include/llvm/IR/IntrinsicsAArch64.td

   1 //===- IntrinsicsAArch64.td - Defines AArch64 intrinsics -----------*- tablegen -*-===//
   2 //
   3 //                     The LLVM Compiler Infrastructure
   4 //
   5 // This file is distributed under the University of Illinois Open Source
   6 // License. See LICENSE.TXT for details.
   7 //
   8 //===----------------------------------------------------------------------===//
   9 //
  10 // This file defines all of the AArch64-specific intrinsics.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14 //===----------------------------------------------------------------------===//
  15 // Advanced SIMD (NEON)
  16
  17 let TargetPrefix = "aarch64" in {  // All intrinsics start with "llvm.aarch64.".
  18
  19 // Vector Absolute Compare (Floating Point)
  20 def int_aarch64_neon_vacgeq :
  21   Intrinsic<[llvm_v2i64_ty], [llvm_v2f64_ty, llvm_v2f64_ty], [IntrNoMem]>;
  22 def int_aarch64_neon_vacgtq :
  23   Intrinsic<[llvm_v2i64_ty], [llvm_v2f64_ty, llvm_v2f64_ty], [IntrNoMem]>;
  24
  25 // Vector saturating accumulate
  26 def int_aarch64_neon_suqadd : Neon_2Arg_Intrinsic;
  27 def int_aarch64_neon_usqadd : Neon_2Arg_Intrinsic;
  28
  29 // Vector Bitwise reverse
  30 def int_aarch64_neon_rbit : Neon_1Arg_Intrinsic;
  31
  32 // Vector extract and narrow
  33 def int_aarch64_neon_xtn :
  34   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  35
  36 // Vector floating-point convert
  37 def int_aarch64_neon_frintn : Neon_1Arg_Intrinsic;
  38 def int_aarch64_neon_fsqrt : Neon_1Arg_Intrinsic;
  39 def int_aarch64_neon_vcvtxn :
  40   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  41 def int_aarch64_neon_vcvtzs :
  42   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  43 def int_aarch64_neon_vcvtzu :
  44   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  45
  46 // Vector maxNum (Floating Point)
  47 def int_aarch64_neon_vmaxnm : Neon_2Arg_Intrinsic;
  48
  49 // Vector minNum (Floating Point)
  50 def int_aarch64_neon_vminnm : Neon_2Arg_Intrinsic;
  51
  52 // Vector Pairwise maxNum (Floating Point)
  53 def int_aarch64_neon_vpmaxnm : Neon_2Arg_Intrinsic;
  54
  55 // Vector Pairwise minNum (Floating Point)
  56 def int_aarch64_neon_vpminnm : Neon_2Arg_Intrinsic;
  57
  58 // Vector Multiply Extended and Scalar Multiply Extended (Floating Point)
  59 def int_aarch64_neon_vmulx  :
  60   Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>]>;
  61
  62 class Neon_N2V_Intrinsic
  63   : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>, llvm_i32_ty],
  64               [IntrNoMem]>;
  65 class Neon_N3V_Intrinsic
  66   : Intrinsic<[llvm_anyvector_ty],
  67               [LLVMMatchType<0>, LLVMMatchType<0>, llvm_i32_ty],
  68               [IntrNoMem]>;
