include/llvm/IR/IntrinsicsARM.td

   1 //===- IntrinsicsARM.td - Defines ARM 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 ARM-specific intrinsics.
  11 //
  12 //===----------------------------------------------------------------------===//
  13
  14
  15 //===----------------------------------------------------------------------===//
  16 // TLS
  17
  18 let TargetPrefix = "arm" in {  // All intrinsics start with "llvm.arm.".
  19
  20 def int_arm_thread_pointer : GCCBuiltin<"__builtin_thread_pointer">,
  21             Intrinsic<[llvm_ptr_ty], [], [IntrNoMem]>;
  22
  23 //===----------------------------------------------------------------------===//
  24 // Saturating Arithmentic
  25
  26 def int_arm_qadd : GCCBuiltin<"__builtin_arm_qadd">,
  27     Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
  28     [IntrNoMem, Commutative]>;
  29 def int_arm_qsub : GCCBuiltin<"__builtin_arm_qsub">,
  30     Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
  31 def int_arm_ssat : GCCBuiltin<"__builtin_arm_ssat">,
  32     Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
  33 def int_arm_usat : GCCBuiltin<"__builtin_arm_usat">,
  34     Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
  35
  36 //===----------------------------------------------------------------------===//
  37 // Load, Store and Clear exclusive
  38
  39 def int_arm_ldrex : Intrinsic<[llvm_i32_ty], [llvm_anyptr_ty]>;
  40 def int_arm_strex : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_anyptr_ty]>;
  41 def int_arm_clrex : Intrinsic<[]>;
  42
  43 def int_arm_strexd : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty,
  44     llvm_ptr_ty]>;
  45 def int_arm_ldrexd : Intrinsic<[llvm_i32_ty, llvm_i32_ty], [llvm_ptr_ty]>;
  46
  47 //===----------------------------------------------------------------------===//
  48 // Data barrier instructions
  49 def int_arm_dmb : GCCBuiltin<"__builtin_arm_dmb">, Intrinsic<[], [llvm_i32_ty]>;
  50 def int_arm_dsb : GCCBuiltin<"__builtin_arm_dsb">, Intrinsic<[], [llvm_i32_ty]>;
  51
  52 //===----------------------------------------------------------------------===//
  53 // VFP
  54
  55 def int_arm_get_fpscr : GCCBuiltin<"__builtin_arm_get_fpscr">,
  56                        Intrinsic<[llvm_i32_ty], [], [IntrNoMem]>;
  57 def int_arm_set_fpscr : GCCBuiltin<"__builtin_arm_set_fpscr">,
  58                        Intrinsic<[], [llvm_i32_ty], []>;
  59 def int_arm_vcvtr     : Intrinsic<[llvm_float_ty], [llvm_anyfloat_ty],
  60                                   [IntrNoMem]>;
  61 def int_arm_vcvtru    : Intrinsic<[llvm_float_ty], [llvm_anyfloat_ty],
  62                                   [IntrNoMem]>;
  63
  64 //===----------------------------------------------------------------------===//
  65 // Coprocessor
  66
  67 // Move to coprocessor
  68 def int_arm_mcr : GCCBuiltin<"__builtin_arm_mcr">,
  69    Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  70                   llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>;
  71 def int_arm_mcr2 : GCCBuiltin<"__builtin_arm_mcr2">,
  72    Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  73                   llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>;
  74
  75 // Move from coprocessor
  76 def int_arm_mrc : GCCBuiltin<"__builtin_arm_mrc">,
  77    Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  78                              llvm_i32_ty, llvm_i32_ty], []>;
  79 def int_arm_mrc2 : GCCBuiltin<"__builtin_arm_mrc2">,
  80    Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  81                              llvm_i32_ty, llvm_i32_ty], []>;
  82
  83 // Coprocessor data processing
  84 def int_arm_cdp : GCCBuiltin<"__builtin_arm_cdp">,
  85    Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  86                   llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>;
  87 def int_arm_cdp2 : GCCBuiltin<"__builtin_arm_cdp2">,
  88    Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  89                   llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], []>;
  90
  91 // Move from two registers to coprocessor
  92 def int_arm_mcrr : GCCBuiltin<"__builtin_arm_mcrr">,
  93    Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  94                   llvm_i32_ty, llvm_i32_ty], []>;
  95 def int_arm_mcrr2 : GCCBuiltin<"__builtin_arm_mcrr2">,
  96    Intrinsic<[], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty,
  97                   llvm_i32_ty, llvm_i32_ty], []>;
  98
  99 //===----------------------------------------------------------------------===//
 100 // CRC32
 101
 102 def int_arm_crc32b  : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
 103     [IntrNoMem]>;
 104 def int_arm_crc32cb : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
 105     [IntrNoMem]>;
 106 def int_arm_crc32h  : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
 107     [IntrNoMem]>;
 108 def int_arm_crc32ch : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
 109     [IntrNoMem]>;
 110 def int_arm_crc32w  : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
 111     [IntrNoMem]>;
 112 def int_arm_crc32cw : Intrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty],
 113     [IntrNoMem]>;
 114
 115 //===----------------------------------------------------------------------===//
 116 // HINT
 117 def int_arm_sevl : Intrinsic<[], []>;
 118
 119 //===----------------------------------------------------------------------===//
 120 // Advanced SIMD (NEON)
 121
 122 // The following classes do not correspond directly to GCC builtins.
