| 1 | /* -*-c-*- |
| 2 | * |
| 3 | * $Id: mpx.c,v 1.3 1999/11/13 01:57:31 mdw Exp $ |
| 4 | * |
| 5 | * Low-level multiprecision arithmetic |
| 6 | * |
| 7 | * (c) 1999 Straylight/Edgeware |
| 8 | */ |
| 9 | |
| 10 | /*----- Licensing notice --------------------------------------------------* |
| 11 | * |
| 12 | * This file is part of Catacomb. |
| 13 | * |
| 14 | * Catacomb is free software; you can redistribute it and/or modify |
| 15 | * it under the terms of the GNU Library General Public License as |
| 16 | * published by the Free Software Foundation; either version 2 of the |
| 17 | * License, or (at your option) any later version. |
| 18 | * |
| 19 | * Catacomb is distributed in the hope that it will be useful, |
| 20 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 21 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 22 | * GNU Library General Public License for more details. |
| 23 | * |
| 24 | * You should have received a copy of the GNU Library General Public |
| 25 | * License along with Catacomb; if not, write to the Free |
| 26 | * Software Foundation, Inc., 59 Temple Place - Suite 330, Boston, |
| 27 | * MA 02111-1307, USA. |
| 28 | */ |
| 29 | |
| 30 | /*----- Revision history --------------------------------------------------* |
| 31 | * |
| 32 | * $Log: mpx.c,v $ |
| 33 | * Revision 1.3 1999/11/13 01:57:31 mdw |
| 34 | * Remove stray debugging code. |
| 35 | * |
| 36 | * Revision 1.2 1999/11/13 01:50:59 mdw |
| 37 | * Multiprecision routines finished and tested. |
| 38 | * |
| 39 | * Revision 1.1 1999/09/03 08:41:12 mdw |
| 40 | * Initial import. |
| 41 | * |
| 42 | */ |
| 43 | |
| 44 | /*----- Header files ------------------------------------------------------*/ |
| 45 | |
| 46 | #include <assert.h> |
| 47 | #include <stdio.h> |
| 48 | #include <stdlib.h> |
| 49 | #include <string.h> |
| 50 | |
| 51 | #include <mLib/bits.h> |
| 52 | |
| 53 | #include "mptypes.h" |
| 54 | #include "mpx.h" |
| 55 | |
| 56 | /*----- Loading and storing -----------------------------------------------*/ |
| 57 | |
| 58 | /* --- @mpx_storel@ --- * |
| 59 | * |
| 60 | * Arguments: @const mpw *v, *vl@ = base and limit of source vector |
| 61 | * @void *pp@ = pointer to octet array |
| 62 | * @size_t sz@ = size of octet array |
| 63 | * |
| 64 | * Returns: --- |
| 65 | * |
| 66 | * Use: Stores an MP in an octet array, least significant octet |
| 67 | * first. High-end octets are silently discarded if there |
| 68 | * isn't enough space for them. |
| 69 | */ |
| 70 | |
| 71 | void mpx_storel(const mpw *v, const mpw *vl, void *pp, size_t sz) |
| 72 | { |
| 73 | mpw n, w = 0; |
| 74 | octet *p = pp, *q = p + sz; |
| 75 | unsigned bits = 0; |
| 76 | |
| 77 | while (p < q) { |
| 78 | if (bits < 8) { |
| 79 | if (v >= vl) { |
| 80 | *p++ = U8(w); |
| 81 | break; |
| 82 | } |
| 83 | n = *v++; |
