1 /* 2 * QEMU float support 3 * 4 * The code in this source file is derived from release 2a of the SoftFloat 5 * IEC/IEEE Floating-point Arithmetic Package. Those parts of the code (and 6 * some later contributions) are provided under that license, as detailed below. 7 * It has subsequently been modified by contributors to the QEMU Project, 8 * so some portions are provided under: 9 * the SoftFloat-2a license 10 * the BSD license 11 * GPL-v2-or-later 12 * 13 * Any future contributions to this file after December 1st 2014 will be 14 * taken to be licensed under the Softfloat-2a license unless specifically 15 * indicated otherwise. 16 */ 17 18 /* 19 =============================================================================== 20 This C header file is part of the SoftFloat IEC/IEEE Floating-point 21 Arithmetic Package, Release 2a. 22 23 Written by John R. Hauser. This work was made possible in part by the 24 International Computer Science Institute, located at Suite 600, 1947 Center 25 Street, Berkeley, California 94704. Funding was partially provided by the 26 National Science Foundation under grant MIP-9311980. The original version 27 of this code was written as part of a project to build a fixed-point vector 28 processor in collaboration with the University of California at Berkeley, 29 overseen by Profs. Nelson Morgan and John Wawrzynek. More information 30 is available through the Web page `http://HTTP.CS.Berkeley.EDU/~jhauser/ 31 arithmetic/SoftFloat.html'. 32 33 THIS SOFTWARE IS DISTRIBUTED AS IS, FOR FREE. Although reasonable effort 34 has been made to avoid it, THIS SOFTWARE MAY CONTAIN FAULTS THAT WILL AT 35 TIMES RESULT IN INCORRECT BEHAVIOR. USE OF THIS SOFTWARE IS RESTRICTED TO 36 PERSONS AND ORGANIZATIONS WHO CAN AND WILL TAKE FULL RESPONSIBILITY FOR ANY 37 AND ALL LOSSES, COSTS, OR OTHER PROBLEMS ARISING FROM ITS USE. 38 39 Derivative works are acceptable, even for commercial purposes, so long as 40 (1) they include prominent notice that the work is derivative, and (2) they 41 include prominent notice akin to these four paragraphs for those parts of 42 this code that are retained. 43 44 =============================================================================== 45 */ 46 47 /* BSD licensing: 48 * Copyright (c) 2006, Fabrice Bellard 49 * All rights reserved. 50 * 51 * Redistribution and use in source and binary forms, with or without 52 * modification, are permitted provided that the following conditions are met: 53 * 54 * 1. Redistributions of source code must retain the above copyright notice, 55 * this list of conditions and the following disclaimer. 56 * 57 * 2. Redistributions in binary form must reproduce the above copyright notice, 58 * this list of conditions and the following disclaimer in the documentation 59 * and/or other materials provided with the distribution. 60 * 61 * 3. Neither the name of the copyright holder nor the names of its contributors 62 * may be used to endorse or promote products derived from this software without 63 * specific prior written permission. 64 * 65 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" 66 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 67 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 68 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE 69 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 70 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 71 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 72 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 73 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 74 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 75 * THE POSSIBILITY OF SUCH DAMAGE. 76 */ 77 78 /* Portions of this work are licensed under the terms of the GNU GPL, 79 * version 2 or later. See the COPYING file in the top-level directory. 80 */ 81 82 #ifndef SOFTFLOAT_H 83 #define SOFTFLOAT_H 84 85 /*---------------------------------------------------------------------------- 86 | Software IEC/IEEE floating-point ordering relations 87 *----------------------------------------------------------------------------*/ 88 89 typedef enum { 90 float_relation_less = -1, 91 float_relation_equal = 0, 92 float_relation_greater = 1, 93 float_relation_unordered = 2 94 } FloatRelation; 95 96 #include "fpu/softfloat-types.h" 97 #include "fpu/softfloat-helpers.h" 98 99 /*---------------------------------------------------------------------------- 100 | Routine to raise any or all of the software IEC/IEEE floating-point 101 | exception flags. 102 *----------------------------------------------------------------------------*/ 103 void float_raise(uint8_t flags, float_status *status); 104 105 /*---------------------------------------------------------------------------- 106 | If `a' is denormal and we are in flush-to-zero mode then set the 107 | input-denormal exception and return zero. Otherwise just return the value. 108 *----------------------------------------------------------------------------*/ 109 float16 float16_squash_input_denormal(float16 a, float_status *status); 110 float32 float32_squash_input_denormal(float32 a, float_status *status); 111 float64 float64_squash_input_denormal(float64 a, float_status *status); 112 113 /*---------------------------------------------------------------------------- 114 | Options to indicate which negations to perform in float*_muladd() 115 | Using these differs from negating an input or output before calling 116 | the muladd function in that this means that a NaN doesn't have its 117 | sign bit inverted before it is propagated. 118 | We also support halving the result before rounding, as a special 119 | case to support the ARM fused-sqrt-step instruction FRSQRTS. 120 *----------------------------------------------------------------------------*/ 121 enum { 122 float_muladd_negate_c = 1, 123 float_muladd_negate_product = 2, 124 float_muladd_negate_result = 4, 125 float_muladd_halve_result = 8, 126 }; 127 128 /*---------------------------------------------------------------------------- 129 | Software IEC/IEEE integer-to-floating-point conversion routines. 