 123 class Neon_1Arg_Intrinsic
 124   : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>], [IntrNoMem]>;
 125 class Neon_1Arg_Narrow_Intrinsic
 126   : Intrinsic<[llvm_anyvector_ty],
 127               [LLVMExtendedElementVectorType<0>], [IntrNoMem]>;
 128 class Neon_2Arg_Intrinsic
 129   : Intrinsic<[llvm_anyvector_ty], [LLVMMatchType<0>, LLVMMatchType<0>],
 130               [IntrNoMem]>;
 131 class Neon_2Arg_Narrow_Intrinsic
 132   : Intrinsic<[llvm_anyvector_ty],
 133               [LLVMExtendedElementVectorType<0>,
 134                LLVMExtendedElementVectorType<0>],
 135               [IntrNoMem]>;
 136 class Neon_2Arg_Long_Intrinsic
 137   : Intrinsic<[llvm_anyvector_ty],
 138               [LLVMTruncatedElementVectorType<0>,
 139                LLVMTruncatedElementVectorType<0>],
 140               [IntrNoMem]>;
 141 class Neon_3Arg_Intrinsic
 142   : Intrinsic<[llvm_anyvector_ty],
 143               [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
 144               [IntrNoMem]>;
 145 class Neon_3Arg_Long_Intrinsic
 146   : Intrinsic<[llvm_anyvector_ty],
 147               [LLVMMatchType<0>,
 148                LLVMTruncatedElementVectorType<0>,
 149                LLVMTruncatedElementVectorType<0>],
 150               [IntrNoMem]>;
 151 class Neon_CvtFxToFP_Intrinsic
 152   : Intrinsic<[llvm_anyfloat_ty], [llvm_anyint_ty, llvm_i32_ty], [IntrNoMem]>;
 153 class Neon_CvtFPToFx_Intrinsic
 154   : Intrinsic<[llvm_anyint_ty], [llvm_anyfloat_ty, llvm_i32_ty], [IntrNoMem]>;
 155 class Neon_CvtFPtoInt_1Arg_Intrinsic
 156   : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty], [IntrNoMem]>;
 157
 158 class Neon_Compare_Intrinsic
 159   : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty, LLVMMatchType<1>],
 160               [IntrNoMem]>;
 161
 162 // The table operands for VTBL and VTBX consist of 1 to 4 v8i8 vectors.
 163 // Besides the table, VTBL has one other v8i8 argument and VTBX has two.
 164 // Overall, the classes range from 2 to 6 v8i8 arguments.
