contrib/llvm/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_fcvtxn :
  40   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  41 def int_aarch64_neon_fcvtns :
  42   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  43 def int_aarch64_neon_fcvtnu :
  44   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  45 def int_aarch64_neon_fcvtps :
  46   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  47 def int_aarch64_neon_fcvtpu :
  48   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  49 def int_aarch64_neon_fcvtms :
  50   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  51 def int_aarch64_neon_fcvtmu :
  52   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  53 def int_aarch64_neon_fcvtas :
  54   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  55 def int_aarch64_neon_fcvtau :
  56   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  57 def int_aarch64_neon_fcvtzs :
  58   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  59 def int_aarch64_neon_fcvtzu :
  60   Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
  61
  62 // Vector maxNum (Floating Point)
  63 def int_aarch64_neon_vmaxnm : Neon_2Arg_Intrinsic;
  64
  65 // Vector minNum (Floating Point)
  66 def int_aarch64_neon_vminnm : Neon_2Arg_Intrinsic;
  67
  68 // Vector Pairwise maxNum (Floating Point)
  69 def int_aarch64_neon_vpmaxnm : Neon_2Arg_Intrinsic;
  70
  71 // Vector Pairwise minNum (Floating Point)
  72 def int_aarch64_neon_vpminnm : Neon_2Arg_Intrinsic;
  73
  74 // Vector Multiply Extended and Scalar Multiply Extended (Floating Point)
  75 def int_aarch64_neon_vmulx  :
  76   Intrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>, LLVMMatchType<0>]>;
  77
  78 class Neon_N2V_Intrinsic
  79   : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>, llvm_i32_ty],
  80               [IntrNoMem]>;
  81 class Neon_N3V_Intrinsic
  82   : Intrinsic<[llvm_anyvector_ty],
  83               [LLVMMatchType<0>, LLVMMatchType<0>, llvm_i32_ty],
  84               [IntrNoMem]>;
  85 class Neon_N2V_Narrow_Intrinsic
  86   : Intrinsic<[llvm_anyvector_ty],
  87               [LLVMExtendedElementVectorType<0>, llvm_i32_ty],
  88               [IntrNoMem]>;
  89
  90 // Vector rounding shift right by immediate (Signed)
  91 def int_aarch64_neon_vsrshr : Neon_N2V_Intrinsic;
  92 def int_aarch64_neon_vurshr : Neon_N2V_Intrinsic;
  93 def int_aarch64_neon_vsqshlu : Neon_N2V_Intrinsic;
  94
  95 def int_aarch64_neon_vsri : Neon_N3V_Intrinsic;
  96 def int_aarch64_neon_vsli : Neon_N3V_Intrinsic;
  97
  98 def int_aarch64_neon_vsqshrun : Neon_N2V_Narrow_Intrinsic;
  99 def int_aarch64_neon_vrshrn : Neon_N2V_Narrow_Intrinsic;
 100 def int_aarch64_neon_vsqrshrun : Neon_N2V_Narrow_Intrinsic;
 101 def int_aarch64_neon_vsqshrn : Neon_N2V_Narrow_Intrinsic;
 102 def int_aarch64_neon_vuqshrn : Neon_N2V_Narrow_Intrinsic;
 103 def int_aarch64_neon_vsqrshrn : Neon_N2V_Narrow_Intrinsic;
 104 def int_aarch64_neon_vuqrshrn : Neon_N2V_Narrow_Intrinsic;
 105
 106 // Vector across
 107 class Neon_Across_Intrinsic
 108   : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
 109
 110 class Neon_2Arg_Across_Float_Intrinsic
 111   : Intrinsic<[llvm_anyvector_ty], [llvm_v4f32_ty], [IntrNoMem]>;
 112
 113 def int_aarch64_neon_saddlv : Neon_Across_Intrinsic;
 114 def int_aarch64_neon_uaddlv : Neon_Across_Intrinsic;
 115 def int_aarch64_neon_smaxv  : Neon_Across_Intrinsic;
 116 def int_aarch64_neon_umaxv  : Neon_Across_Intrinsic;
 117 def int_aarch64_neon_sminv  : Neon_Across_Intrinsic;
 118 def int_aarch64_neon_uminv  : Neon_Across_Intrinsic;
 119 def int_aarch64_neon_vaddv  : Neon_Across_Intrinsic;
 120 def int_aarch64_neon_vmaxv  : Neon_Across_Intrinsic;
 121 def int_aarch64_neon_vminv  : Neon_Across_Intrinsic;
 122 def int_aarch64_neon_vmaxnmv : Neon_Across_Intrinsic;
 123 def int_aarch64_neon_vminnmv : Neon_Across_Intrinsic;
 124
 125 // Vector Table Lookup.