| 84 | *p++ = U8(w | n << bits); |
| 85 | w = n >> (8 - bits); |
| 86 | bits += MPW_BITS - 8; |
| 87 | } else { |
| 88 | *p++ = U8(w); |
| 89 | w >>= 8; |
| 90 | bits -= 8; |
| 91 | } |
| 92 | } |
| 93 | memset(p, 0, q - p); |
| 94 | } |
| 95 | |
| 96 | /* --- @mpx_loadl@ --- * |
| 97 | * |
| 98 | * Arguments: @mpw *v, *vl@ = base and limit of destination vector |
| 99 | * @const void *pp@ = pointer to octet array |
| 100 | * @size_t sz@ = size of octet array |
| 101 | * |
| 102 | * Returns: --- |
| 103 | * |
| 104 | * Use: Loads an MP in an octet array, least significant octet |
| 105 | * first. High-end octets are ignored if there isn't enough |
| 106 | * space for them. |
| 107 | */ |
| 108 | |
| 109 | void mpx_loadl(mpw *v, mpw *vl, const void *pp, size_t sz) |
| 110 | { |
| 111 | unsigned n; |
| 112 | mpw w = 0; |
| 113 | const octet *p = pp, *q = p + sz; |
| 114 | unsigned bits = 0; |
| 115 | |
| 116 | if (v >= vl) |
| 117 | return; |
| 118 | while (p < q) { |
| 119 | n = U8(*p++); |
| 120 | w |= n << bits; |
| 121 | bits += 8; |
| 122 | if (bits >= MPW_BITS) { |
| 123 | *v++ = MPW(w); |
| 124 | w = n >> (MPW_BITS - bits + 8); |
| 125 | bits -= MPW_BITS; |
| 126 | if (v >= vl) |
| 127 | return; |
| 128 | } |
| 129 | } |
| 130 | *v++ = w; |
| 131 | MPX_ZERO(v, vl); |
| 132 | } |
| 133 | |
| 134 | /* --- @mpx_storeb@ --- * |
| 135 | * |
| 136 | * Arguments: @const mpw *v, *vl@ = base and limit of source vector |
| 137 | * @void *pp@ = pointer to octet array |
| 138 | * @size_t sz@ = size of octet array |
| 139 | * |
| 140 | * Returns: --- |
| 141 | * |
| 142 | * Use: Stores an MP in an octet array, most significant octet |
| 143 | * first. High-end octets are silently discarded if there |
| 144 | * isn't enough space for them. |
| 145 | */ |
| 146 | |
| 147 | void mpx_storeb(const mpw *v, const mpw *vl, void *pp, size_t sz) |
| 148 | { |
| 149 | mpw n, w = 0; |
| 150 | octet *p = pp, *q = p + sz; |
| 151 | unsigned bits = 0; |
| 152 | |
| 153 | while (q > p) { |
| 154 | if (bits < 8) { |
| 155 | if (v >= vl) { |
| 156 | *--q = U8(w); |
| 157 | break; |
| 158 | } |
| 159 | n = *v++; |
| 160 | *--q = U8(w | n << bits); |
| 161 | w = n >> (8 - bits); |
| 162 | bits += MPW_BITS - 8; |
| 163 | } else { |
| 164 | *--q = U8(w); |
| 165 | w >>= 8; |
| 166 | bits -= 8; |
| 167 | } |
| 168 | } |
| 169 | memset(p, 0, q - p); |
| 170 | } |
| 171 | |
| 172 | /* --- @mpx_loadb@ --- * |
| 173 | * |
| 174 | * Arguments: @mpw *v, *vl@ = base and limit of destination vector |
| 175 | * @const void *pp@ = pointer to octet array |
| 176 | * @size_t sz@ = size of octet array |
| 177 | * |
| 178 | * Returns: --- |
| 179 | * |
| 180 | * Use: Loads an MP in an octet array, most significant octet |
| 181 | * first. High-end octets are ignored if there isn't enough |
| 182 | * space for them. |
| 183 | */ |
| 184 | |