 165 class Neon_Tbl2Arg_Intrinsic
 166   : Intrinsic<[llvm_v8i8_ty],
 167               [llvm_v8i8_ty, llvm_v8i8_ty], [IntrNoMem]>;
 168 class Neon_Tbl3Arg_Intrinsic
 169   : Intrinsic<[llvm_v8i8_ty],
 170               [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty], [IntrNoMem]>;
 171 class Neon_Tbl4Arg_Intrinsic
 172   : Intrinsic<[llvm_v8i8_ty],
 173               [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty],
 174               [IntrNoMem]>;
 175 class Neon_Tbl5Arg_Intrinsic
 176   : Intrinsic<[llvm_v8i8_ty],
 177               [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty,
 178                llvm_v8i8_ty], [IntrNoMem]>;
 179 class Neon_Tbl6Arg_Intrinsic
 180   : Intrinsic<[llvm_v8i8_ty],
 181               [llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty, llvm_v8i8_ty,
 182                llvm_v8i8_ty, llvm_v8i8_ty], [IntrNoMem]>;
 183
 184 // Arithmetic ops
 185
 186 let Properties = [IntrNoMem, Commutative] in {
 187
 188   // Vector Add.
 189   def int_arm_neon_vhadds : Neon_2Arg_Intrinsic;
 190   def int_arm_neon_vhaddu : Neon_2Arg_Intrinsic;
 191   def int_arm_neon_vrhadds : Neon_2Arg_Intrinsic;
 192   def int_arm_neon_vrhaddu : Neon_2Arg_Intrinsic;
 193   def int_arm_neon_vqadds : Neon_2Arg_Intrinsic;
 194   def int_arm_neon_vqaddu : Neon_2Arg_Intrinsic;
 195   def int_arm_neon_vraddhn : Neon_2Arg_Narrow_Intrinsic;
 196
 197   // Vector Multiply.
 198   def int_arm_neon_vmulp : Neon_2Arg_Intrinsic;
 199   def int_arm_neon_vqdmulh : Neon_2Arg_Intrinsic;
 200   def int_arm_neon_vqrdmulh : Neon_2Arg_Intrinsic;
 201   def int_arm_neon_vmulls : Neon_2Arg_Long_Intrinsic;
 202   def int_arm_neon_vmullu : Neon_2Arg_Long_Intrinsic;
 203   def int_arm_neon_vmullp : Neon_2Arg_Long_Intrinsic;
 204   def int_arm_neon_vqdmull : Neon_2Arg_Long_Intrinsic;
 205
 206   // Vector Maximum.
 207   def int_arm_neon_vmaxs : Neon_2Arg_Intrinsic;
 208   def int_arm_neon_vmaxu : Neon_2Arg_Intrinsic;
 209   def int_arm_neon_vmaxnm : Neon_2Arg_Intrinsic;
 210
 211   // Vector Minimum.
 212   def int_arm_neon_vmins : Neon_2Arg_Intrinsic;
 213   def int_arm_neon_vminu : Neon_2Arg_Intrinsic;
 214   def int_arm_neon_vminnm : Neon_2Arg_Intrinsic;
 215
 216   // Vector Reciprocal Step.
 217   def int_arm_neon_vrecps : Neon_2Arg_Intrinsic;
 218
 219   // Vector Reciprocal Square Root Step.
 220   def int_arm_neon_vrsqrts : Neon_2Arg_Intrinsic;
 221 }
 222
 223 // Vector Subtract.
 224 def int_arm_neon_vhsubs : Neon_2Arg_Intrinsic;
 225 def int_arm_neon_vhsubu : Neon_2Arg_Intrinsic;
 226 def int_arm_neon_vqsubs : Neon_2Arg_Intrinsic;
 227 def int_arm_neon_vqsubu : Neon_2Arg_Intrinsic;
 228 def int_arm_neon_vrsubhn : Neon_2Arg_Narrow_Intrinsic;
 229
 230 // Vector Absolute Compare.
 231 def int_arm_neon_vacge : Neon_Compare_Intrinsic;
 232 def int_arm_neon_vacgt : Neon_Compare_Intrinsic;
 233
 234 // Vector Absolute Differences.
 235 def int_arm_neon_vabds : Neon_2Arg_Intrinsic;
 236 def int_arm_neon_vabdu : Neon_2Arg_Intrinsic;
 237
 238 // Vector Pairwise Add.
 239 def int_arm_neon_vpadd : Neon_2Arg_Intrinsic;
 240
 241 // Vector Pairwise Add Long.
 242 // Note: This is different than the other "long" NEON intrinsics because
 243 // the result vector has half as many elements as the source vector.
 244 // The source and destination vector types must be specified separately.
