371} 372 373define <4 x i16> @vrshls16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 374;CHECK: vrshls16: 375;CHECK: vrshl.s16 376 %tmp1 = load <4 x i16>* %A 377 %tmp2 = load <4 x i16>* %B 378 %tmp3 = call <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2) 379 ret <4 x i16> %tmp3 380} 381 382define <2 x i32> @vrshls32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 383;CHECK: vrshls32: 384;CHECK: vrshl.s32 385 %tmp1 = load <2 x i32>* %A 386 %tmp2 = load <2 x i32>* %B 387 %tmp3 = call <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2) 388 ret <2 x i32> %tmp3 389} 390 391define <1 x i64> @vrshls64(<1 x i64>* %A, <1 x i64>* %B) nounwind { 392;CHECK: vrshls64: 393;CHECK: vrshl.s64 394 %tmp1 = load <1 x i64>* %A 395 %tmp2 = load <1 x i64>* %B 396 %tmp3 = call <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2) 397 ret <1 x i64> %tmp3 398} 399 400define <8 x i8> @vrshlu8(<8 x i8>* %A, <8 x i8>* %B) nounwind { 401;CHECK: vrshlu8: 402;CHECK: vrshl.u8 403 %tmp1 = load <8 x i8>* %A 404 %tmp2 = load <8 x i8>* %B 405 %tmp3 = call <8 x i8> @llvm.arm.neon.vrshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> %tmp2) 406 ret <8 x i8> %tmp3 407} 408 409define <4 x i16> @vrshlu16(<4 x i16>* %A, <4 x i16>* %B) nounwind { 410;CHECK: vrshlu16: 411;CHECK: vrshl.u16 412 %tmp1 = load <4 x i16>* %A 413 %tmp2 = load <4 x i16>* %B 414 %tmp3 = call <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> %tmp2) 415 ret <4 x i16> %tmp3 416} 417 418define <2 x i32> @vrshlu32(<2 x i32>* %A, <2 x i32>* %B) nounwind { 419;CHECK: vrshlu32: 420;CHECK: vrshl.u32 421 %tmp1 = load <2 x i32>* %A 422 %tmp2 = load <2 x i32>* %B 423 %tmp3 = call <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> %tmp2) 424 ret <2 x i32> %tmp3 425} 426 427define <1 x i64> @vrshlu64(<1 x i64>* %A, <1 x i64>* %B) nounwind { 428;CHECK: vrshlu64: 429;CHECK: vrshl.u64 430 %tmp1 = load <1 x i64>* %A 431 %tmp2 = load <1 x i64>* %B 432 %tmp3 = call <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> %tmp2) 433 ret <1 x i64> %tmp3 434} 435 436define <16 x i8> @vrshlQs8(<16 x i8>* %A, <16 x i8>* %B) nounwind { 437;CHECK: vrshlQs8: 438;CHECK: vrshl.s8 439 %tmp1 = load <16 x i8>* %A 440 %tmp2 = load <16 x i8>* %B 441 %tmp3 = call <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2) 442 ret <16 x i8> %tmp3 443} 444 445define <8 x i16> @vrshlQs16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 446;CHECK: vrshlQs16: 447;CHECK: vrshl.s16 448 %tmp1 = load <8 x i16>* %A 449 %tmp2 = load <8 x i16>* %B 450 %tmp3 = call <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2) 451 ret <8 x i16> %tmp3 452} 453 454define <4 x i32> @vrshlQs32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 455;CHECK: vrshlQs32: 456;CHECK: vrshl.s32 457 %tmp1 = load <4 x i32>* %A 458 %tmp2 = load <4 x i32>* %B 459 %tmp3 = call <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2) 460 ret <4 x i32> %tmp3 461} 462 463define <2 x i64> @vrshlQs64(<2 x i64>* %A, <2 x i64>* %B) nounwind { 464;CHECK: vrshlQs64: 465;CHECK: vrshl.s64 466 %tmp1 = load <2 x i64>* %A 467 %tmp2 = load <2 x i64>* %B 468 %tmp3 = call <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2) 469 ret <2 x i64> %tmp3 470} 471 472define <16 x i8> @vrshlQu8(<16 x i8>* %A, <16 x i8>* %B) nounwind { 473;CHECK: vrshlQu8: 474;CHECK: vrshl.u8 475 %tmp1 = load <16 x i8>* %A 476 %tmp2 = load <16 x i8>* %B 477 %tmp3 = call <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> %tmp2) 478 ret <16 x i8> %tmp3 479} 480 481define <8 x i16> @vrshlQu16(<8 x i16>* %A, <8 x i16>* %B) nounwind { 482;CHECK: vrshlQu16: 483;CHECK: vrshl.u16 484 %tmp1 = load <8 x i16>* %A 485 %tmp2 = load <8 x i16>* %B 486 %tmp3 = call <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> %tmp2) 487 ret <8 x i16> %tmp3 488} 489 490define <4 x i32> @vrshlQu32(<4 x i32>* %A, <4 x i32>* %B) nounwind { 491;CHECK: vrshlQu32: 492;CHECK: vrshl.u32 493 %tmp1 = load <4 x i32>* %A 494 %tmp2 = load <4 x i32>* %B 495 %tmp3 = call <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> %tmp2) 496 ret <4 x i32> %tmp3 497} 498 499define <2 x i64> @vrshlQu64(<2 x i64>* %A, <2 x i64>* %B) nounwind { 500;CHECK: vrshlQu64: 501;CHECK: vrshl.u64 502 %tmp1 = load <2 x i64>* %A 503 %tmp2 = load <2 x i64>* %B 504 %tmp3 = call <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> %tmp2) 505 ret <2 x i64> %tmp3 506} 507 508define <8 x i8> @vrshrs8(<8 x i8>* %A) nounwind { 509;CHECK: vrshrs8: 510;CHECK: vrshr.s8 511 %tmp1 = load <8 x i8>* %A 512 %tmp2 = call <8 x i8> @llvm.arm.neon.vrshifts.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 513 ret <8 x i8> %tmp2 514} 515 516define <4 x i16> @vrshrs16(<4 x i16>* %A) nounwind { 517;CHECK: vrshrs16: 518;CHECK: vrshr.s16 519 %tmp1 = load <4 x i16>* %A 520 %tmp2 = call <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >) 521 ret <4 x i16> %tmp2 522} 523 524define <2 x i32> @vrshrs32(<2 x i32>* %A) nounwind { 525;CHECK: vrshrs32: 526;CHECK: vrshr.s32 527 %tmp1 = load <2 x i32>* %A 528 %tmp2 = call <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >) 529 ret <2 x i32> %tmp2 530} 531 532define <1 x i64> @vrshrs64(<1 x i64>* %A) nounwind { 533;CHECK: vrshrs64: 534;CHECK: vrshr.s64 535 %tmp1 = load <1 x i64>* %A 536 %tmp2 = call <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >) 537 ret <1 x i64> %tmp2 538} 539 540define <8 x i8> @vrshru8(<8 x i8>* %A) nounwind { 541;CHECK: vrshru8: 542;CHECK: vrshr.u8 543 %tmp1 = load <8 x i8>* %A 544 %tmp2 = call <8 x i8> @llvm.arm.neon.vrshiftu.v8i8(<8 x i8> %tmp1, <8 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 545 ret <8 x i8> %tmp2 546} 547 548define <4 x i16> @vrshru16(<4 x i16>* %A) nounwind { 549;CHECK: vrshru16: 550;CHECK: vrshr.u16 551 %tmp1 = load <4 x i16>* %A 552 %tmp2 = call <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16> %tmp1, <4 x i16> < i16 -16, i16 -16, i16 -16, i16 -16 >) 553 ret <4 x i16> %tmp2 554} 555 556define <2 x i32> @vrshru32(<2 x i32>* %A) nounwind { 557;CHECK: vrshru32: 558;CHECK: vrshr.u32 559 %tmp1 = load <2 x i32>* %A 560 %tmp2 = call <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32> %tmp1, <2 x i32> < i32 -32, i32 -32 >) 561 ret <2 x i32> %tmp2 562} 563 564define <1 x i64> @vrshru64(<1 x i64>* %A) nounwind { 565;CHECK: vrshru64: 566;CHECK: vrshr.u64 567 %tmp1 = load <1 x i64>* %A 568 %tmp2 = call <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64> %tmp1, <1 x i64> < i64 -64 >) 569 ret <1 x i64> %tmp2 570} 571 572define <16 x i8> @vrshrQs8(<16 x i8>* %A) nounwind { 573;CHECK: vrshrQs8: 574;CHECK: vrshr.s8 575 %tmp1 = load <16 x i8>* %A 576 %tmp2 = call <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 577 ret <16 x i8> %tmp2 578} 579 580define <8 x i16> @vrshrQs16(<8 x