 126 def int_aarch64_neon_vtbl1 :
 127   Intrinsic<[llvm_anyvector_ty],
 128             [llvm_anyvector_ty, LLVMMatchType<0>], [IntrNoMem]>;
 129
 130 def int_aarch64_neon_vtbl2 :
 131   Intrinsic<[llvm_anyvector_ty],
 132             [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<0>],
 133             [IntrNoMem]>;
 134
 135 def int_aarch64_neon_vtbl3 :
 136   Intrinsic<[llvm_anyvector_ty],
 137             [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<1>,
 138             LLVMMatchType<0>], [IntrNoMem]>;
 139
 140 def int_aarch64_neon_vtbl4 :
 141   Intrinsic<[llvm_anyvector_ty],
 142             [llvm_anyvector_ty, LLVMMatchType<1>, LLVMMatchType<1>,
 143             LLVMMatchType<1>, LLVMMatchType<0>], [IntrNoMem]>;
 144
 145 // Vector Table Extension.
 146 // Some elements of the destination vector may not be updated, so the original
 147 // value of that vector is passed as the first argument.  The next 1-4
 148 // arguments after that are the table.
 149 def int_aarch64_neon_vtbx1 :
 150   Intrinsic<[llvm_anyvector_ty],
 151             [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<0>],
 152             [IntrNoMem]>;
 153
 154 def int_aarch64_neon_vtbx2 :
 155   Intrinsic<[llvm_anyvector_ty],
 156             [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>,
 157             LLVMMatchType<0>], [IntrNoMem]>;
 158
 159 def int_aarch64_neon_vtbx3 :
 160   Intrinsic<[llvm_anyvector_ty],
 161             [LLVMMatchType<0>, llvm_anyvector_ty, LLVMMatchType<1>,
 162             LLVMMatchType<1>, LLVMMatchType<0>], [IntrNoMem]>;
 163
 164 def int_aarch64_neon_vtbx4 :
 165   Intrinsic<[llvm_anyvector_ty],
 166             [LLVMMatchType<0>, llvm_anyvector_ty,  LLVMMatchType<1>,
 167             LLVMMatchType<1>,  LLVMMatchType<1>, LLVMMatchType<0>],
 168             [IntrNoMem]>;
 169
 170 // Vector Load/store
 171 def int_aarch64_neon_vld1x2 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>],
 172                                         [llvm_ptr_ty, llvm_i32_ty],
 173                                         [IntrReadArgMem]>;
 174 def int_aarch64_neon_vld1x3 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 175                                          LLVMMatchType<0>],
 176                                         [llvm_ptr_ty, llvm_i32_ty],
 177                                         [IntrReadArgMem]>;
 178 def int_aarch64_neon_vld1x4 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 179                                          LLVMMatchType<0>, LLVMMatchType<0>],
 180                                         [llvm_ptr_ty, llvm_i32_ty],
 181                                         [IntrReadArgMem]>;
 182
 183 def int_aarch64_neon_vst1x2 : Intrinsic<[],
 184                                         [llvm_ptr_ty, llvm_anyvector_ty,
 185                                          LLVMMatchType<0>, llvm_i32_ty],
 186                                         [IntrReadWriteArgMem]>;
 187 def int_aarch64_neon_vst1x3 : Intrinsic<[],
 188                                         [llvm_ptr_ty, llvm_anyvector_ty,
 189                                          LLVMMatchType<0>, LLVMMatchType<0>,
 190                                          llvm_i32_ty], [IntrReadWriteArgMem]>;
 191 def int_aarch64_neon_vst1x4 : Intrinsic<[],
 192                                         [llvm_ptr_ty, llvm_anyvector_ty,
 193                                          LLVMMatchType<0>, LLVMMatchType<0>,
 194                                          LLVMMatchType<0>, llvm_i32_ty],
 195                                         [IntrReadWriteArgMem]>;
 196
 197 // Scalar Add
 198 def int_aarch64_neon_vaddds :
 199   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 200 def int_aarch64_neon_vadddu :
 201   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 202
 203
 204 // Scalar Sub
 205 def int_aarch64_neon_vsubds :
 206   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 207 def int_aarch64_neon_vsubdu :
 208   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 209
 210
 211 // Scalar Shift
 212 // Scalar Shift Left
 213 def int_aarch64_neon_vshlds :
 214   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 215 def int_aarch64_neon_vshldu :
 216   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 217
 218 // Scalar Saturating Shift Left
 219 def int_aarch64_neon_vqshls : Neon_2Arg_Intrinsic;
 220 def int_aarch64_neon_vqshlu : Neon_2Arg_Intrinsic;
 221
 222 // Scalar Shift Rouding Left
 223 def int_aarch64_neon_vrshlds :
 224   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 225 def int_aarch64_neon_vrshldu :
 226   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty], [IntrNoMem]>;
 227
 228 // Scalar Saturating Rounding Shift Left
 229 def int_aarch64_neon_vqrshls : Neon_2Arg_Intrinsic;
 230 def int_aarch64_neon_vqrshlu : Neon_2Arg_Intrinsic;
 231
 232 // Scalar Reduce Pairwise Add.