130 *----------------------------------------------------------------------------*/ 131 132 float16 int16_to_float16_scalbn(int16_t a, int, float_status *status); 133 float16 int32_to_float16_scalbn(int32_t a, int, float_status *status); 134 float16 int64_to_float16_scalbn(int64_t a, int, float_status *status); 135 float16 uint16_to_float16_scalbn(uint16_t a, int, float_status *status); 136 float16 uint32_to_float16_scalbn(uint32_t a, int, float_status *status); 137 float16 uint64_to_float16_scalbn(uint64_t a, int, float_status *status); 138 139 float16 int16_to_float16(int16_t a, float_status *status); 140 float16 int32_to_float16(int32_t a, float_status *status); 141 float16 int64_to_float16(int64_t a, float_status *status); 142 float16 uint16_to_float16(uint16_t a, float_status *status); 143 float16 uint32_to_float16(uint32_t a, float_status *status); 144 float16 uint64_to_float16(uint64_t a, float_status *status); 145 146 float32 int16_to_float32_scalbn(int16_t, int, float_status *status); 147 float32 int32_to_float32_scalbn(int32_t, int, float_status *status); 148 float32 int64_to_float32_scalbn(int64_t, int, float_status *status); 149 float32 uint16_to_float32_scalbn(uint16_t, int, float_status *status); 150 float32 uint32_to_float32_scalbn(uint32_t, int, float_status *status); 151 float32 uint64_to_float32_scalbn(uint64_t, int, float_status *status); 152 153 float32 int16_to_float32(int16_t, float_status *status); 154 float32 int32_to_float32(int32_t, float_status *status); 155 float32 int64_to_float32(int64_t, float_status *status); 156 float32 uint16_to_float32(uint16_t, float_status *status); 157 float32 uint32_to_float32(uint32_t, float_status *status); 158 float32 uint64_to_float32(uint64_t, float_status *status); 159 160 float64 int16_to_float64_scalbn(int16_t, int, float_status *status); 161 float64 int32_to_float64_scalbn(int32_t, int, float_status *status); 162 float64 int64_to_float64_scalbn(int64_t, int, float_status *status); 163 float64 uint16_to_float64_scalbn(uint16_t, int, float_status *status); 164 float64 uint32_to_float64_scalbn(uint32_t, int, float_status *status); 165 float64 uint64_to_float64_scalbn(uint64_t, int, float_status *status); 166 167 float64 int16_to_float64(int16_t, float_status *status); 168 float64 int32_to_float64(int32_t, float_status *status); 169 float64 int64_to_float64(int64_t, float_status *status); 170 float64 uint16_to_float64(uint16_t, float_status *status); 171 float64 uint32_to_float64(uint32_t, float_status *status); 172 float64 uint64_to_float64(uint64_t, float_status *status); 173 174 floatx80 int32_to_floatx80(int32_t, float_status *status); 175 floatx80 int64_to_floatx80(int64_t, float_status *status); 176 177 float128 int32_to_float128(int32_t, float_status *status); 178 float128 int64_to_float128(int64_t, float_status *status); 179 float128 uint64_to_float128(uint64_t, float_status *status); 180 181 /*---------------------------------------------------------------------------- 182 | Software half-precision conversion routines. 183 *----------------------------------------------------------------------------*/ 184 185 float16 float32_to_float16(float32, bool ieee, float_status *status); 186 float32 float16_to_float32(float16, bool ieee, float_status *status); 187 float16 float64_to_float16(float64 a, bool ieee, float_status *status); 188 float64 float16_to_float64(float16 a, bool ieee, float_status *status); 189 190 int16_t float16_to_int16_scalbn(float16, FloatRoundMode, int, float_status *); 191 int32_t float16_to_int32_scalbn(float16, FloatRoundMode, int, float_status *); 192 int64_t float16_to_int64_scalbn(float16, FloatRoundMode, int, float_status *); 193 194 int16_t float16_to_int16(float16, float_status *status); 195 int32_t float16_to_int32(float16, float_status *status); 196 int64_t float16_to_int64(float16, float_status *status); 197 198 int16_t float16_to_int16_round_to_zero(float16, float_status *status); 199 int32_t float16_to_int32_round_to_zero(float16, float_status *status); 200 int64_t float16_to_int64_round_to_zero(float16, float_status *status); 201 202 uint16_t float16_to_uint16_scalbn(float16 a, FloatRoundMode, 203 int, float_status *status); 204 uint32_t float16_to_uint32_scalbn(float16 a, FloatRoundMode, 205 int, float_status *status); 206 uint64_t float16_to_uint64_scalbn(float16 a, FloatRoundMode, 207 int, float_status *status); 208 209 uint16_t float16_to_uint16(float16 a, float_status *status); 210 uint32_t float16_to_uint32(float16 a, float_status *status); 211 uint64_t float16_to_uint64(float16 a, float_status *status); 212 213 uint16_t float16_to_uint16_round_to_zero(float16 a, float_status *status); 214 uint32_t float16_to_uint32_round_to_zero(float16 a, float_status *status); 215 uint64_t float16_to_uint64_round_to_zero(float16 a, float_status *status); 216 217 /*---------------------------------------------------------------------------- 218 | Software half-precision operations. 219 *----------------------------------------------------------------------------*/ 220 221 float16 float16_round_to_int(float16, float_status *status); 222 float16 float16_add(float16, float16, float_status *status); 223 float16 float16_sub(float16, float16, float_status *status); 224 float16 float16_mul(float16, float16, float_status *status); 225 float16 float16_muladd(float16, float16, float16, int, float_status *status); 226 float16 float16_div(float16, float16, float_status *status); 227 float16 float16_scalbn(float16, int, float_status *status); 228 float16 float16_min(float16, float16, float_status *status); 229 float16 float16_max(float16, float16, float_status *status); 230 float16 float16_minnum(float16, float16, float_status *status); 231 float16 float16_maxnum(float16, float16, float_status *status); 232 float16 float16_minnummag(float16, float16, float_status *status); 233 float16 float16_maxnummag(float16, float16, float_status *status); 234 float16 float16_sqrt(float16, float_status *status); 235 FloatRelation float16_compare(float16, float16, float_status *status); 236 FloatRelation float16_compare_quiet(float16, float16, float_status *status); 237 238 bool float16_is_quiet_nan(float16, float_status *status); 239 bool float16_is_signaling_nan(float16, float_status *status); 240 float16 float16_silence_nan(float16, float_status *status); 241 242 static inline bool float16_is_any_nan(float16 a) 243 { 244 return ((float16_val(a) & ~0x8000) > 0x7c00); 245 } 246 247 static inline bool float16_is_neg(float16 a) 248 { 249 return float16_val(a) >> 15; 250 } 251 252 static inline bool float16_is_infinity(float16 a) 253 { 254 return (float16_val(a) & 0x7fff) == 0x7c00; 255 } 256 257 static inline bool float16_is_zero(float16 a) 258 { 259 return (float16_val(a) & 0x7fff) == 0; 260 } 261 262 static inline bool float16_is_zero_or_denormal(float16 a) 263 { 264 return (float16_val(a) & 0x7c00) == 0; 265 } 266 267 static inline float16 float16_abs(float16 a) 268 { 269 /* Note that abs does *not* handle NaN specially, nor does 270 * it flush denormal inputs to zero. 271 */ 272 return make_float16(float16_val(a) & 0x7fff); 273 } 274 275 static inline float16 float16_chs(float16 a) 276 { 277 /* Note that chs does *not* handle NaN specially, nor does 278 * it flush denormal inputs to zero. 279 */ 280 return make_float16(float16_val(a) ^ 0x8000); 281 } 282 283 static inline float16 float16_set_sign(float16 a, int sign) 284 { 285 return make_float16((float16_val(a) & 0x7fff) | (sign << 15)); 286 } 287 288 static inline bool float16_eq(float16 a, float16 b, float_status *s) 289 { 290 return float16_compare(a, b, s) == float_relation_equal; 291 } 292 293 static inline bool float16_le(float16 a, float16 b, float_status *s) 294 { 295 return float16_compare(a, b, s) <= float_relation_equal; 296 } 297 298 static inline bool float16_lt(float16 a, float16 b, float_status *s) 299 { 300 return float16_compare(a, b, s) < float_relation_equal; 301 } 302 303 static inline bool float16_unordered(float16 a, float16 b, float_status *s) 304 { 305 return float16_compare(a, b, s) == float_relation_unordered; 306 } 307 308 static inline bool float16_eq_quiet(float16 a, float16 b, float_status *s) 309 { 310 return float16_compare_quiet(a, b, s) == float_relation_equal; 311 } 312 313 static inline bool float16_le_quiet(float16 a, float16 b, float_status *s) 314 { 315 return float16_compare_quiet(a, b, s) <= float_relation_equal; 316 } 317 318 static inline bool float16_lt_quiet(float16 a, float16 b, float_status *s) 319 { 320 return float16_compare_quiet(a, b, s) < float_relation_equal; 321 } 322 323 static inline bool float16_unordered_quiet(float16 a, float16 b, 324 float_status *s) 325 { 326 return float16_compare_quiet(a, b, s) == float_relation_unordered; 327 } 328 329 #define float16_zero make_float16(0) 330 #define float16_half make_float16(0x3800) 331 #define float16_one make_float16(0x3c00) 332 #define float16_one_point_five make_float16(0x3e00) 333 #define float16_two make_float16(0x4000) 334 #define float16_three make_float16(0x4200) 335 #define float16_infinity make_float16(0x7c00) 336 337 /*---------------------------------------------------------------------------- 338 | The pattern for a default generated half-precision NaN. 339 *----------------------------------------------------------------------------*/ 340 float16 float16_default_nan(float_status *status); 341 342 /*---------------------------------------------------------------------------- 343 | Software IEC/IEEE single-precision conversion routines. 344 *----------------------------------------------------------------------------*/ 345 346 int16_t float32_to_int16_scalbn(float32, FloatRoundMode, int, float_status *); 347 int32_t float32_to_int32_scalbn(float32, FloatRoundMode, int, float_status *); 348 int64_t float32_to_int64_scalbn(float32, FloatRoundMode, int, float_status *); 349 350 int16_t float32_to_int16(float32, float_status *status); 351 int32_t float32_to_int32(float32, float_status *status); 352 int64_t float32_to_int64(float32, float_status *status); 353 354 int16_t float32_to_int16_round_to_zero(float32, float_status *status); 355 int32_t float32_to_int32_round_to_zero(float32, float_status *status); 356 int64_t float32_to_int64_round_to_zero(float32, float_status *status); 357 358 uint16_t float32_to_uint16_scalbn(float32, FloatRoundMode, int, float_status *); 359 uint32_t float32_to_uint32_scalbn(float32, FloatRoundMode, int, float_status *); 360 uint64_t float32_to_uint64_scalbn(float32, FloatRoundMode, int, float_status *); 361 362 uint16_t float32_to_uint16(float32, float_status *status); 363 uint32_t float32_to_uint32(float32, float_status *status); 364 uint64_t float32_to_uint64(float32, float_status *status); 365 366 uint16_t float32_to_uint16_round_to_zero(float32, float_status *status); 367 uint32_t float32_to_uint32_round_to_zero(float32, float_status *status); 368 uint64_t float32_to_uint64_round_to_zero(float32, float_status *status); 369 370 float64 float32_to_float64(float32, float_status *status); 371 floatx80 float32_to_floatx80(float32, float_status *status); 372 float128 float32_to_float128(float32, float_status *status); 373 374 /*---------------------------------------------------------------------------- 375 | Software IEC/IEEE single-precision operations. 