| 185 | void mpx_loadb(mpw *v, mpw *vl, const void *pp, size_t sz) |
| 186 | { |
| 187 | unsigned n; |
| 188 | mpw w = 0; |
| 189 | const octet *p = pp, *q = p + sz; |
| 190 | unsigned bits = 0; |
| 191 | |
| 192 | if (v >= vl) |
| 193 | return; |
| 194 | while (q > p) { |
| 195 | n = U8(*--q); |
| 196 | w |= n << bits; |
| 197 | bits += 8; |
| 198 | if (bits >= MPW_BITS) { |
| 199 | *v++ = MPW(w); |
| 200 | w = n >> (MPW_BITS - bits + 8); |
| 201 | bits -= MPW_BITS; |
| 202 | if (v >= vl) |
| 203 | return; |
| 204 | } |
| 205 | } |
| 206 | *v++ = w; |
| 207 | MPX_ZERO(v, vl); |
| 208 | } |
| 209 | |
| 210 | /*----- Logical shifting --------------------------------------------------*/ |
| 211 | |
| 212 | /* --- @mpx_lsl@ --- * |
| 213 | * |
| 214 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
| 215 | * @const mpw *av, *avl@ = source vector base and limit |
| 216 | * @size_t n@ = number of bit positions to shift by |
| 217 | * |
| 218 | * Returns: --- |
| 219 | * |
| 220 | * Use: Performs a logical shift left operation on an integer. |
| 221 | */ |
| 222 | |
| 223 | void mpx_lsl(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n) |
| 224 | { |
| 225 | size_t nw; |
| 226 | unsigned nb; |
| 227 | |
| 228 | /* --- Trivial special case --- */ |
| 229 | |
| 230 | if (n == 0) |
| 231 | MPX_COPY(dv, dvl, av, avl); |
| 232 | |
| 233 | /* --- Single bit shifting --- */ |
| 234 | |
| 235 | else if (n == 1) { |
| 236 | mpw w = 0; |
| 237 | while (av < avl) { |
| 238 | mpw t; |
| 239 | if (dv >= dvl) |
| 240 | goto done; |
| 241 | t = *av++; |
| 242 | *dv++ = MPW((t << 1) | w); |
| 243 | w = t >> (MPW_BITS - 1); |
| 244 | } |
| 245 | if (dv >= dvl) |
| 246 | goto done; |
| 247 | *dv++ = MPW(w); |
| 248 | MPX_ZERO(dv, dvl); |
| 249 | goto done; |
| 250 | } |
| 251 | |
| 252 | /* --- Break out word and bit shifts for more sophisticated work --- */ |
| 253 | |
| 254 | nw = n / MPW_BITS; |
| 255 | nb = n % MPW_BITS; |
| 256 | |
| 257 | /* --- Handle a shift by a multiple of the word size --- */ |
| 258 | |
| 259 | if (nb == 0) { |
| 260 | MPX_COPY(dv + nw, dvl, av, avl); |
| 261 | memset(dv, 0, MPWS(nw)); |
| 262 | } |
| 263 | |
| 264 | /* --- And finally the difficult case --- * |
| 265 | * |
| 266 | * This is a little convoluted, because I have to start from the end and |
| 267 | * work backwards to avoid overwriting the source, if they're both the same |
| 268 | * block of memory. |
| 269 | */ |
| 270 | |
| 271 | else { |
| 272 | mpw w; |
| 273 | size_t nr = MPW_BITS - nb; |
| 274 | size_t dvn = dvl - dv; |
| 275 | size_t avn = avl - av; |
| 276 | |
| 277 | if (dvn <= nw) { |
| 278 | MPX_ZERO(dv, dvl); |
| 279 | goto done; |
| 280 | } |
| 281 | |
| 282 | if (dvn > avn + nw) { |
| 283 | size_t off = avn + nw + 1; |
| 284 | MPX_ZERO(dv + off, dvl); |
| 285 | dvl = dv + off; |
| 286 | w = 0; |
| 287 | } else { |
| 288 | avl = av + dvn - nw; |