 233 def int_aarch64_neon_vpadd :
 234   Intrinsic<[llvm_v1i64_ty], [llvm_v2i64_ty],[IntrNoMem]>;
 235 def int_aarch64_neon_vpfadd :
 236   Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>;
 237 def int_aarch64_neon_vpfaddq :
 238   Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>;
 239
 240 // Scalar Reduce Pairwise Floating Point Max/Min.
 241 def int_aarch64_neon_vpmax :
 242   Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>;
 243 def int_aarch64_neon_vpmaxq :
 244   Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>;
 245 def int_aarch64_neon_vpmin :
 246   Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>;
 247 def int_aarch64_neon_vpminq :
 248   Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>;
 249
 250 // Scalar Reduce Pairwise Floating Point Maxnm/Minnm.
 251 def int_aarch64_neon_vpfmaxnm :
 252   Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>;
 253 def int_aarch64_neon_vpfmaxnmq :
 254   Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>;
 255 def int_aarch64_neon_vpfminnm :
 256   Intrinsic<[llvm_v1f32_ty], [llvm_v2f32_ty], [IntrNoMem]>;
 257 def int_aarch64_neon_vpfminnmq :
 258   Intrinsic<[llvm_v1f64_ty], [llvm_v2f64_ty], [IntrNoMem]>;
 259
 260 // Scalar Signed Integer Convert To Floating-point
 261 def int_aarch64_neon_vcvtf32_s32 :
 262   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty], [IntrNoMem]>;
 263 def int_aarch64_neon_vcvtf64_s64 :
 264   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 265
 266 // Scalar Unsigned Integer Convert To Floating-point
 267 def int_aarch64_neon_vcvtf32_u32 :
 268   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty], [IntrNoMem]>;
 269 def int_aarch64_neon_vcvtf64_u64 :
 270   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 271
 272 // Scalar Floating-point Reciprocal Exponent
 273 def int_aarch64_neon_vrecpx : Neon_1Arg_Intrinsic;
 274
 275 class Neon_Cmp_Intrinsic
 276   : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty, llvm_anyvector_ty],
 277               [IntrNoMem]>;
 278
 279 // Scalar Compare Equal
 280 def int_aarch64_neon_vceq : Neon_Cmp_Intrinsic;
 281
 282 // Scalar Compare Greater-Than or Equal
 283 def int_aarch64_neon_vcge : Neon_Cmp_Intrinsic;
 284 def int_aarch64_neon_vchs : Neon_Cmp_Intrinsic;
 285
 286 // Scalar Compare Less-Than or Equal
 287 def int_aarch64_neon_vclez : Neon_Cmp_Intrinsic;
 288
 289 // Scalar Compare Less-Than
 290 def int_aarch64_neon_vcltz : Neon_Cmp_Intrinsic;
 291
 292 // Scalar Compare Greater-Than
 293 def int_aarch64_neon_vcgt : Neon_Cmp_Intrinsic;
 294 def int_aarch64_neon_vchi : Neon_Cmp_Intrinsic;
 295
 296 // Scalar Compare Bitwise Test Bits
 297 def int_aarch64_neon_vtstd : Neon_Cmp_Intrinsic;
 298
 299 // Scalar Floating-point Absolute Compare Greater Than Or Equal
 300 def int_aarch64_neon_vcage : Neon_Cmp_Intrinsic;
 301
 302 // Scalar Floating-point Absolute Compare Greater Than
 303 def int_aarch64_neon_vcagt : Neon_Cmp_Intrinsic;
 304
 305 // Scalar Signed Saturating Accumulated of Unsigned Value
 306 def int_aarch64_neon_vuqadd : Neon_2Arg_Intrinsic;
 307
 308 // Scalar Unsigned Saturating Accumulated of Signed Value
 309 def int_aarch64_neon_vsqadd : Neon_2Arg_Intrinsic;
 310
 311 // Scalar Absolute Value
 312 def int_aarch64_neon_vabs :
 313   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 314
