gcc/config/aarch64/arm_fp16.h

   1 /* ARM FP16 scalar intrinsics include file.
   2
   3    Copyright (C) 2016-2021 Free Software Foundation, Inc.
   4    Contributed by ARM Ltd.
   5
   6    This file is part of GCC.
   7
   8    GCC is free software; you can redistribute it and/or modify it
   9    under the terms of the GNU General Public License as published
  10    by the Free Software Foundation; either version 3, or (at your
  11    option) any later version.
  12
  13    GCC is distributed in the hope that it will be useful, but WITHOUT
  14    ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
  15    or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public
  16    License for more details.
  17
  18    Under Section 7 of GPL version 3, you are granted additional
  19    permissions described in the GCC Runtime Library Exception, version
  20    3.1, as published by the Free Software Foundation.
  21
  22    You should have received a copy of the GNU General Public License and
  23    a copy of the GCC Runtime Library Exception along with this program;
  24    see the files COPYING3 and COPYING.RUNTIME respectively.  If not, see
  25    <http://www.gnu.org/licenses/>.  */
  26
  27 #ifndef _AARCH64_FP16_H_
  28 #define _AARCH64_FP16_H_
  29
  30 #include <stdint.h>
  31
  32 #pragma GCC push_options
  33 #pragma GCC target ("arch=armv8.2-a+fp16")
  34
  35 typedef __fp16 float16_t;
  36
  37 /* ARMv8.2-A FP16 one operand scalar intrinsics.  */
  38
  39 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
  40 vabsh_f16 (float16_t __a)
  41 {
  42   return __builtin_aarch64_abshf (__a);
  43 }
  44
  45 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
  46 vceqzh_f16 (float16_t __a)
  47 {
  48   return __builtin_aarch64_cmeqhf_uss (__a, 0.0f);
  49 }
  50
  51 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
  52 vcgezh_f16 (float16_t __a)
  53 {
  54   return __builtin_aarch64_cmgehf_uss (__a, 0.0f);
  55 }
  56
  57 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
  58 vcgtzh_f16 (float16_t __a)
  59 {
  60   return __builtin_aarch64_cmgthf_uss (__a, 0.0f);
  61 }
  62
  63 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
  64 vclezh_f16 (float16_t __a)
  65 {
  66   return __builtin_aarch64_cmlehf_uss (__a, 0.0f);
  67 }
  68
  69 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
  70 vcltzh_f16 (float16_t __a)
  71 {
  72   return __builtin_aarch64_cmlthf_uss (__a, 0.0f);
  73 }
  74
  75 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
  76 vcvth_f16_s16 (int16_t __a)
  77 {
  78   return __builtin_aarch64_floathihf (__a);
  79 }
  80
  81 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
  82 vcvth_f16_s32 (int32_t __a)
  83 {
  84   return __builtin_aarch64_floatsihf (__a);
  85 }
  86
  87 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
  88 vcvth_f16_s64 (int64_t __a)
  89 {
  90   return __builtin_aarch64_floatdihf (__a);
  91 }
  92
  93 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
  94 vcvth_f16_u16 (uint16_t __a)
  95 {
  96   return __builtin_aarch64_floatunshihf_us (__a);
  97 }
  98
  99 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 100 vcvth_f16_u32 (uint32_t __a)
 101 {
 102   return __builtin_aarch64_floatunssihf_us (__a);
 103 }
 104
 105 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 106 vcvth_f16_u64 (uint64_t __a)
 107 {
 108   return __builtin_aarch64_floatunsdihf_us (__a);
 109 }
 110
 111 __extension__ static __inline int16_t __attribute__ ((__always_inline__))
 112 vcvth_s16_f16 (float16_t __a)
 113 {
 114   return __builtin_aarch64_fix_trunchfhi (__a);
 115 }
 116
 117 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
 118 vcvth_s32_f16 (float16_t __a)
 119 {
 120   return __builtin_aarch64_fix_trunchfsi (__a);
 121 }
 122
 123 __extension__ static __inline int64_t __attribute__ ((__always_inline__))
 124 vcvth_s64_f16 (float16_t __a)
 125 {
 126   return __builtin_aarch64_fix_trunchfdi (__a);
 127 }
 128
 129 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 130 vcvth_u16_f16 (float16_t __a)
 131 {
 132   return __builtin_aarch64_fixuns_trunchfhi_us (__a);
 133 }
 134
 135 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
 136 vcvth_u32_f16 (float16_t __a)
 137 {