| 289 | w = *--avl << nb; |
| 290 | } |
| 291 | |
| 292 | while (avl > av) { |
| 293 | mpw t = *--avl; |
| 294 | *--dvl = (t >> nr) | w; |
| 295 | w = t << nb; |
| 296 | } |
| 297 | |
| 298 | *--dvl = w; |
| 299 | MPX_ZERO(dv, dvl); |
| 300 | } |
| 301 | |
| 302 | done:; |
| 303 | } |
| 304 | |
| 305 | /* --- @mpx_lsr@ --- * |
| 306 | * |
| 307 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
| 308 | * @const mpw *av, *avl@ = source vector base and limit |
| 309 | * @size_t n@ = number of bit positions to shift by |
| 310 | * |
| 311 | * Returns: --- |
| 312 | * |
| 313 | * Use: Performs a logical shift right operation on an integer. |
| 314 | */ |
| 315 | |
| 316 | void mpx_lsr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n) |
| 317 | { |
| 318 | size_t nw; |
| 319 | unsigned nb; |
| 320 | |
| 321 | /* --- Trivial special case --- */ |
| 322 | |
| 323 | if (n == 0) |
| 324 | MPX_COPY(dv, dvl, av, avl); |
| 325 | |
| 326 | /* --- Single bit shifting --- */ |
| 327 | |
| 328 | else if (n == 1) { |
| 329 | mpw w = *av++ >> 1; |
| 330 | while (av < avl) { |
| 331 | mpw t; |
| 332 | if (dv >= dvl) |
| 333 | goto done; |
| 334 | t = *av++; |
| 335 | *dv++ = MPW((t << (MPW_BITS - 1)) | w); |
| 336 | w = t >> 1; |
| 337 | } |
| 338 | if (dv >= dvl) |
| 339 | goto done; |
| 340 | *dv++ = MPW(w); |
| 341 | MPX_ZERO(dv, dvl); |
| 342 | goto done; |
| 343 | } |
| 344 | |
| 345 | /* --- Break out word and bit shifts for more sophisticated work --- */ |
| 346 | |
| 347 | nw = n / MPW_BITS; |
| 348 | nb = n % MPW_BITS; |
| 349 | |
| 350 | /* --- Handle a shift by a multiple of the word size --- */ |
| 351 | |
| 352 | if (nb == 0) |
| 353 | MPX_COPY(dv, dvl, av + nw, avl); |
| 354 | |
| 355 | /* --- And finally the difficult case --- */ |
| 356 | |
| 357 | else { |
| 358 | mpw w; |
| 359 | size_t nr = MPW_BITS - nb; |
| 360 | |
| 361 | av += nw; |
| 362 | w = *av++; |
| 363 | while (av < avl) { |
| 364 | mpw t; |
| 365 | if (dv >= dvl) |
| 366 | goto done; |
| 367 | t = *av++; |
| 368 | *dv++ = MPW((w >> nb) | (t << nr)); |
| 369 | w = t; |
| 370 | } |
| 371 | if (dv < dvl) { |
| 372 | *dv++ = MPW(w >> nb); |
| 373 | MPX_ZERO(dv, dvl); |
| 374 | } |
| 375 | } |
| 376 | |
| 377 | done:; |
| 378 | } |
| 379 | |
| 380 | /*----- Unsigned arithmetic -----------------------------------------------*/ |
| 381 | |
| 382 | /* --- @mpx_ucmp@ --- * |
| 383 | * |
| 384 | * Arguments: @const mpw *av, *avl@ = first argument vector base and limit |
| 385 | * @const mpw *bv, *bvl@ = second argument vector base and limit |
| 386 | * |
| 387 | * Returns: Less than, equal to, or greater than zero depending on |
| 388 | * whether @a@ is less than, equal to or greater than @b@, |
| 389 | * respectively. |
| 390 | * |
| 391 | * Use: Performs an unsigned integer comparison. |
| 392 | */ |
| 393 | |