  69 class Neon_N2V_Narrow_Intrinsic
  70   : Intrinsic<[llvm_anyvector_ty],
  71               [LLVMExtendedElementVectorType<0>, llvm_i32_ty],
  72               [IntrNoMem]>;
  73
  74 // Vector rounding shift right by immediate (Signed)
  75 def int_aarch64_neon_vsrshr : Neon_N2V_Intrinsic;
  76 def int_aarch64_neon_vurshr : Neon_N2V_Intrinsic;
  77 def int_aarch64_neon_vsqshlu : Neon_N2V_Intrinsic;
  78
  79 def int_aarch64_neon_vsri : Neon_N3V_Intrinsic;
  80 def int_aarch64_neon_vsli : Neon_N3V_Intrinsic;
  81
  82 def int_aarch64_neon_vsqshrun : Neon_N2V_Narrow_Intrinsic;
  83 def int_aarch64_neon_vrshrn : Neon_N2V_Narrow_Intrinsic;
  84 def int_aarch64_neon_vsqrshrun : Neon_N2V_Narrow_Intrinsic;
  85 def int_aarch64_neon_vsqshrn : Neon_N2V_Narrow_Intrinsic;
  86 def int_aarch64_neon_vuqshrn : Neon_N2V_Narrow_Intrinsic;
  87 def int_aarch64_neon_vsqrshrn : Neon_N2V_Narrow_Intrinsic;
  88 def int_aarch64_neon_vuqrshrn : Neon_N2V_Narrow_Intrinsic;
  89
  90 // Vector across
  91 class Neon_Across_Intrinsic
  92   : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  93
  94 def int_aarch64_neon_saddlv : Neon_Across_Intrinsic;
  95 def int_aarch64_neon_uaddlv : Neon_Across_Intrinsic;
  96 def int_aarch64_neon_smaxv  : Neon_Across_Intrinsic;
  97 def int_aarch64_neon_umaxv  : Neon_Across_Intrinsic;
  98 def int_aarch64_neon_sminv  : Neon_Across_Intrinsic;
  99 def int_aarch64_neon_uminv  : Neon_Across_Intrinsic;
 100 def int_aarch64_neon_vaddv  : Neon_Across_Intrinsic;
 101 def int_aarch64_neon_vmaxv  : Neon_Across_Intrinsic;
 102 def int_aarch64_neon_vminv  : Neon_Across_Intrinsic;
 103 def int_aarch64_neon_vmaxnmv : Neon_Across_Intrinsic;
 104 def int_aarch64_neon_vminnmv : Neon_Across_Intrinsic;
 105
 106 // Vector Table Lookup.
 107 def int_aarch64_neon_vtbl1 :
 108   Intrinsic<[llvm_anyvector_ty],
 109             [llvm_anyvector_ty, LLVMMatchType<0>], [IntrNoMem]>;
 110
 111 def int_aarch64_neon_vtbl2 :
 112   Intrinsic<[llvm_anyvector_ty],
 113             [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<0>],
 114             [IntrNoMem]>;
 115
 116 def int_aarch64_neon_vtbl3 :
 117   Intrinsic<[llvm_anyvector_ty],
 118             [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<1>,
 119             LLVMMatchType<0>], [IntrNoMem]>;
 120
 121 def int_aarch64_neon_vtbl4 :
 122   Intrinsic<[llvm_anyvector_ty],
 123             [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<1>,
 124             LLVMMatchType<1>, LLVMMatchType<0>], [IntrNoMem]>;
 125
 126 // Vector Table Extension.
 127 // Some elements of the destination vector may not be updated, so the original
 128 // value of that vector is passed as the first argument.  The next 1-4
 129 // arguments after that are the table.