376 *----------------------------------------------------------------------------*/ 377 float32 float32_round_to_int(float32, float_status *status); 378 float32 float32_add(float32, float32, float_status *status); 379 float32 float32_sub(float32, float32, float_status *status); 380 float32 float32_mul(float32, float32, float_status *status); 381 float32 float32_div(float32, float32, float_status *status); 382 float32 float32_rem(float32, float32, float_status *status); 383 float32 float32_muladd(float32, float32, float32, int, float_status *status); 384 float32 float32_sqrt(float32, float_status *status); 385 float32 float32_exp2(float32, float_status *status); 386 float32 float32_log2(float32, float_status *status); 387 FloatRelation float32_compare(float32, float32, float_status *status); 388 FloatRelation float32_compare_quiet(float32, float32, float_status *status); 389 float32 float32_min(float32, float32, float_status *status); 390 float32 float32_max(float32, float32, float_status *status); 391 float32 float32_minnum(float32, float32, float_status *status); 392 float32 float32_maxnum(float32, float32, float_status *status); 393 float32 float32_minnummag(float32, float32, float_status *status); 394 float32 float32_maxnummag(float32, float32, float_status *status); 395 bool float32_is_quiet_nan(float32, float_status *status); 396 bool float32_is_signaling_nan(float32, float_status *status); 397 float32 float32_silence_nan(float32, float_status *status); 398 float32 float32_scalbn(float32, int, float_status *status); 399 400 static inline float32 float32_abs(float32 a) 401 { 402 /* Note that abs does *not* handle NaN specially, nor does 403 * it flush denormal inputs to zero. 404 */ 405 return make_float32(float32_val(a) & 0x7fffffff); 406 } 407 408 static inline float32 float32_chs(float32 a) 409 { 410 /* Note that chs does *not* handle NaN specially, nor does 411 * it flush denormal inputs to zero. 412 */ 413 return make_float32(float32_val(a) ^ 0x80000000); 414 } 415 416 static inline bool float32_is_infinity(float32 a) 417 { 418 return (float32_val(a) & 0x7fffffff) == 0x7f800000; 419 } 420 421 static inline bool float32_is_neg(float32 a) 422 { 423 return float32_val(a) >> 31; 424 } 425 426 static inline bool float32_is_zero(float32 a) 427 { 428 return (float32_val(a) & 0x7fffffff) == 0; 429 } 430 431 static inline bool float32_is_any_nan(float32 a) 432 { 433 return ((float32_val(a) & ~(1 << 31)) > 0x7f800000UL); 434 } 435 436 static inline bool float32_is_zero_or_denormal(float32 a) 437 { 438 return (float32_val(a) & 0x7f800000) == 0; 439 } 440 441 static inline bool float32_is_normal(float32 a) 442 { 443 return (((float32_val(a) >> 23) + 1) & 0xff) >= 2; 444 } 445 446 static inline bool float32_is_denormal(float32 a) 447 { 448 return float32_is_zero_or_denormal(a) && !float32_is_zero(a); 449 } 450 451 static inline bool float32_is_zero_or_normal(float32 a) 452 { 453 return float32_is_normal(a) || float32_is_zero(a); 454 } 455 456 static inline float32 float32_set_sign(float32 a, int sign) 457 { 458 return make_float32((float32_val(a) & 0x7fffffff) | (sign << 31)); 459 } 460 461 static inline bool float32_eq(float32 a, float32 b, float_status *s) 462 { 463 return float32_compare(a, b, s) == float_relation_equal; 464 } 465 466 static inline bool float32_le(float32 a, float32 b, float_status *s) 467 { 468 return float32_compare(a, b, s) <= float_relation_equal; 469 } 470 471 static inline bool float32_lt(float32 a, float32 b, float_status *s) 472 { 473 return float32_compare(a, b, s) < float_relation_equal; 474 } 475 476 static inline bool float32_unordered(float32 a, float32 b, float_status *s) 477 { 478 return float32_compare(a, b, s) == float_relation_unordered; 479 } 480 481 static inline bool float32_eq_quiet(float32 a, float32 b, float_status *s) 482 { 483 return float32_compare_quiet(a, b, s) == float_relation_equal; 484 } 485 486 static inline bool float32_le_quiet(float32 a, float32 b, float_status *s) 487 { 488 return float32_compare_quiet(a, b, s) <= float_relation_equal; 489 } 490 491 static inline bool float32_lt_quiet(float32 a, float32 b, float_status *s) 492 { 493 return float32_compare_quiet(a, b, s) < float_relation_equal; 494 } 495 496 static inline bool float32_unordered_quiet(float32 a, float32 b, 497 float_status *s) 498 { 499 return float32_compare_quiet(a, b, s) == float_relation_unordered; 500 } 501 502 #define float32_zero make_float32(0) 503 #define float32_half make_float32(0x3f000000) 504 #define float32_one make_float32(0x3f800000) 505 #define float32_one_point_five make_float32(0x3fc00000) 506 #define float32_two make_float32(0x40000000) 507 #define float32_three make_float32(0x40400000) 508 #define float32_infinity make_float32(0x7f800000) 509 510 /*---------------------------------------------------------------------------- 511 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into a 512 | single-precision floating-point value, returning the result. After being 513 | shifted into the proper positions, the three fields are simply added 514 | together to form the result. This means that any integer portion of `zSig' 515 | will be added into the exponent. Since a properly normalized significand 516 | will have an integer portion equal to 1, the `zExp' input should be 1 less 517 | than the desired result exponent whenever `zSig' is a complete, normalized 518 | significand. 519 *----------------------------------------------------------------------------*/ 520 521 static inline float32 packFloat32(bool zSign, int zExp, uint32_t zSig) 522 { 523 return make_float32( 524 (((uint32_t)zSign) << 31) + (((uint32_t)zExp) << 23) + zSig); 525 } 526 527 /*---------------------------------------------------------------------------- 528 | The pattern for a default generated single-precision NaN. 529 *----------------------------------------------------------------------------*/ 530 float32 float32_default_nan(float_status *status); 531 532 /*---------------------------------------------------------------------------- 533 | Software IEC/IEEE double-precision conversion routines. 