| 394 | int mpx_ucmp(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl) |
| 395 | { |
| 396 | MPX_SHRINK(av, avl); |
| 397 | MPX_SHRINK(bv, bvl); |
| 398 | |
| 399 | if (avl - av > bvl - bv) |
| 400 | return (+1); |
| 401 | else if (avl - av < bvl - bv) |
| 402 | return (-1); |
| 403 | else while (avl > av) { |
| 404 | mpw a = *--avl, b = *--bvl; |
| 405 | if (a > b) |
| 406 | return (+1); |
| 407 | else if (a < b) |
| 408 | return (-1); |
| 409 | } |
| 410 | return (0); |
| 411 | } |
| 412 | |
| 413 | /* --- @mpx_uadd@ --- * |
| 414 | * |
| 415 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
| 416 | * @const mpw *av, *avl@ = first addend vector base and limit |
| 417 | * @const mpw *bv, *bvl@ = second addend vector base and limit |
| 418 | * |
| 419 | * Returns: --- |
| 420 | * |
| 421 | * Use: Performs unsigned integer addition. If the result overflows |
| 422 | * the destination vector, high-order bits are discarded. This |
| 423 | * means that two's complement addition happens more or less for |
| 424 | * free, although that's more a side-effect than anything else. |
| 425 | * The result vector may be equal to either or both source |
| 426 | * vectors, but may not otherwise overlap them. |
| 427 | */ |
| 428 | |
| 429 | void mpx_uadd(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, |
| 430 | const mpw *bv, const mpw *bvl) |
| 431 | { |
| 432 | mpw c = 0; |
| 433 | |
| 434 | while (av < avl || bv < bvl) { |
| 435 | mpw a, b; |
| 436 | mpd x; |
| 437 | if (dv >= dvl) |
| 438 | return; |
| 439 | a = (av < avl) ? *av++ : 0; |
| 440 | b = (bv < bvl) ? *bv++ : 0; |
| 441 | x = (mpd)a + (mpd)b + c; |
| 442 | *dv++ = MPW(x); |
| 443 | c = x >> MPW_BITS; |
| 444 | } |
| 445 | if (dv < dvl) { |
| 446 | *dv++ = c; |
| 447 | MPX_ZERO(dv, dvl); |
| 448 | } |
| 449 | } |
| 450 | |
| 451 | /* --- @mpx_usub@ --- * |
| 452 | * |
| 453 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
| 454 | * @const mpw *av, *avl@ = first argument vector base and limit |
| 455 | * @const mpw *bv, *bvl@ = second argument vector base and limit |
| 456 | * |
| 457 | * Returns: --- |
| 458 | * |
| 459 | * Use: Performs unsigned integer subtraction. If the result |
| 460 | * overflows the destination vector, high-order bits are |
| 461 | * discarded. This means that two's complement subtraction |
| 462 | * happens more or less for free, althuogh that's more a side- |
| 463 | * effect than anything else. The result vector may be equal to |
| 464 | * either or both source vectors, but may not otherwise overlap |
| 465 | * them. |
| 466 | */ |
| 467 | |
| 468 | void mpx_usub(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, |
| 469 | const mpw *bv, const mpw *bvl) |
| 470 | { |
| 471 | mpw c = 0; |
| 472 | |
| 473 | while (av < avl || bv < bvl) { |
| 474 | mpw a, b; |
| 475 | mpd x; |
| 476 | if (dv >= dvl) |
| 477 | return; |