 130 def int_aarch64_neon_vtbx1 :
 131   Intrinsic<[llvm_anyvector_ty],
 132             [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<0>],
 133             [IntrNoMem]>;
 134
 135 def int_aarch64_neon_vtbx2 :
 136   Intrinsic<[llvm_anyvector_ty],
 137             [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>,
 138             LLVMMatchType<0>], [IntrNoMem]>;
 139
 140 def int_aarch64_neon_vtbx3 :
 141   Intrinsic<[llvm_anyvector_ty],
 142             [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>,
 143             LLVMMatchType<1>, LLVMMatchType<0>], [IntrNoMem]>;
 144
 145 def int_aarch64_neon_vtbx4 :
 146   Intrinsic<[llvm_anyvector_ty],
 147             [LLVMMatchType<0>, llvm_anyvector_ty,  LLVMMatchType<1>,
 148             LLVMMatchType<1>,  LLVMMatchType<1>, LLVMMatchType<0>],
 149             [IntrNoMem]>;
 150
 151 // Vector Load/store
 152 def int_aarch64_neon_vld1x2 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>],
 153                                         [llvm_ptr_ty, llvm_i32_ty],
 154                                         [IntrReadArgMem]>;
 155 def int_aarch64_neon_vld1x3 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 156                                          LLVMMatchType<0>],
 157                                         [llvm_ptr_ty, llvm_i32_ty],
 158                                         [IntrReadArgMem]>;
 159 def int_aarch64_neon_vld1x4 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 160                                          LLVMMatchType<0>, LLVMMatchType<0>],
 161                                         [llvm_ptr_ty, llvm_i32_ty],
 162                                         [IntrReadArgMem]>;
 163
 164 def int_aarch64_neon_vst1x2 : Intrinsic<[],
 165                                         [llvm_ptr_ty, llvm_anyvector_ty,
 166                                          LLVMMatchType<0>, llvm_i32_ty],
 167                                         [IntrReadWriteArgMem]>;
 168 def int_aarch64_neon_vst1x3 : Intrinsic<[],
 169                                         [llvm_ptr_ty, llvm_anyvector_ty,
 170                                          LLVMMatchType<0>, LLVMMatchType<0>,
 171                                          llvm_i32_ty], [IntrReadWriteArgMem]>;
 172 def int_aarch64_neon_vst1x4 : Intrinsic<[],
 173                                         [llvm_ptr_ty, llvm_anyvector_ty,
 174                                          LLVMMatchType<0>, LLVMMatchType<0>,
 175                                          LLVMMatchType<0>, llvm_i32_ty],
 176                                         [IntrReadWriteArgMem]>;
 177
 178 // Scalar Add
 179 def int_aarch64_neon_vaddds :
 180   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 181 def int_aarch64_neon_vadddu :
 182   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 183
 184
 185 // Scalar Sub
 186 def int_aarch64_neon_vsubds :
 187   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 188 def int_aarch64_neon_vsubdu :
 189   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 190
 191
 192 // Scalar Shift
 193 // Scalar Shift Left
 194 def int_aarch64_neon_vshlds :
 195   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 196 def int_aarch64_neon_vshldu :
 197   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 198
 199 // Scalar Saturating Shift Left
 200 def int_aarch64_neon_vqshls : Neon_2Arg_Intrinsic;
 201 def int_aarch64_neon_vqshlu : Neon_2Arg_Intrinsic;
 202
 203 // Scalar Shift Rouding Left
 204 def int_aarch64_neon_vrshlds :
 205   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 206 def int_aarch64_neon_vrshldu :
 207   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 208
 209 // Scalar Saturating Rounding Shift Left
 210 def int_aarch64_neon_vqrshls : Neon_2Arg_Intrinsic;
 211 def int_aarch64_neon_vqrshlu : Neon_2Arg_Intrinsic;
 212
 213 // Scalar Reduce Pairwise Add.
 214 def int_aarch64_neon_vpadd :
 215   Intrinsic<[llvm_v1i64_ty], [llvm_v2i64_ty],[IntrNoMem]>;
 216 def int_aarch64_neon_vpfadd :
 217   Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>;
 218
 219 // Scalar Reduce Pairwise Floating Point Max/Min.
 220 def int_aarch64_neon_vpmax :
 221   Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>;
 222 def int_aarch64_neon_vpmin :
 223   Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>;
 224
 225 // Scalar Reduce Pairwise Floating Point Maxnm/Minnm.