534 *----------------------------------------------------------------------------*/ 535 536 int16_t float64_to_int16_scalbn(float64, FloatRoundMode, int, float_status *); 537 int32_t float64_to_int32_scalbn(float64, FloatRoundMode, int, float_status *); 538 int64_t float64_to_int64_scalbn(float64, FloatRoundMode, int, float_status *); 539 540 int16_t float64_to_int16(float64, float_status *status); 541 int32_t float64_to_int32(float64, float_status *status); 542 int64_t float64_to_int64(float64, float_status *status); 543 544 int16_t float64_to_int16_round_to_zero(float64, float_status *status); 545 int32_t float64_to_int32_round_to_zero(float64, float_status *status); 546 int64_t float64_to_int64_round_to_zero(float64, float_status *status); 547 548 uint16_t float64_to_uint16_scalbn(float64, FloatRoundMode, int, float_status *); 549 uint32_t float64_to_uint32_scalbn(float64, FloatRoundMode, int, float_status *); 550 uint64_t float64_to_uint64_scalbn(float64, FloatRoundMode, int, float_status *); 551 552 uint16_t float64_to_uint16(float64, float_status *status); 553 uint32_t float64_to_uint32(float64, float_status *status); 554 uint64_t float64_to_uint64(float64, float_status *status); 555 556 uint16_t float64_to_uint16_round_to_zero(float64, float_status *status); 557 uint32_t float64_to_uint32_round_to_zero(float64, float_status *status); 558 uint64_t float64_to_uint64_round_to_zero(float64, float_status *status); 559 560 float32 float64_to_float32(float64, float_status *status); 561 floatx80 float64_to_floatx80(float64, float_status *status); 562 float128 float64_to_float128(float64, float_status *status); 563 564 /*---------------------------------------------------------------------------- 565 | Software IEC/IEEE double-precision operations. 566 *----------------------------------------------------------------------------*/ 567 float64 float64_round_to_int(float64, float_status *status); 568 float64 float64_add(float64, float64, float_status *status); 569 float64 float64_sub(float64, float64, float_status *status); 570 float64 float64_mul(float64, float64, float_status *status); 571 float64 float64_div(float64, float64, float_status *status); 572 float64 float64_rem(float64, float64, float_status *status); 573 float64 float64_muladd(float64, float64, float64, int, float_status *status); 574 float64 float64_sqrt(float64, float_status *status); 575 float64 float64_log2(float64, float_status *status); 576 FloatRelation float64_compare(float64, float64, float_status *status); 577 FloatRelation float64_compare_quiet(float64, float64, float_status *status); 578 float64 float64_min(float64, float64, float_status *status); 579 float64 float64_max(float64, float64, float_status *status); 580 float64 float64_minnum(float64, float64, float_status *status); 581 float64 float64_maxnum(float64, float64, float_status *status); 582 float64 float64_minnummag(float64, float64, float_status *status); 583 float64 float64_maxnummag(float64, float64, float_status *status); 584 bool float64_is_quiet_nan(float64 a, float_status *status); 585 bool float64_is_signaling_nan(float64, float_status *status); 586 float64 float64_silence_nan(float64, float_status *status); 587 float64 float64_scalbn(float64, int, float_status *status); 588 589 static inline float64 float64_abs(float64 a) 590 { 591 /* Note that abs does *not* handle NaN specially, nor does 592 * it flush denormal inputs to zero. 593 */ 594 return make_float64(float64_val(a) & 0x7fffffffffffffffLL); 595 } 596 597 static inline float64 float64_chs(float64 a) 598 { 599 /* Note that chs does *not* handle NaN specially, nor does 600 * it flush denormal inputs to zero. 601 */ 602 return make_float64(float64_val(a) ^ 0x8000000000000000LL); 603 } 604 605 static inline bool float64_is_infinity(float64 a) 606 { 607 return (float64_val(a) & 0x7fffffffffffffffLL ) == 0x7ff0000000000000LL; 608 } 609 610 static inline bool float64_is_neg(float64 a) 611 { 612 return float64_val(a) >> 63; 613 } 614 615 static inline bool float64_is_zero(float64 a) 616 { 617 return (float64_val(a) & 0x7fffffffffffffffLL) == 0; 618 } 619 620 static inline bool float64_is_any_nan(float64 a) 621 { 622 return ((float64_val(a) & ~(1ULL << 63)) > 0x7ff0000000000000ULL); 623 } 624 625 static inline bool float64_is_zero_or_denormal(float64 a) 626 { 627 return (float64_val(a) & 0x7ff0000000000000LL) == 0; 628 } 629 630 static inline bool float64_is_normal(float64 a) 631 { 632 return (((float64_val(a) >> 52) + 1) & 0x7ff) >= 2; 633 } 634 635 static inline bool float64_is_denormal(float64 a) 636 { 637 return float64_is_zero_or_denormal(a) && !float64_is_zero(a); 638 } 639 640 static inline bool float64_is_zero_or_normal(float64 a) 641 { 642 return float64_is_normal(a) || float64_is_zero(a); 643 } 644 645 static inline float64 float64_set_sign(float64 a, int sign) 646 { 647 return make_float64((float64_val(a) & 0x7fffffffffffffffULL) 648 | ((int64_t)sign << 63)); 649 } 650 651 static inline bool float64_eq(float64 a, float64 b, float_status *s) 652 { 653 return float64_compare(a, b, s) == float_relation_equal; 654 } 655 656 static inline bool float64_le(float64 a, float64 b, float_status *s) 657 { 658 return float64_compare(a, b, s) <= float_relation_equal; 659 } 660 661 static inline bool float64_lt(float64 a, float64 b, float_status *s) 662 { 663 return float64_compare(a, b, s) < float_relation_equal; 664 } 665 666 static inline bool float64_unordered(float64 a, float64 b, float_status *s) 667 { 668 return float64_compare(a, b, s) == float_relation_unordered; 669 } 670 671 static inline bool float64_eq_quiet(float64 a, float64 b, float_status *s) 672 { 673 return float64_compare_quiet(a, b, s) == float_relation_equal; 674 } 675 676 static inline bool float64_le_quiet(float64 a, float64 b, float_status *s) 677 { 678 return float64_compare_quiet(a, b, s) <= float_relation_equal; 679 } 680 681 static inline bool float64_lt_quiet(float64 a, float64 b, float_status *s) 682 { 683 return float64_compare_quiet(a, b, s) < float_relation_equal; 684 } 685 686 static inline bool float64_unordered_quiet(float64 a, float64 b, 687 float_status *s) 688 { 689 return float64_compare_quiet(a, b, s) == float_relation_unordered; 690 } 691 692 #define float64_zero make_float64(0) 693 #define float64_half make_float64(0x3fe0000000000000LL) 694 #define float64_one make_float64(0x3ff0000000000000LL) 695 #define float64_one_point_five make_float64(0x3FF8000000000000ULL) 696 #define float64_two make_float64(0x4000000000000000ULL) 697 #define float64_three make_float64(0x4008000000000000ULL) 698 #define float64_ln2 make_float64(0x3fe62e42fefa39efLL) 699 #define float64_infinity make_float64(0x7ff0000000000000LL) 700 701 /*---------------------------------------------------------------------------- 702 | The pattern for a default generated double-precision NaN. 703 *----------------------------------------------------------------------------*/ 704 float64 float64_default_nan(float_status *status); 705 706 /*---------------------------------------------------------------------------- 707 | Software IEC/IEEE extended double-precision conversion routines. 