 245 def int_arm_neon_vpaddls : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty],
 246                                      [IntrNoMem]>;
 247 def int_arm_neon_vpaddlu : Intrinsic<[llvm_anyvector_ty], [llvm_anyvector_ty],
 248                                      [IntrNoMem]>;
 249
 250 // Vector Pairwise Add and Accumulate Long.
 251 // Note: This is similar to vpaddl but the destination vector also appears
 252 // as the first argument.
 253 def int_arm_neon_vpadals : Intrinsic<[llvm_anyvector_ty],
 254                                      [LLVMMatchType<0>, llvm_anyvector_ty],
 255                                      [IntrNoMem]>;
 256 def int_arm_neon_vpadalu : Intrinsic<[llvm_anyvector_ty],
 257                                      [LLVMMatchType<0>, llvm_anyvector_ty],
 258                                      [IntrNoMem]>;
 259
 260 // Vector Pairwise Maximum and Minimum.
 261 def int_arm_neon_vpmaxs : Neon_2Arg_Intrinsic;
 262 def int_arm_neon_vpmaxu : Neon_2Arg_Intrinsic;
 263 def int_arm_neon_vpmins : Neon_2Arg_Intrinsic;
 264 def int_arm_neon_vpminu : Neon_2Arg_Intrinsic;
 265
 266 // Vector Shifts:
 267 //
 268 // The various saturating and rounding vector shift operations need to be
 269 // represented by intrinsics in LLVM, and even the basic VSHL variable shift
 270 // operation cannot be safely translated to LLVM's shift operators.  VSHL can
 271 // be used for both left and right shifts, or even combinations of the two,
 272 // depending on the signs of the shift amounts.  It also has well-defined
 273 // behavior for shift amounts that LLVM leaves undefined.  Only basic shifts
 274 // by constants can be represented with LLVM's shift operators.
 275 //
 276 // The shift counts for these intrinsics are always vectors, even for constant
 277 // shifts, where the constant is replicated.  For consistency with VSHL (and
 278 // other variable shift instructions), left shifts have positive shift counts
 279 // and right shifts have negative shift counts.  This convention is also used
 280 // for constant right shift intrinsics, and to help preserve sanity, the
 281 // intrinsic names use "shift" instead of either "shl" or "shr".  Where
 282 // applicable, signed and unsigned versions of the intrinsics are
 283 // distinguished with "s" and "u" suffixes.  A few NEON shift instructions,
 284 // such as VQSHLU, take signed operands but produce unsigned results; these
 285 // use a "su" suffix.
 286
 287 // Vector Shift.
 288 def int_arm_neon_vshifts : Neon_2Arg_Intrinsic;
 289 def int_arm_neon_vshiftu : Neon_2Arg_Intrinsic;
 290
 291 // Vector Rounding Shift.
 292 def int_arm_neon_vrshifts : Neon_2Arg_Intrinsic;
 293 def int_arm_neon_vrshiftu : Neon_2Arg_Intrinsic;
 294 def int_arm_neon_vrshiftn : Neon_2Arg_Narrow_Intrinsic;
 295
 296 // Vector Saturating Shift.
 297 def int_arm_neon_vqshifts : Neon_2Arg_Intrinsic;
 298 def int_arm_neon_vqshiftu : Neon_2Arg_Intrinsic;
 299 def int_arm_neon_vqshiftsu : Neon_2Arg_Intrinsic;
 300 def int_arm_neon_vqshiftns : Neon_2Arg_Narrow_Intrinsic;
 301 def int_arm_neon_vqshiftnu : Neon_2Arg_Narrow_Intrinsic;
 302 def int_arm_neon_vqshiftnsu : Neon_2Arg_Narrow_Intrinsic;
 303
 304 // Vector Saturating Rounding Shift.
 305 def int_arm_neon_vqrshifts : Neon_2Arg_Intrinsic;
 306 def int_arm_neon_vqrshiftu : Neon_2Arg_Intrinsic;
 307 def int_arm_neon_vqrshiftns : Neon_2Arg_Narrow_Intrinsic;
 308 def int_arm_neon_vqrshiftnu : Neon_2Arg_Narrow_Intrinsic;
 309 def int_arm_neon_vqrshiftnsu : Neon_2Arg_Narrow_Intrinsic;
 310
 311 // Vector Shift and Insert.