 226 def int_aarch64_neon_vpfmaxnm :
 227   Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>;
 228 def int_aarch64_neon_vpfminnm :
 229   Intrinsic<[llvm_anyfloat_ty], [llvm_anyvector_ty], [IntrNoMem]>;
 230
 231 // Scalar Signed Integer Convert To Floating-point
 232 def int_aarch64_neon_vcvtf32_s32 :
 233   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty], [IntrNoMem]>;
 234 def int_aarch64_neon_vcvtf64_s64 :
 235   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 236
 237 // Scalar Unsigned Integer Convert To Floating-point
 238 def int_aarch64_neon_vcvtf32_u32 :
 239   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty], [IntrNoMem]>;
 240 def int_aarch64_neon_vcvtf64_u64 :
 241   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 242
 243 // Scalar Floating-point Convert
 244 def int_aarch64_neon_fcvtxn :
 245   Intrinsic<[llvm_float_ty], [llvm_double_ty], [IntrNoMem]>;
 246 def int_aarch64_neon_fcvtns :
 247   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 248 def int_aarch64_neon_fcvtnu :
 249   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 250 def int_aarch64_neon_fcvtps :
 251   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 252 def int_aarch64_neon_fcvtpu :
 253   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 254 def int_aarch64_neon_fcvtms :
 255   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 256 def int_aarch64_neon_fcvtmu :
 257   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 258 def int_aarch64_neon_fcvtas :
 259   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 260 def int_aarch64_neon_fcvtau :
 261   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 262 def int_aarch64_neon_fcvtzs :
 263   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 264 def int_aarch64_neon_fcvtzu :
 265   Intrinsic<[llvm_anyvector_ty], [llvm_anyfloat_ty], [IntrNoMem]>;
 266
 267 // Scalar Floating-point Reciprocal Exponent
 268 def int_aarch64_neon_vrecpx : Neon_1Arg_Intrinsic;
 269
 270 class Neon_Cmp_Intrinsic
 271   : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty, llvm_anyvector_ty],
 272               [IntrNoMem]>;
 273
 274 // Scalar Compare Equal
 275 def int_aarch64_neon_vceq : Neon_Cmp_Intrinsic;
 276
 277 // Scalar Compare Greater-Than or Equal
 278 def int_aarch64_neon_vcge : Neon_Cmp_Intrinsic;
 279 def int_aarch64_neon_vchs : Neon_Cmp_Intrinsic;
 280
 281 // Scalar Compare Less-Than or Equal
 282 def int_aarch64_neon_vclez : Neon_Cmp_Intrinsic;
 283
 284 // Scalar Compare Less-Than
 285 def int_aarch64_neon_vcltz : Neon_Cmp_Intrinsic;
 286
 287 // Scalar Compare Greater-Than
 288 def int_aarch64_neon_vcgt : Neon_Cmp_Intrinsic;
 289 def int_aarch64_neon_vchi : Neon_Cmp_Intrinsic;
 290
 291 // Scalar Compare Bitwise Test Bits
 292 def int_aarch64_neon_vtstd : Neon_Cmp_Intrinsic;
 293
 294 // Scalar Floating-point Absolute Compare Greater Than Or Equal
 295 def int_aarch64_neon_vcage : Neon_Cmp_Intrinsic;
 296
 297 // Scalar Floating-point Absolute Compare Greater Than
 298 def int_aarch64_neon_vcagt : Neon_Cmp_Intrinsic;
 299
 300 // Scalar Signed Saturating Accumulated of Unsigned Value
 301 def int_aarch64_neon_vuqadd : Neon_2Arg_Intrinsic;
 302
 303 // Scalar Unsigned Saturating Accumulated of Signed Value
 304 def int_aarch64_neon_vsqadd : Neon_2Arg_Intrinsic;
 305
 306 // Scalar Absolute Value
 307 def int_aarch64_neon_vabs :
 308   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 309