708 *----------------------------------------------------------------------------*/ 709 int32_t floatx80_to_int32(floatx80, float_status *status); 710 int32_t floatx80_to_int32_round_to_zero(floatx80, float_status *status); 711 int64_t floatx80_to_int64(floatx80, float_status *status); 712 int64_t floatx80_to_int64_round_to_zero(floatx80, float_status *status); 713 float32 floatx80_to_float32(floatx80, float_status *status); 714 float64 floatx80_to_float64(floatx80, float_status *status); 715 float128 floatx80_to_float128(floatx80, float_status *status); 716 717 /*---------------------------------------------------------------------------- 718 | The pattern for an extended double-precision inf. 719 *----------------------------------------------------------------------------*/ 720 extern const floatx80 floatx80_infinity; 721 722 /*---------------------------------------------------------------------------- 723 | Software IEC/IEEE extended double-precision operations. 724 *----------------------------------------------------------------------------*/ 725 floatx80 floatx80_round(floatx80 a, float_status *status); 726 floatx80 floatx80_round_to_int(floatx80, float_status *status); 727 floatx80 floatx80_add(floatx80, floatx80, float_status *status); 728 floatx80 floatx80_sub(floatx80, floatx80, float_status *status); 729 floatx80 floatx80_mul(floatx80, floatx80, float_status *status); 730 floatx80 floatx80_div(floatx80, floatx80, float_status *status); 731 floatx80 floatx80_modrem(floatx80, floatx80, bool, uint64_t *, 732 float_status *status); 733 floatx80 floatx80_mod(floatx80, floatx80, float_status *status); 734 floatx80 floatx80_rem(floatx80, floatx80, float_status *status); 735 floatx80 floatx80_sqrt(floatx80, float_status *status); 736 FloatRelation floatx80_compare(floatx80, floatx80, float_status *status); 737 FloatRelation floatx80_compare_quiet(floatx80, floatx80, float_status *status); 738 int floatx80_is_quiet_nan(floatx80, float_status *status); 739 int floatx80_is_signaling_nan(floatx80, float_status *status); 740 floatx80 floatx80_silence_nan(floatx80, float_status *status); 741 floatx80 floatx80_scalbn(floatx80, int, float_status *status); 742 743 static inline floatx80 floatx80_abs(floatx80 a) 744 { 745 a.high &= 0x7fff; 746 return a; 747 } 748 749 static inline floatx80 floatx80_chs(floatx80 a) 750 { 751 a.high ^= 0x8000; 752 return a; 753 } 754 755 static inline bool floatx80_is_infinity(floatx80 a) 756 { 757 #if defined(TARGET_M68K) 758 return (a.high & 0x7fff) == floatx80_infinity.high && !(a.low << 1); 759 #else 760 return (a.high & 0x7fff) == floatx80_infinity.high && 761 a.low == floatx80_infinity.low; 762 #endif 763 } 764 765 static inline bool floatx80_is_neg(floatx80 a) 766 { 767 return a.high >> 15; 768 } 769 770 static inline bool floatx80_is_zero(floatx80 a) 771 { 772 return (a.high & 0x7fff) == 0 && a.low == 0; 773 } 774 775 static inline bool floatx80_is_zero_or_denormal(floatx80 a) 776 { 777 return (a.high & 0x7fff) == 0; 778 } 779 780 static inline bool floatx80_is_any_nan(floatx80 a) 781 { 782 return ((a.high & 0x7fff) == 0x7fff) && (a.low<<1); 783 } 784 785 static inline bool floatx80_eq(floatx80 a, floatx80 b, float_status *s) 786 { 787 return floatx80_compare(a, b, s) == float_relation_equal; 788 } 789 790 static inline bool floatx80_le(floatx80 a, floatx80 b, float_status *s) 791 { 792 return floatx80_compare(a, b, s) <= float_relation_equal; 793 } 794 795 static inline bool floatx80_lt(floatx80 a, floatx80 b, float_status *s) 796 { 797 return floatx80_compare(a, b, s) < float_relation_equal; 798 } 799 800 static inline bool floatx80_unordered(floatx80 a, floatx80 b, float_status *s) 801 { 802 return floatx80_compare(a, b, s) == float_relation_unordered; 803 } 804 805 static inline bool floatx80_eq_quiet(floatx80 a, floatx80 b, float_status *s) 806 { 807 return floatx80_compare_quiet(a, b, s) == float_relation_equal; 808 } 809 810 static inline bool floatx80_le_quiet(floatx80 a, floatx80 b, float_status *s) 811 { 812 return floatx80_compare_quiet(a, b, s) <= float_relation_equal; 813 } 814 815 static inline bool floatx80_lt_quiet(floatx80 a, floatx80 b, float_status *s) 816 { 817 return floatx80_compare_quiet(a, b, s) < float_relation_equal; 818 } 819 820 static inline bool floatx80_unordered_quiet(floatx80 a, floatx80 b, 821 float_status *s) 822 { 823 return floatx80_compare_quiet(a, b, s) == float_relation_unordered; 824 } 825 826 /*---------------------------------------------------------------------------- 827 | Return whether the given value is an invalid floatx80 encoding. 828 | Invalid floatx80 encodings arise when the integer bit is not set, but 829 | the exponent is not zero. The only times the integer bit is permitted to 830 | be zero is in subnormal numbers and the value zero. 831 | This includes what the Intel software developer's manual calls pseudo-NaNs, 832 | pseudo-infinities and un-normal numbers. It does not include 833 | pseudo-denormals, which must still be correctly handled as inputs even 834 | if they are never generated as outputs. 