 315 // Scalar Absolute Difference
 316 def int_aarch64_neon_vabd : Neon_2Arg_Intrinsic;
 317
 318 // Scalar Negate Value
 319 def int_aarch64_neon_vneg :
 320   Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty], [IntrNoMem]>;
 321
 322 // Signed Saturating Doubling Multiply-Add Long
 323 def int_aarch64_neon_vqdmlal : Neon_3Arg_Long_Intrinsic;
 324
 325 // Signed Saturating Doubling Multiply-Subtract Long
 326 def int_aarch64_neon_vqdmlsl : Neon_3Arg_Long_Intrinsic;
 327
 328 class Neon_2Arg_ShiftImm_Intrinsic
 329   : Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>;
 330
 331 class Neon_3Arg_ShiftImm_Intrinsic
 332   : Intrinsic<[llvm_v1i64_ty], [llvm_v1i64_ty, llvm_v1i64_ty, llvm_i32_ty],
 333               [IntrNoMem]>;
 334
 335 // Scalar Shift Right (Immediate)
 336 def int_aarch64_neon_vshrds_n : Neon_2Arg_ShiftImm_Intrinsic;
 337 def int_aarch64_neon_vshrdu_n : Neon_2Arg_ShiftImm_Intrinsic;
 338
 339 // Scalar Shift Right and Accumulate (Immediate)
 340 def int_aarch64_neon_vsrads_n : Neon_3Arg_ShiftImm_Intrinsic;
 341 def int_aarch64_neon_vsradu_n : Neon_3Arg_ShiftImm_Intrinsic;
 342
 343 // Scalar Rounding Shift Right and Accumulate (Immediate)
 344 def int_aarch64_neon_vrsrads_n : Neon_3Arg_ShiftImm_Intrinsic;
 345 def int_aarch64_neon_vrsradu_n : Neon_3Arg_ShiftImm_Intrinsic;
 346
 347 // Scalar Shift Left (Immediate)
 348 def int_aarch64_neon_vshld_n : Neon_2Arg_ShiftImm_Intrinsic;
 349
 350 // Scalar Saturating Shift Left (Immediate)
 351 def int_aarch64_neon_vqshls_n : Neon_N2V_Intrinsic;
 352 def int_aarch64_neon_vqshlu_n : Neon_N2V_Intrinsic;
 353
 354 // Scalar Signed Saturating Shift Left Unsigned (Immediate)
 355 def int_aarch64_neon_vqshlus_n : Neon_N2V_Intrinsic;
 356
 357 // Scalar Signed Fixed-point Convert To Floating-Point (Immediate)
 358 def int_aarch64_neon_vcvtf32_n_s32 :
 359   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty, llvm_i32_ty], [IntrNoMem]>;
 360 def int_aarch64_neon_vcvtf64_n_s64 :
 361   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>;
 362
 363 // Scalar Unsigned Fixed-point Convert To Floating-Point (Immediate)
 364 def int_aarch64_neon_vcvtf32_n_u32 :
 365   Intrinsic<[llvm_float_ty], [llvm_v1i32_ty, llvm_i32_ty], [IntrNoMem]>;
 366 def int_aarch64_neon_vcvtf64_n_u64 :
 367   Intrinsic<[llvm_double_ty], [llvm_v1i64_ty, llvm_i32_ty], [IntrNoMem]>;
 368
 369 // Scalar Floating-point Convert To Signed Fixed-point (Immediate)
 370 def int_aarch64_neon_vcvts_n_s32_f32 :
 371   Intrinsic<[llvm_v1i32_ty], [llvm_v1f32_ty, llvm_i32_ty], [IntrNoMem]>;
 372 def int_aarch64_neon_vcvtd_n_s64_f64 :
 373   Intrinsic<[llvm_v1i64_ty], [llvm_v1f64_ty, llvm_i32_ty], [IntrNoMem]>;
 374
 375 // Scalar Floating-point Convert To Unsigned Fixed-point (Immediate)
 376 def int_aarch64_neon_vcvts_n_u32_f32 :
 377   Intrinsic<[llvm_v1i32_ty], [llvm_v1f32_ty, llvm_i32_ty], [IntrNoMem]>;
 378 def int_aarch64_neon_vcvtd_n_u64_f64 :
 379   Intrinsic<[llvm_v1i64_ty], [llvm_v1f64_ty, llvm_i32_ty], [IntrNoMem]>;
 380
 381 class Neon_SHA_Intrinsic
 382   : Intrinsic<[llvm_v4i32_ty], [llvm_v4i32_ty, llvm_v1i32_ty, llvm_v4i32_ty],
 383               [IntrNoMem]>;
 384
 385 def int_aarch64_neon_sha1c : Neon_SHA_Intrinsic;
 386 def int_aarch64_neon_sha1m : Neon_SHA_Intrinsic;
 387 def int_aarch64_neon_sha1p : Neon_SHA_Intrinsic;
 388 }