| 478 | a = (av < avl) ? *av++ : 0; |
| 479 | b = (bv < bvl) ? *bv++ : 0; |
| 480 | x = (mpd)a - (mpd)b - c; |
| 481 | *dv++ = MPW(x); |
| 482 | if (x >> MPW_BITS) |
| 483 | c = 1; |
| 484 | else |
| 485 | c = 0; |
| 486 | } |
| 487 | if (c) |
| 488 | c = MPW_MAX; |
| 489 | while (dv < dvl) |
| 490 | *dv++ = c; |
| 491 | } |
| 492 | |
| 493 | /* --- @mpx_umul@ --- * |
| 494 | * |
| 495 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
| 496 | * @const mpw *av, *avl@ = multiplicand vector base and limit |
| 497 | * @const mpw *bv, *bvl@ = multiplier vector base and limit |
| 498 | * |
| 499 | * Returns: --- |
| 500 | * |
| 501 | * Use: Performs unsigned integer multiplication. If the result |
| 502 | * overflows the desination vector, high-order bits are |
| 503 | * discarded. The result vector may not overlap the argument |
| 504 | * vectors in any way. |
| 505 | */ |
| 506 | |
| 507 | void mpx_umul(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, |
| 508 | const mpw *bv, const mpw *bvl) |
| 509 | { |
| 510 | /* --- This is probably worthwhile on a multiply --- */ |
| 511 | |
| 512 | MPX_SHRINK(av, avl); |
| 513 | MPX_SHRINK(bv, bvl); |
| 514 | |
| 515 | /* --- Deal with a multiply by zero --- */ |
| 516 | |
| 517 | if (bv == bvl) { |
| 518 | MPX_ZERO(dv, dvl); |
| 519 | return; |
| 520 | } |
| 521 | |
| 522 | /* --- Do the initial multiply and initialize the accumulator --- */ |
| 523 | |
| 524 | MPX_UMULN(dv, dvl, av, avl, *bv++); |
| 525 | |
| 526 | /* --- Do the remaining multiply/accumulates --- */ |
| 527 | |
| 528 | while (dv < dvl && bv < bvl) { |
| 529 | mpw m = *bv++; |
| 530 | mpw c = 0; |
| 531 | const mpw *avv = av; |
| 532 | mpw *dvv = ++dv; |
| 533 | |
| 534 | while (avv < avl) { |
| 535 | mpd x; |
| 536 | if (dvv >= dvl) |
| 537 | goto next; |
| 538 | x = (mpd)*dvv + (mpd)m * (mpd)*avv++ + c; |
| 539 | *dvv++ = MPW(x); |
| 540 | c = x >> MPW_BITS; |
| 541 | } |
| 542 | MPX_UADDN(dvv, dvl, c); |
| 543 | next:; |
| 544 | } |
| 545 | } |
| 546 | |
| 547 | /* --- @mpx_usqr@ --- * |
| 548 | * |
| 549 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
| 550 | * @const mpw *av, *av@ = source vector base and limit |
| 551 | * |
| 552 | * Returns: --- |
| 553 | * |
| 554 | * Use: Performs unsigned integer squaring. The result vector must |
| 555 | * not overlap the source vector in any way. |
| 556 | */ |
| 557 | |
| 558 | void mpx_usqr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl) |
| 559 | { |
| 560 | MPX_ZERO(dv, dvl); |
| 561 | |
| 562 | /* --- Main loop --- */ |
| 563 | |
| 564 | while (av < avl) { |
| 565 | const mpw *avv = av; |
| 566 | mpw *dvv = dv; |
| 567 | mpw a = *av; |
| 568 | mpd c; |
| 569 | |
| 570 | /* --- Stop if I've run out of destination --- */ |
| 571 | |
| 572 | if (dvv >= dvl) |
| 573 | break; |
| 574 | |
| 575 | /* --- Work out the square at this point in the proceedings --- */ |