835 *----------------------------------------------------------------------------*/ 836 static inline bool floatx80_invalid_encoding(floatx80 a) 837 { 838 #if defined(TARGET_M68K) 839 /*------------------------------------------------------------------------- 840 | With m68k, the explicit integer bit can be zero in the case of: 841 | - zeros (exp == 0, mantissa == 0) 842 | - denormalized numbers (exp == 0, mantissa != 0) 843 | - unnormalized numbers (exp != 0, exp < 0x7FFF) 844 | - infinities (exp == 0x7FFF, mantissa == 0) 845 | - not-a-numbers (exp == 0x7FFF, mantissa != 0) 846 | 847 | For infinities and NaNs, the explicit integer bit can be either one or 848 | zero. 849 | 850 | The IEEE 754 standard does not define a zero integer bit. Such a number 851 | is an unnormalized number. Hardware does not directly support 852 | denormalized and unnormalized numbers, but implicitly supports them by 853 | trapping them as unimplemented data types, allowing efficient conversion 854 | in software. 855 | 856 | See "M68000 FAMILY PROGRAMMER’S REFERENCE MANUAL", 857 | "1.6 FLOATING-POINT DATA TYPES" 858 *------------------------------------------------------------------------*/ 859 return false; 860 #else 861 return (a.low & (1ULL << 63)) == 0 && (a.high & 0x7FFF) != 0; 862 #endif 863 } 864 865 #define floatx80_zero make_floatx80(0x0000, 0x0000000000000000LL) 866 #define floatx80_zero_init make_floatx80_init(0x0000, 0x0000000000000000LL) 867 #define floatx80_one make_floatx80(0x3fff, 0x8000000000000000LL) 868 #define floatx80_ln2 make_floatx80(0x3ffe, 0xb17217f7d1cf79acLL) 869 #define floatx80_pi make_floatx80(0x4000, 0xc90fdaa22168c235LL) 870 #define floatx80_half make_floatx80(0x3ffe, 0x8000000000000000LL) 871 872 /*---------------------------------------------------------------------------- 873 | Returns the fraction bits of the extended double-precision floating-point 874 | value `a'. 875 *----------------------------------------------------------------------------*/ 876 877 static inline uint64_t extractFloatx80Frac(floatx80 a) 878 { 879 return a.low; 880 } 881 882 /*---------------------------------------------------------------------------- 883 | Returns the exponent bits of the extended double-precision floating-point 884 | value `a'. 885 *----------------------------------------------------------------------------*/ 886 887 static inline int32_t extractFloatx80Exp(floatx80 a) 888 { 889 return a.high & 0x7FFF; 890 } 891 892 /*---------------------------------------------------------------------------- 893 | Returns the sign bit of the extended double-precision floating-point value 894 | `a'. 895 *----------------------------------------------------------------------------*/ 896 897 static inline bool extractFloatx80Sign(floatx80 a) 898 { 899 return a.high >> 15; 900 } 901 902 /*---------------------------------------------------------------------------- 903 | Packs the sign `zSign', exponent `zExp', and significand `zSig' into an 904 | extended double-precision floating-point value, returning the result. 905 *----------------------------------------------------------------------------*/ 906 907 static inline floatx80 packFloatx80(bool zSign, int32_t zExp, uint64_t zSig) 908 { 909 floatx80 z; 910 911 z.low = zSig; 912 z.high = (((uint16_t)zSign) << 15) + zExp; 913 return z; 914 } 915 916 /*---------------------------------------------------------------------------- 917 | Normalizes the subnormal extended double-precision floating-point value 918 | represented by the denormalized significand `aSig'. The normalized exponent 919 | and significand are stored at the locations pointed to by `zExpPtr' and 920 | `zSigPtr', respectively. 921 *----------------------------------------------------------------------------*/ 922 923 void normalizeFloatx80Subnormal(uint64_t aSig, int32_t *zExpPtr, 924 uint64_t *zSigPtr); 925 926 /*---------------------------------------------------------------------------- 927 | Takes two extended double-precision floating-point values `a' and `b', one 928 | of which is a NaN, and returns the appropriate NaN result. If either `a' or 929 | `b' is a signaling NaN, the invalid exception is raised. 930 *----------------------------------------------------------------------------*/ 931 932 floatx80 propagateFloatx80NaN(floatx80 a, floatx80 b, float_status *status); 933 934 /*---------------------------------------------------------------------------- 935 | Takes an abstract floating-point value having sign `zSign', exponent `zExp', 936 | and extended significand formed by the concatenation of `zSig0' and `zSig1', 937 | and returns the proper extended double-precision floating-point value 938 | corresponding to the abstract input. Ordinarily, the abstract value is 939 | rounded and packed into the extended double-precision format, with the 940 | inexact exception raised if the abstract input cannot be represented 941 | exactly. However, if the abstract value is too large, the overflow and 942 | inexact exceptions are raised and an infinity or maximal finite value is 943 | returned. If the abstract value is too small, the input value is rounded to 944 | a subnormal number, and the underflow and inexact exceptions are raised if 945 | the abstract input cannot be represented exactly as a subnormal extended 946 | double-precision floating-point number. 947 | If `roundingPrecision' is 32 or 64, the result is rounded to the same 948 | number of bits as single or double precision, respectively. Otherwise, the 949 | result is rounded to the full precision of the extended double-precision 950 | format. 951 | The input significand must be normalized or smaller. If the input 952 | significand is not normalized, `zExp' must be 0; in that case, the result 953 | returned is a subnormal number, and it must not require rounding. The 954 | handling of underflow and overflow follows the IEC/IEEE Standard for Binary 955 | Floating-Point Arithmetic. 956 *----------------------------------------------------------------------------*/ 957 958 floatx80 roundAndPackFloatx80(int8_t roundingPrecision, bool zSign, 959 int32_t zExp, uint64_t zSig0, uint64_t zSig1, 960 float_status *status); 961 962 /*---------------------------------------------------------------------------- 963 | Takes an abstract floating-point value having sign `zSign', exponent 964 | `zExp', and significand formed by the concatenation of `zSig0' and `zSig1', 965 | and returns the proper extended double-precision floating-point value 966 | corresponding to the abstract input. This routine is just like 967 | `roundAndPackFloatx80' except that the input significand does not have to be 968 | normalized. 