 310 // Scalar Absolute Difference
 311 def int_aarch64_neon_vabd : Neon_2Arg_Intrinsic;
 312
 313 // Scalar Negate Value
 314 def int_aarch64_neon_vneg :
 315   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 316
 317 // Signed Saturating Doubling Multiply-Add Long
 318 def int_aarch64_neon_vqdmlal : Neon_3Arg_Long_Intrinsic;
 319
 320 // Signed Saturating Doubling Multiply-Subtract Long
 321 def int_aarch64_neon_vqdmlsl : Neon_3Arg_Long_Intrinsic;
 322
 323 def int_aarch64_neon_vmull_p64 :
 324   Intrinsic<[llvm_v16i8_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 325
 326 class Neon_2Arg_ShiftImm_Intrinsic
 327   : Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>;
 328
 329 class Neon_3Arg_ShiftImm_Intrinsic
 330   : Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty, llvm_i32_ty],
 331               [IntrNoMem]>;
 332
 333 // Scalar Shift Right (Immediate)
 334 def int_aarch64_neon_vshrds_n : Neon_2Arg_ShiftImm_Intrinsic;
 335 def int_aarch64_neon_vshrdu_n : Neon_2Arg_ShiftImm_Intrinsic;
 336
 337 // Scalar Shift Right and Accumulate (Immediate)
 338 def int_aarch64_neon_vsrads_n : Neon_3Arg_ShiftImm_Intrinsic;
 339 def int_aarch64_neon_vsradu_n : Neon_3Arg_ShiftImm_Intrinsic;
 340
 341 // Scalar Rounding Shift Right and Accumulate (Immediate)
 342 def int_aarch64_neon_vrsrads_n : Neon_3Arg_ShiftImm_Intrinsic;
 343 def int_aarch64_neon_vrsradu_n : Neon_3Arg_ShiftImm_Intrinsic;
 344
 345 // Scalar Shift Left (Immediate)
 346 def int_aarch64_neon_vshld_n : Neon_2Arg_ShiftImm_Intrinsic;
 347
 348 // Scalar Saturating Shift Left (Immediate)
 349 def int_aarch64_neon_vqshls_n : Neon_N2V_Intrinsic;
 350 def int_aarch64_neon_vqshlu_n : Neon_N2V_Intrinsic;
 351
 352 // Scalar Signed Saturating Shift Left Unsigned (Immediate)
 353 def int_aarch64_neon_vqshlus_n : Neon_N2V_Intrinsic;
 354
 355 // Scalar Signed Fixed-point Convert To Floating-Point (Immediate)
 356 def int_aarch64_neon_vcvtf32_n_s32 :
 357   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty, llvm_i32_ty], [IntrNoMem]>;
 358 def int_aarch64_neon_vcvtf64_n_s64 :
 359   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>;
 360
 361 // Scalar Unsigned Fixed-point Convert To Floating-Point (Immediate)
 362 def int_aarch64_neon_vcvtf32_n_u32 :
 363   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty, llvm_i32_ty], [IntrNoMem]>;
 364 def int_aarch64_neon_vcvtf64_n_u64 :
 365   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>;
 366
 367 // Scalar Floating-point Convert To Signed Fixed-point (Immediate)
 368 def int_aarch64_neon_vcvts_n_s32_f32 :
 369   Intrinsic<[llvm_v1i32_ty], [llvm_v1f32_ty, llvm_i32_ty], [IntrNoMem]>;
 370 def int_aarch64_neon_vcvtd_n_s64_f64 :
 371   Intrinsic<[llvm_v1i64_ty], [llvm_v1f64_ty, llvm_i32_ty], [IntrNoMem]>;
 372
 373 // Scalar Floating-point Convert To Unsigned Fixed-point (Immediate)
 374 def int_aarch64_neon_vcvts_n_u32_f32 :
 375   Intrinsic<[llvm_v1i32_ty], [llvm_v1f32_ty, llvm_i32_ty], [IntrNoMem]>;
 376 def int_aarch64_neon_vcvtd_n_u64_f64 :
 377   Intrinsic<[llvm_v1i64_ty], [llvm_v1f64_ty, llvm_i32_ty], [IntrNoMem]>;
 378
 379 class Neon_SHA_Intrinsic
 380   : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v1i32_ty, llvm_v4i32_ty],
 381               [IntrNoMem]>;
 382
 383 def int_aarch64_neon_sha1c : Neon_SHA_Intrinsic;
 384 def int_aarch64_neon_sha1m : Neon_SHA_Intrinsic;
 385 def int_aarch64_neon_sha1p : Neon_SHA_Intrinsic;
 386 }