| 576 | |
| 577 | { |
| 578 | mpd x = (mpd)a * (mpd)a + *dvv; |
| 579 | *dvv++ = MPW(x); |
| 580 | c = MPW(x >> MPW_BITS); |
| 581 | } |
| 582 | |
| 583 | /* --- Now fix up the rest of the vector upwards --- */ |
| 584 | |
| 585 | avv++; |
| 586 | while (dvv < dvl && avv < avl) { |
| 587 | mpd x = (mpd)a * (mpd)*avv++; |
| 588 | mpd y = ((x << 1) & MPW_MAX) + c + *dvv; |
| 589 | c = (x >> (MPW_BITS - 1)) + (y >> MPW_BITS); |
| 590 | *dvv++ = MPW(y); |
| 591 | } |
| 592 | while (dvv < dvl && c) { |
| 593 | mpd x = c + *dvv; |
| 594 | *dvv++ = MPW(x); |
| 595 | c = x >> MPW_BITS; |
| 596 | } |
| 597 | |
| 598 | /* --- Get ready for the next round --- */ |
| 599 | |
| 600 | av++; |
| 601 | dv += 2; |
| 602 | } |
| 603 | } |
| 604 | |
| 605 | /* --- @mpx_udiv@ --- * |
| 606 | * |
| 607 | * Arguments: @mpw *qv, *qvl@ = quotient vector base and limit |
| 608 | * @mpw *rv, *rvl@ = dividend/remainder vector base and limit |
| 609 | * @const mpw *dv, *dvl@ = divisor vector base and limit |
| 610 | * @mpw *sv, *svl@ = scratch workspace |
| 611 | * |
| 612 | * Returns: --- |
| 613 | * |
| 614 | * Use: Performs unsigned integer division. If the result overflows |
| 615 | * the quotient vector, high-order bits are discarded. (Clearly |
| 616 | * the remainder vector can't overflow.) The various vectors |
| 617 | * may not overlap in any way. Yes, I know it's a bit odd |
| 618 | * requiring the dividend to be in the result position but it |
| 619 | * does make some sense really. The remainder must have |
| 620 | * headroom for at least two extra words. The scratch space |
| 621 | * must be at least two words larger than twice the size of the |
| 622 | * divisor. |
| 623 | */ |
| 624 | |
| 625 | void mpx_udiv(mpw *qv, mpw *qvl, mpw *rv, mpw *rvl, |
| 626 | const mpw *dv, const mpw *dvl, |
| 627 | mpw *sv, mpw *svl) |
| 628 | { |
| 629 | unsigned norm = 0; |
| 630 | size_t scale; |
| 631 | mpw d, dd; |
| 632 | |
| 633 | /* --- Initialize the quotient --- */ |
| 634 | |
| 635 | MPX_ZERO(qv, qvl); |
| 636 | |
| 637 | /* --- Perform some sanity checks --- */ |
| 638 | |
| 639 | MPX_SHRINK(dv, dvl); |
| 640 | assert(((void)"division by zero in mpx_udiv", dv < dvl)); |
| 641 | |
| 642 | /* --- Normalize the divisor --- * |
| 643 | * |
| 644 | * The algorithm requires that the divisor be at least two digits long. |
| 645 | * This is easy to fix. |
| 646 | */ |
| 647 | |
| 648 | { |
| 649 | unsigned b; |
| 650 | |
| 651 | d = dvl[-1]; |
| 652 | for (b = MPW_BITS / 2; b; b >>= 1) { |
| 653 | if (d < (MPW_MAX >> b)) { |
| 654 | d <<= b; |
| 655 | norm += b; |
| 656 | } |
| 657 | } |
| 658 | if (dv + 1 == dvl) |
| 659 | norm += MPW_BITS; |
| 660 | } |
| 661 | |
| 662 | /* --- Normalize the dividend/remainder to match --- */ |
| 663 | |
| 664 | if (norm) { |
| 665 | mpw *svvl = sv + (dvl - dv) + 1; |