 138   return __builtin_aarch64_fixuns_trunchfsi_us (__a);
 139 }
 140
 141 __extension__ static __inline uint64_t __attribute__ ((__always_inline__))
 142 vcvth_u64_f16 (float16_t __a)
 143 {
 144   return __builtin_aarch64_fixuns_trunchfdi_us (__a);
 145 }
 146
 147 __extension__ static __inline int16_t __attribute__ ((__always_inline__))
 148 vcvtah_s16_f16 (float16_t __a)
 149 {
 150   return __builtin_aarch64_lroundhfhi (__a);
 151 }
 152
 153 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
 154 vcvtah_s32_f16 (float16_t __a)
 155 {
 156   return __builtin_aarch64_lroundhfsi (__a);
 157 }
 158
 159 __extension__ static __inline int64_t __attribute__ ((__always_inline__))
 160 vcvtah_s64_f16 (float16_t __a)
 161 {
 162   return __builtin_aarch64_lroundhfdi (__a);
 163 }
 164
 165 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 166 vcvtah_u16_f16 (float16_t __a)
 167 {
 168   return __builtin_aarch64_lrounduhfhi_us (__a);
 169 }
 170
 171 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
 172 vcvtah_u32_f16 (float16_t __a)
 173 {
 174   return __builtin_aarch64_lrounduhfsi_us (__a);
 175 }
 176
 177 __extension__ static __inline uint64_t __attribute__ ((__always_inline__))
 178 vcvtah_u64_f16 (float16_t __a)
 179 {
 180   return __builtin_aarch64_lrounduhfdi_us (__a);
 181 }
 182
 183 __extension__ static __inline int16_t __attribute__ ((__always_inline__))
 184 vcvtmh_s16_f16 (float16_t __a)
 185 {
 186   return __builtin_aarch64_lfloorhfhi (__a);
 187 }
 188
 189 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
 190 vcvtmh_s32_f16 (float16_t __a)
 191 {
 192   return __builtin_aarch64_lfloorhfsi (__a);
 193 }
 194
 195 __extension__ static __inline int64_t __attribute__ ((__always_inline__))
 196 vcvtmh_s64_f16 (float16_t __a)
 197 {
 198   return __builtin_aarch64_lfloorhfdi (__a);
 199 }
 200
 201 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 202 vcvtmh_u16_f16 (float16_t __a)
 203 {
 204   return __builtin_aarch64_lflooruhfhi_us (__a);
 205 }
 206
 207 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
 208 vcvtmh_u32_f16 (float16_t __a)
 209 {
 210   return __builtin_aarch64_lflooruhfsi_us (__a);
 211 }
 212
 213 __extension__ static __inline uint64_t __attribute__ ((__always_inline__))
 214 vcvtmh_u64_f16 (float16_t __a)
 215 {
 216   return __builtin_aarch64_lflooruhfdi_us (__a);
 217 }
 218
 219 __extension__ static __inline int16_t __attribute__ ((__always_inline__))
 220 vcvtnh_s16_f16 (float16_t __a)
 221 {
 222   return __builtin_aarch64_lfrintnhfhi (__a);
 223 }
 224
 225 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
 226 vcvtnh_s32_f16 (float16_t __a)
 227 {
 228   return __builtin_aarch64_lfrintnhfsi (__a);
 229 }
 230
 231 __extension__ static __inline int64_t __attribute__ ((__always_inline__))
 232 vcvtnh_s64_f16 (float16_t __a)
 233 {
 234   return __builtin_aarch64_lfrintnhfdi (__a);
 235 }
 236
 237 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 238 vcvtnh_u16_f16 (float16_t __a)
 239 {
 240   return __builtin_aarch64_lfrintnuhfhi_us (__a);
 241 }
 242
 243 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
 244 vcvtnh_u32_f16 (float16_t __a)
 245 {
 246   return __builtin_aarch64_lfrintnuhfsi_us (__a);
 247 }
 248
 249 __extension__ static __inline uint64_t __attribute__ ((__always_inline__))
 250 vcvtnh_u64_f16 (float16_t __a)
 251 {
 252   return __builtin_aarch64_lfrintnuhfdi_us (__a);
 253 }
 254
 255 __extension__ static __inline int16_t __attribute__ ((__always_inline__))
 256 vcvtph_s16_f16 (float16_t __a)
 257 {
 258   return __builtin_aarch64_lceilhfhi (__a);
 259 }
 260
 261 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
 262 vcvtph_s32_f16 (float16_t __a)
 263 {
 264   return __builtin_aarch64_lceilhfsi (__a);
 265 }
 266
 267 __extension__ static __inline int64_t __attribute__ ((__always_inline__))
 268 vcvtph_s64_f16 (float16_t __a)
 269 {
 270   return __builtin_aarch64_lceilhfdi (__a);
 271 }
 272
 273 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 274 vcvtph_u16_f16 (float16_t __a)
 275 {
 276   return __builtin_aarch64_lceiluhfhi_us (__a);
 277 }
 278
 279 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
 280 vcvtph_u32_f16 (float16_t __a)
 281 {
 282   return __builtin_aarch64_lceiluhfsi_us (__a);
 283 }
 284