 312 def int_arm_neon_vshiftins : Neon_3Arg_Intrinsic;
 313
 314 // Vector Absolute Value and Saturating Absolute Value.
 315 def int_arm_neon_vabs : Neon_1Arg_Intrinsic;
 316 def int_arm_neon_vqabs : Neon_1Arg_Intrinsic;
 317
 318 // Vector Saturating Negate.
 319 def int_arm_neon_vqneg : Neon_1Arg_Intrinsic;
 320
 321 // Vector Count Leading Sign/Zero Bits.
 322 def int_arm_neon_vcls : Neon_1Arg_Intrinsic;
 323 def int_arm_neon_vclz : Neon_1Arg_Intrinsic;
 324
 325 // Vector Count One Bits.
 326 def int_arm_neon_vcnt : Neon_1Arg_Intrinsic;
 327
 328 // Vector Reciprocal Estimate.
 329 def int_arm_neon_vrecpe : Neon_1Arg_Intrinsic;
 330
 331 // Vector Reciprocal Square Root Estimate.
 332 def int_arm_neon_vrsqrte : Neon_1Arg_Intrinsic;
 333
 334 // Vector Conversions Between Floating-point and Integer
 335 def int_arm_neon_vcvtau : Neon_CvtFPtoInt_1Arg_Intrinsic;
 336 def int_arm_neon_vcvtas : Neon_CvtFPtoInt_1Arg_Intrinsic;
 337 def int_arm_neon_vcvtnu : Neon_CvtFPtoInt_1Arg_Intrinsic;
 338 def int_arm_neon_vcvtns : Neon_CvtFPtoInt_1Arg_Intrinsic;
 339 def int_arm_neon_vcvtpu : Neon_CvtFPtoInt_1Arg_Intrinsic;
 340 def int_arm_neon_vcvtps : Neon_CvtFPtoInt_1Arg_Intrinsic;
 341 def int_arm_neon_vcvtmu : Neon_CvtFPtoInt_1Arg_Intrinsic;
 342 def int_arm_neon_vcvtms : Neon_CvtFPtoInt_1Arg_Intrinsic;
 343
 344 // Vector Conversions Between Floating-point and Fixed-point.
 345 def int_arm_neon_vcvtfp2fxs : Neon_CvtFPToFx_Intrinsic;
 346 def int_arm_neon_vcvtfp2fxu : Neon_CvtFPToFx_Intrinsic;
 347 def int_arm_neon_vcvtfxs2fp : Neon_CvtFxToFP_Intrinsic;
 348 def int_arm_neon_vcvtfxu2fp : Neon_CvtFxToFP_Intrinsic;
 349
 350 // Vector Conversions Between Half-Precision and Single-Precision.
 351 def int_arm_neon_vcvtfp2hf
 352     : Intrinsic<[llvm_v4i16_ty], [llvm_v4f32_ty], [IntrNoMem]>;
 353 def int_arm_neon_vcvthf2fp
 354     : Intrinsic<[llvm_v4f32_ty], [llvm_v4i16_ty], [IntrNoMem]>;
 355
 356 // Narrowing Saturating Vector Moves.
 357 def int_arm_neon_vqmovns : Neon_1Arg_Narrow_Intrinsic;
 358 def int_arm_neon_vqmovnu : Neon_1Arg_Narrow_Intrinsic;
 359 def int_arm_neon_vqmovnsu : Neon_1Arg_Narrow_Intrinsic;
 360
 361 // Vector Table Lookup.
 362 // The first 1-4 arguments are the table.
 363 def int_arm_neon_vtbl1 : Neon_Tbl2Arg_Intrinsic;
 364 def int_arm_neon_vtbl2 : Neon_Tbl3Arg_Intrinsic;
 365 def int_arm_neon_vtbl3 : Neon_Tbl4Arg_Intrinsic;
 366 def int_arm_neon_vtbl4 : Neon_Tbl5Arg_Intrinsic;
 367
 368 // Vector Table Extension.
 369 // Some elements of the destination vector may not be updated, so the original
 370 // value of that vector is passed as the first argument.  The next 1-4
 371 // arguments after that are the table.