969 *----------------------------------------------------------------------------*/ 970 971 floatx80 normalizeRoundAndPackFloatx80(int8_t roundingPrecision, 972 bool zSign, int32_t zExp, 973 uint64_t zSig0, uint64_t zSig1, 974 float_status *status); 975 976 /*---------------------------------------------------------------------------- 977 | The pattern for a default generated extended double-precision NaN. 978 *----------------------------------------------------------------------------*/ 979 floatx80 floatx80_default_nan(float_status *status); 980 981 /*---------------------------------------------------------------------------- 982 | Software IEC/IEEE quadruple-precision conversion routines. 983 *----------------------------------------------------------------------------*/ 984 int32_t float128_to_int32(float128, float_status *status); 985 int32_t float128_to_int32_round_to_zero(float128, float_status *status); 986 int64_t float128_to_int64(float128, float_status *status); 987 int64_t float128_to_int64_round_to_zero(float128, float_status *status); 988 uint64_t float128_to_uint64(float128, float_status *status); 989 uint64_t float128_to_uint64_round_to_zero(float128, float_status *status); 990 uint32_t float128_to_uint32(float128, float_status *status); 991 uint32_t float128_to_uint32_round_to_zero(float128, float_status *status); 992 float32 float128_to_float32(float128, float_status *status); 993 float64 float128_to_float64(float128, float_status *status); 994 floatx80 float128_to_floatx80(float128, float_status *status); 995 996 /*---------------------------------------------------------------------------- 997 | Software IEC/IEEE quadruple-precision operations. 998 *----------------------------------------------------------------------------*/ 999 float128 float128_round_to_int(float128, float_status *status); 1000 float128 float128_add(float128, float128, float_status *status); 1001 float128 float128_sub(float128, float128, float_status *status); 1002 float128 float128_mul(float128, float128, float_status *status); 1003 float128 float128_div(float128, float128, float_status *status); 1004 float128 float128_rem(float128, float128, float_status *status); 1005 float128 float128_sqrt(float128, float_status *status); 1006 FloatRelation float128_compare(float128, float128, float_status *status); 1007 FloatRelation float128_compare_quiet(float128, float128, float_status *status); 1008 bool float128_is_quiet_nan(float128, float_status *status); 1009 bool float128_is_signaling_nan(float128, float_status *status); 1010 float128 float128_silence_nan(float128, float_status *status); 1011 float128 float128_scalbn(float128, int, float_status *status); 1012 1013 static inline float128 float128_abs(float128 a) 1014 { 1015 a.high &= 0x7fffffffffffffffLL; 1016 return a; 1017 } 1018 1019 static inline float128 float128_chs(float128 a) 1020 { 1021 a.high ^= 0x8000000000000000LL; 1022 return a; 1023 } 1024 1025 static inline bool float128_is_infinity(float128 a) 1026 { 1027 return (a.high & 0x7fffffffffffffffLL) == 0x7fff000000000000LL && a.low == 0; 1028 } 1029 1030 static inline bool float128_is_neg(float128 a) 1031 { 1032 return a.high >> 63; 1033 } 1034 1035 static inline bool float128_is_zero(float128 a) 1036 { 1037 return (a.high & 0x7fffffffffffffffLL) == 0 && a.low == 0; 1038 } 1039 1040 static inline bool float128_is_zero_or_denormal(float128 a) 1041 { 1042 return (a.high & 0x7fff000000000000LL) == 0; 1043 } 1044 1045 static inline bool float128_is_normal(float128 a) 1046 { 1047 return (((a.high >> 48) + 1) & 0x7fff) >= 2; 1048 } 1049 1050 static inline bool float128_is_denormal(float128 a) 1051 { 1052 return float128_is_zero_or_denormal(a) && !float128_is_zero(a); 1053 } 1054 1055 static inline bool float128_is_any_nan(float128 a) 1056 { 1057 return ((a.high >> 48) & 0x7fff) == 0x7fff && 1058 ((a.low != 0) || ((a.high & 0xffffffffffffLL) != 0)); 1059 } 1060 1061 static inline bool float128_eq(float128 a, float128 b, float_status *s) 1062 { 1063 return float128_compare(a, b, s) == float_relation_equal; 1064 } 1065 1066 static inline bool float128_le(float128 a, float128 b, float_status *s) 1067 { 1068 return float128_compare(a, b, s) <= float_relation_equal; 1069 } 1070 1071 static inline bool float128_lt(float128 a, float128 b, float_status *s) 1072 { 1073 return float128_compare(a, b, s) < float_relation_equal; 1074 } 1075 1076 static inline bool float128_unordered(float128 a, float128 b, float_status *s) 1077 { 1078 return float128_compare(a, b, s) == float_relation_unordered; 1079 } 1080 1081 static inline bool float128_eq_quiet(float128 a, float128 b, float_status *s) 1082 { 1083 return float128_compare_quiet(a, b, s) == float_relation_equal; 1084 } 1085 1086 static inline bool float128_le_quiet(float128 a, float128 b, float_status *s) 1087 { 1088 return float128_compare_quiet(a, b, s) <= float_relation_equal; 1089 } 1090 1091 static inline bool float128_lt_quiet(float128 a, float128 b, float_status *s) 1092 { 1093 return float128_compare_quiet(a, b, s) < float_relation_equal; 1094 } 1095 1096 static inline bool float128_unordered_quiet(float128 a, float128 b, 1097 float_status *s) 1098 { 1099 return float128_compare_quiet(a, b, s) == float_relation_unordered; 1100 } 1101 1102 #define float128_zero make_float128(0, 0) 1103 1104 /*---------------------------------------------------------------------------- 1105 | The pattern for a default generated quadruple-precision NaN. 1106 *----------------------------------------------------------------------------*/ 1107 float128 float128_default_nan(float_status *status); 1108 1109 #endif /* SOFTFLOAT_H */ 1110