 285 __extension__ static __inline uint64_t __attribute__ ((__always_inline__))
 286 vcvtph_u64_f16 (float16_t __a)
 287 {
 288   return __builtin_aarch64_lceiluhfdi_us (__a);
 289 }
 290
 291 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 292 vnegh_f16 (float16_t __a)
 293 {
 294   return __builtin_aarch64_neghf (__a);
 295 }
 296
 297 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 298 vrecpeh_f16 (float16_t __a)
 299 {
 300   return __builtin_aarch64_frecpehf (__a);
 301 }
 302
 303 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 304 vrecpxh_f16 (float16_t __a)
 305 {
 306   return __builtin_aarch64_frecpxhf (__a);
 307 }
 308
 309 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 310 vrndh_f16 (float16_t __a)
 311 {
 312   return __builtin_aarch64_btrunchf (__a);
 313 }
 314
 315 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 316 vrndah_f16 (float16_t __a)
 317 {
 318   return __builtin_aarch64_roundhf (__a);
 319 }
 320
 321 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 322 vrndih_f16 (float16_t __a)
 323 {
 324   return __builtin_aarch64_nearbyinthf (__a);
 325 }
 326
 327 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 328 vrndmh_f16 (float16_t __a)
 329 {
 330   return __builtin_aarch64_floorhf (__a);
 331 }
 332
 333 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 334 vrndnh_f16 (float16_t __a)
 335 {
 336   return __builtin_aarch64_frintnhf (__a);
 337 }
 338
 339 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 340 vrndph_f16 (float16_t __a)
 341 {
 342   return __builtin_aarch64_ceilhf (__a);
 343 }
 344
 345 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 346 vrndxh_f16 (float16_t __a)
 347 {
 348   return __builtin_aarch64_rinthf (__a);
 349 }
 350
 351 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 352 vrsqrteh_f16 (float16_t __a)
 353 {
 354   return __builtin_aarch64_rsqrtehf (__a);
 355 }
 356
 357 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 358 vsqrth_f16 (float16_t __a)
 359 {
 360   return __builtin_aarch64_sqrthf (__a);
 361 }
 362
 363 /* ARMv8.2-A FP16 two operands scalar intrinsics.  */
 364
 365 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 366 vaddh_f16 (float16_t __a, float16_t __b)
 367 {
 368   return __a + __b;
 369 }
 370
 371 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 372 vabdh_f16 (float16_t __a, float16_t __b)
 373 {
 374   return __builtin_aarch64_fabdhf (__a, __b);
 375 }
 376
 377 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 378 vcageh_f16 (float16_t __a, float16_t __b)
 379 {
 380   return __builtin_aarch64_facgehf_uss (__a, __b);
 381 }
 382
 383 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 384 vcagth_f16 (float16_t __a, float16_t __b)
 385 {
 386   return __builtin_aarch64_facgthf_uss (__a, __b);
 387 }
 388
 389 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 390 vcaleh_f16 (float16_t __a, float16_t __b)
 391 {
 392   return __builtin_aarch64_faclehf_uss (__a, __b);
 393 }
 394
 395 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 396 vcalth_f16 (float16_t __a, float16_t __b)
 397 {
 398   return __builtin_aarch64_faclthf_uss (__a, __b);
 399 }
 400
 401 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 402 vceqh_f16 (float16_t __a, float16_t __b)
 403 {
 404   return __builtin_aarch64_cmeqhf_uss (__a, __b);
 405 }
 406
 407 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 408 vcgeh_f16 (float16_t __a, float16_t __b)
 409 {
 410   return __builtin_aarch64_cmgehf_uss (__a, __b);
 411 }
 412
 413 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 414 vcgth_f16 (float16_t __a, float16_t __b)
 415 {
 416   return __builtin_aarch64_cmgthf_uss (__a, __b);
 417 }
 418
 419 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 420 vcleh_f16 (float16_t __a, float16_t __b)
 421 {
 422   return __builtin_aarch64_cmlehf_uss (__a, __b);
 423 }
 424
 425 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 426 vclth_f16 (float16_t __a, float16_t __b)
 427 {
 428   return __builtin_aarch64_cmlthf_uss (__a, __b);
 429 }
 430
 431 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 432 vcvth_n_f16_s16 (int16_t __a, const int __b)
 433 {
 434   return __builtin_aarch64_scvtfhi (__a, __b);
 435 }
 436
 437 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 438 vcvth_n_f16_s32 (int32_t __a, const int __b)