 372 def int_arm_neon_vtbx1 : Neon_Tbl3Arg_Intrinsic;
 373 def int_arm_neon_vtbx2 : Neon_Tbl4Arg_Intrinsic;
 374 def int_arm_neon_vtbx3 : Neon_Tbl5Arg_Intrinsic;
 375 def int_arm_neon_vtbx4 : Neon_Tbl6Arg_Intrinsic;
 376
 377 // Vector Rounding
 378 def int_arm_neon_vrintn : Neon_1Arg_Intrinsic;
 379 def int_arm_neon_vrintx : Neon_1Arg_Intrinsic;
 380 def int_arm_neon_vrinta : Neon_1Arg_Intrinsic;
 381 def int_arm_neon_vrintz : Neon_1Arg_Intrinsic;
 382 def int_arm_neon_vrintm : Neon_1Arg_Intrinsic;
 383 def int_arm_neon_vrintp : Neon_1Arg_Intrinsic;
 384
 385 // De-interleaving vector loads from N-element structures.
 386 // Source operands are the address and alignment.
 387 def int_arm_neon_vld1 : Intrinsic<[llvm_anyvector_ty],
 388                                   [llvm_ptr_ty, llvm_i32_ty],
 389                                   [IntrReadArgMem]>;
 390 def int_arm_neon_vld2 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>],
 391                                   [llvm_ptr_ty, llvm_i32_ty],
 392                                   [IntrReadArgMem]>;
 393 def int_arm_neon_vld3 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 394                                    LLVMMatchType<0>],
 395                                   [llvm_ptr_ty, llvm_i32_ty],
 396                                   [IntrReadArgMem]>;
 397 def int_arm_neon_vld4 : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 398                                    LLVMMatchType<0>, LLVMMatchType<0>],
 399                                   [llvm_ptr_ty, llvm_i32_ty],
 400                                   [IntrReadArgMem]>;
 401
 402 // Vector load N-element structure to one lane.
 403 // Source operands are: the address, the N input vectors (since only one
 404 // lane is assigned), the lane number, and the alignment.
 405 def int_arm_neon_vld2lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>],
 406                                       [llvm_ptr_ty, LLVMMatchType<0>,
 407                                        LLVMMatchType<0>, llvm_i32_ty,
 408                                        llvm_i32_ty], [IntrReadArgMem]>;
 409 def int_arm_neon_vld3lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 410                                        LLVMMatchType<0>],
 411                                       [llvm_ptr_ty, LLVMMatchType<0>,
 412                                        LLVMMatchType<0>, LLVMMatchType<0>,
 413                                        llvm_i32_ty, llvm_i32_ty],
 414                                       [IntrReadArgMem]>;
 415 def int_arm_neon_vld4lane : Intrinsic<[llvm_anyvector_ty, LLVMMatchType<0>,
 416                                        LLVMMatchType<0>, LLVMMatchType<0>],
 417                                       [llvm_ptr_ty, LLVMMatchType<0>,
 418                                        LLVMMatchType<0>, LLVMMatchType<0>,
 419                                        LLVMMatchType<0>, llvm_i32_ty,
 420                                        llvm_i32_ty], [IntrReadArgMem]>;
 421
 422 // Interleaving vector stores from N-element structures.
 423 // Source operands are: the address, the N vectors, and the alignment.
 424 def int_arm_neon_vst1 : Intrinsic<[],
 425                                   [llvm_ptr_ty, llvm_anyvector_ty,
 426                                    llvm_i32_ty], [IntrReadWriteArgMem]>;
 427 def int_arm_neon_vst2 : Intrinsic<[],
 428                                   [llvm_ptr_ty, llvm_anyvector_ty,
 429                                    LLVMMatchType<0>, llvm_i32_ty],
 430                                   [IntrReadWriteArgMem]>;
 431 def int_arm_neon_vst3 : Intrinsic<[],
 432                                   [llvm_ptr_ty, llvm_anyvector_ty,
 433                                    LLVMMatchType<0>, LLVMMatchType<0>,
 434                                    llvm_i32_ty], [IntrReadWriteArgMem]>;
 435 def int_arm_neon_vst4 : Intrinsic<[],
 436                                   [llvm_ptr_ty, llvm_anyvector_ty,
 437                                    LLVMMatchType<0>, LLVMMatchType<0>,
 438                                    LLVMMatchType<0>, llvm_i32_ty],
 439                                   [IntrReadWriteArgMem]>;
 440
 441 // Vector store N-element structure from one lane.