i16>* %A) nounwind { 581;CHECK: vrshrQs16: 582;CHECK: vrshr.s16 583 %tmp1 = load <8 x i16>* %A 584 %tmp2 = call <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >) 585 ret <8 x i16> %tmp2 586} 587 588define <4 x i32> @vrshrQs32(<4 x i32>* %A) nounwind { 589;CHECK: vrshrQs32: 590;CHECK: vrshr.s32 591 %tmp1 = load <4 x i32>* %A 592 %tmp2 = call <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >) 593 ret <4 x i32> %tmp2 594} 595 596define <2 x i64> @vrshrQs64(<2 x i64>* %A) nounwind { 597;CHECK: vrshrQs64: 598;CHECK: vrshr.s64 599 %tmp1 = load <2 x i64>* %A 600 %tmp2 = call <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >) 601 ret <2 x i64> %tmp2 602} 603 604define <16 x i8> @vrshrQu8(<16 x i8>* %A) nounwind { 605;CHECK: vrshrQu8: 606;CHECK: vrshr.u8 607 %tmp1 = load <16 x i8>* %A 608 %tmp2 = call <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8> %tmp1, <16 x i8> < i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8, i8 -8 >) 609 ret <16 x i8> %tmp2 610} 611 612define <8 x i16> @vrshrQu16(<8 x i16>* %A) nounwind { 613;CHECK: vrshrQu16: 614;CHECK: vrshr.u16 615 %tmp1 = load <8 x i16>* %A 616 %tmp2 = call <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16> %tmp1, <8 x i16> < i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16, i16 -16 >) 617 ret <8 x i16> %tmp2 618} 619 620define <4 x i32> @vrshrQu32(<4 x i32>* %A) nounwind { 621;CHECK: vrshrQu32: 622;CHECK: vrshr.u32 623 %tmp1 = load <4 x i32>* %A 624 %tmp2 = call <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32> %tmp1, <4 x i32> < i32 -32, i32 -32, i32 -32, i32 -32 >) 625 ret <4 x i32> %tmp2 626} 627 628define <2 x i64> @vrshrQu64(<2 x i64>* %A) nounwind { 629;CHECK: vrshrQu64: 630;CHECK: vrshr.u64 631 %tmp1 = load <2 x i64>* %A 632 %tmp2 = call <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64> %tmp1, <2 x i64> < i64 -64, i64 -64 >) 633 ret <2 x i64> %tmp2 634} 635 636declare <8 x i8> @llvm.arm.neon.vrshifts.v8i8(<8 x i8>, <8 x i8>) nounwind readnone 637declare <4 x i16> @llvm.arm.neon.vrshifts.v4i16(<4 x i16>, <4 x i16>) nounwind readnone 638declare <2 x i32> @llvm.arm.neon.vrshifts.v2i32(<2 x i32>, <2 x i32>) nounwind readnone 639declare <1 x i64> @llvm.arm.neon.vrshifts.v1i64(<1 x i64>, <1 x i64>) nounwind readnone 640 641declare <8 x i8> @llvm.arm.neon.vrshiftu.v8i8(<8 x i8>, <8 x i8>) nounwind readnone 642declare <4 x i16> @llvm.arm.neon.vrshiftu.v4i16(<4 x i16>, <4 x i16>) nounwind readnone 643declare <2 x i32> @llvm.arm.neon.vrshiftu.v2i32(<2 x i32>, <2 x i32>) nounwind readnone 644declare <1 x i64> @llvm.arm.neon.vrshiftu.v1i64(<1 x i64>, <1 x i64>) nounwind readnone 645 646declare <16 x i8> @llvm.arm.neon.vrshifts.v16i8(<16 x i8>, <16 x i8>) nounwind readnone 647declare <8 x i16> @llvm.arm.neon.vrshifts.v8i16(<8 x i16>, <8 x i16>) nounwind readnone 648declare <4 x i32> @llvm.arm.neon.vrshifts.v4i32(<4 x i32>, <4 x i32>) nounwind readnone 649declare <2 x i64> @llvm.arm.neon.vrshifts.v2i64(<2 x i64>, <2 x i64>) nounwind readnone 650 651declare <16 x i8> @llvm.arm.neon.vrshiftu.v16i8(<16 x i8>, <16 x i8>) nounwind readnone 652declare <8 x i16> @llvm.arm.neon.vrshiftu.v8i16(<8 x i16>, <8 x i16>) nounwind readnone 653declare <4 x i32> @llvm.arm.neon.vrshiftu.v4i32(<4 x i32>, <4 x i32>) nounwind readnone 654declare <2 x i64> @llvm.arm.neon.vrshiftu.v2i64(<2 x i64>, <2 x i64>) nounwind readnone 655