 439 {
 440   return __builtin_aarch64_scvtfsihf (__a, __b);
 441 }
 442
 443 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 444 vcvth_n_f16_s64 (int64_t __a, const int __b)
 445 {
 446   return __builtin_aarch64_scvtfdihf (__a, __b);
 447 }
 448
 449 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 450 vcvth_n_f16_u16 (uint16_t __a, const int __b)
 451 {
 452   return __builtin_aarch64_ucvtfhi_sus (__a, __b);
 453 }
 454
 455 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 456 vcvth_n_f16_u32 (uint32_t __a, const int __b)
 457 {
 458   return __builtin_aarch64_ucvtfsihf_sus (__a, __b);
 459 }
 460
 461 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 462 vcvth_n_f16_u64 (uint64_t __a, const int __b)
 463 {
 464   return __builtin_aarch64_ucvtfdihf_sus (__a, __b);
 465 }
 466
 467 __extension__ static __inline int16_t __attribute__ ((__always_inline__))
 468 vcvth_n_s16_f16 (float16_t __a, const int __b)
 469 {
 470   return __builtin_aarch64_fcvtzshf (__a, __b);
 471 }
 472
 473 __extension__ static __inline int32_t __attribute__ ((__always_inline__))
 474 vcvth_n_s32_f16 (float16_t __a, const int __b)
 475 {
 476   return __builtin_aarch64_fcvtzshfsi (__a, __b);
 477 }
 478
 479 __extension__ static __inline int64_t __attribute__ ((__always_inline__))
 480 vcvth_n_s64_f16 (float16_t __a, const int __b)
 481 {
 482   return __builtin_aarch64_fcvtzshfdi (__a, __b);
 483 }
 484
 485 __extension__ static __inline uint16_t __attribute__ ((__always_inline__))
 486 vcvth_n_u16_f16 (float16_t __a, const int __b)
 487 {
 488   return __builtin_aarch64_fcvtzuhf_uss (__a, __b);
 489 }
 490
 491 __extension__ static __inline uint32_t __attribute__ ((__always_inline__))
 492 vcvth_n_u32_f16 (float16_t __a, const int __b)
 493 {
 494   return __builtin_aarch64_fcvtzuhfsi_uss (__a, __b);
 495 }
 496
 497 __extension__ static __inline uint64_t __attribute__ ((__always_inline__))
 498 vcvth_n_u64_f16 (float16_t __a, const int __b)
 499 {
 500   return __builtin_aarch64_fcvtzuhfdi_uss (__a, __b);
 501 }
 502
 503 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 504 vdivh_f16 (float16_t __a, float16_t __b)
 505 {
 506   return __a / __b;
 507 }
 508
 509 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 510 vmaxh_f16 (float16_t __a, float16_t __b)
 511 {
 512   return __builtin_aarch64_fmaxhf (__a, __b);
 513 }
 514
 515 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 516 vmaxnmh_f16 (float16_t __a, float16_t __b)
 517 {
 518   return __builtin_aarch64_fmaxhf (__a, __b);
 519 }
 520
 521 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 522 vminh_f16 (float16_t __a, float16_t __b)
 523 {
 524   return __builtin_aarch64_fminhf (__a, __b);
 525 }
 526
 527 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 528 vminnmh_f16 (float16_t __a, float16_t __b)
 529 {
 530   return __builtin_aarch64_fminhf (__a, __b);
 531 }
 532
 533 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 534 vmulh_f16 (float16_t __a, float16_t __b)
 535 {
 536   return __a * __b;
 537 }
 538
 539 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 540 vmulxh_f16 (float16_t __a, float16_t __b)
 541 {
 542   return __builtin_aarch64_fmulxhf (__a, __b);
 543 }
 544
 545 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 546 vrecpsh_f16 (float16_t __a, float16_t __b)
 547 {
 548   return __builtin_aarch64_frecpshf (__a, __b);
 549 }
 550
 551 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 552 vrsqrtsh_f16 (float16_t __a, float16_t __b)
 553 {
 554   return __builtin_aarch64_rsqrtshf (__a, __b);
 555 }
 556
 557 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 558 vsubh_f16 (float16_t __a, float16_t __b)
 559 {
 560   return __a - __b;
 561 }
 562
 563 /* ARMv8.2-A FP16 three operands scalar intrinsics.  */
 564
 565 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 566 vfmah_f16 (float16_t __a, float16_t __b, float16_t __c)
 567 {
 568   return __builtin_aarch64_fmahf (__b, __c, __a);
 569 }
 570
 571 __extension__ static __inline float16_t __attribute__ ((__always_inline__))
 572 vfmsh_f16 (float16_t __a, float16_t __b, float16_t __c)
 573 {
 574   return __builtin_aarch64_fnmahf (__b, __c, __a);
 575 }
 576
 577 #pragma GCC pop_options
 578
 579 #endif