 442 // Source operands are: the address, the N vectors, the lane number, and
 443 // the alignment.
 444 def int_arm_neon_vst2lane : Intrinsic<[],
 445                                       [llvm_ptr_ty, llvm_anyvector_ty,
 446                                        LLVMMatchType<0>, llvm_i32_ty,
 447                                        llvm_i32_ty], [IntrReadWriteArgMem]>;
 448 def int_arm_neon_vst3lane : Intrinsic<[],
 449                                       [llvm_ptr_ty, llvm_anyvector_ty,
 450                                        LLVMMatchType<0>, LLVMMatchType<0>,
 451                                        llvm_i32_ty, llvm_i32_ty],
 452                                       [IntrReadWriteArgMem]>;
 453 def int_arm_neon_vst4lane : Intrinsic<[],
 454                                       [llvm_ptr_ty, llvm_anyvector_ty,
 455                                        LLVMMatchType<0>, LLVMMatchType<0>,
 456                                        LLVMMatchType<0>, llvm_i32_ty,
 457                                        llvm_i32_ty], [IntrReadWriteArgMem]>;
 458
 459 // Vector bitwise select.
 460 def int_arm_neon_vbsl : Intrinsic<[llvm_anyvector_ty],
 461                         [LLVMMatchType<0>, LLVMMatchType<0>, LLVMMatchType<0>],
 462                         [IntrNoMem]>;
 463
 464
 465 // Crypto instructions
 466 class AES_1Arg_Intrinsic : Intrinsic<[llvm_v16i8_ty],
 467                                      [llvm_v16i8_ty], [IntrNoMem]>;
 468 class AES_2Arg_Intrinsic : Intrinsic<[llvm_v16i8_ty],
 469                                      [llvm_v16i8_ty, llvm_v16i8_ty],
 470                                      [IntrNoMem]>;
 471
 472 class SHA_1Arg_Intrinsic : Intrinsic<[llvm_i32_ty], [llvm_i32_ty],
 473                                      [IntrNoMem]>;
 474 class SHA_2Arg_Intrinsic : Intrinsic<[llvm_v4i32_ty],
 475                                      [llvm_v4i32_ty, llvm_v4i32_ty],
 476                                      [IntrNoMem]>;
 477 class SHA_3Arg_i32_Intrinsic : Intrinsic<[llvm_v4i32_ty],
 478                                    [llvm_v4i32_ty, llvm_i32_ty, llvm_v4i32_ty],
 479                                    [IntrNoMem]>;
 480 class SHA_3Arg_v4i32_Intrinsic : Intrinsic<[llvm_v4i32_ty],
 481                                    [llvm_v4i32_ty, llvm_v4i32_ty,llvm_v4i32_ty],
 482                                    [IntrNoMem]>;
 483
 484 def int_arm_neon_aesd : AES_2Arg_Intrinsic;
 485 def int_arm_neon_aese : AES_2Arg_Intrinsic;
 486 def int_arm_neon_aesimc : AES_1Arg_Intrinsic;
 487 def int_arm_neon_aesmc : AES_1Arg_Intrinsic;
 488 def int_arm_neon_sha1h : SHA_1Arg_Intrinsic;
 489 def int_arm_neon_sha1su1 : SHA_2Arg_Intrinsic;
 490 def int_arm_neon_sha256su0 : SHA_2Arg_Intrinsic;
 491 def int_arm_neon_sha1c : SHA_3Arg_i32_Intrinsic;
 492 def int_arm_neon_sha1m : SHA_3Arg_i32_Intrinsic;
 493 def int_arm_neon_sha1p : SHA_3Arg_i32_Intrinsic;
 494 def int_arm_neon_sha1su0: SHA_3Arg_v4i32_Intrinsic;
 495 def int_arm_neon_sha256h: SHA_3Arg_v4i32_Intrinsic;
 496 def int_arm_neon_sha256h2: SHA_3Arg_v4i32_Intrinsic;
 497 def int_arm_neon_sha256su1: SHA_3Arg_v4i32_Intrinsic;
 498
 499 } // end TargetPrefix