| 666 | mpx_lsl(rv, rvl, rv, rvl, norm); |
| 667 | mpx_lsl(sv, svvl, dv, dvl, norm); |
| 668 | dv = sv; |
| 669 | sv = svvl; |
| 670 | dvl = svvl; |
| 671 | MPX_SHRINK(dv, dvl); |
| 672 | } |
| 673 | |
| 674 | MPX_SHRINK(rv, rvl); |
| 675 | d = dvl[-1]; |
| 676 | dd = dvl[-2]; |
| 677 | |
| 678 | /* --- Work out the relative scales --- */ |
| 679 | |
| 680 | { |
| 681 | size_t rvn = rvl - rv; |
| 682 | size_t dvn = dvl - dv; |
| 683 | |
| 684 | /* --- If the divisor is clearly larger, notice this --- */ |
| 685 | |
| 686 | if (dvn > rvn) { |
| 687 | mpx_lsr(rv, rvl, rv, rvl, norm); |
| 688 | return; |
| 689 | } |
| 690 | |
| 691 | scale = rvn - dvn; |
| 692 | } |
| 693 | |
| 694 | /* --- Calculate the most significant quotient digit --- * |
| 695 | * |
| 696 | * Because the divisor has its top bit set, this can only happen once. The |
| 697 | * pointer arithmetic is a little contorted, to make sure that the |
| 698 | * behaviour is defined. |
| 699 | */ |
| 700 | |
| 701 | if (MPX_UCMP(rv + scale, rvl, >=, dv, dvl)) { |
| 702 | mpx_usub(rv + scale, rvl, rv + scale, rvl, dv, dvl); |
| 703 | if (qvl - qv > scale) |
| 704 | qv[scale] = 1; |
| 705 | } |
| 706 | |
| 707 | /* --- Now for the main loop --- */ |
| 708 | |
| 709 | { |
| 710 | mpw *rvv = rvl - 2; |
| 711 | |
| 712 | while (scale) { |
| 713 | mpw q; |
| 714 | mpd rh; |
| 715 | |
| 716 | /* --- Get an estimate for the next quotient digit --- */ |
| 717 | |
| 718 | mpw r = rvv[1]; |
| 719 | mpw rr = rvv[0]; |
| 720 | mpw rrr = *--rvv; |
| 721 | |
| 722 | scale--; |
| 723 | rh = ((mpd)r << MPW_BITS) | rr; |
| 724 | if (r == d) |
| 725 | q = MPW_MAX; |
| 726 | else |
| 727 | q = MPW(rh / d); |
| 728 | |
| 729 | /* --- Refine the estimate --- */ |
| 730 | |
| 731 | { |
| 732 | mpd yh = (mpd)d * q; |
| 733 | mpd yl = (mpd)dd * q; |
| 734 | |
| 735 | if (yl > MPW_MAX) { |
| 736 | yh += yl >> MPW_BITS; |
| 737 | yl &= MPW_MAX; |
| 738 | } |
| 739 | |
| 740 | while (yh > rh || (yh == rh && yl > rrr)) { |
| 741 | q--; |
| 742 | yh -= d; |
| 743 | if (yl < dd) { |
| 744 | yh++; |
| 745 | yl += MPW_MAX; |
| 746 | } |
| 747 | yl -= dd; |
| 748 | } |
| 749 | } |
| 750 | |
| 751 | /* --- Remove a chunk from the dividend --- */ |
| 752 | |
| 753 | { |
| 754 | mpw *svv; |
| 755 | const mpw *dvv; |
| 756 | mpw c = 0; |
| 757 | |
| 758 | /* --- Calculate the size of the chunk --- */ |
| 759 | |
| 760 | for (svv = sv, dvv = dv; dvv < dvl; svv++, dvv++) { |
| 761 | mpd x = (mpd)*dvv * (mpd)q + c; |
| 762 | *svv = MPW(x); |
| 763 | c = x >> MPW_BITS; |
| 764 | } |
| 765 | if (c) |
| 766 | *svv++ = c; |
| 767 | |
| 768 | /* --- Now make sure that we can cope with the difference --- * |
| 769 | * |
| 770 | * Take advantage of the fact that subtraction works two's- |
| 771 | * complement. |
| 772 | */ |
| 773 | |
| 774 | mpx_usub(rv + scale, rvl, rv + scale, rvl, sv, svv); |
| 775 | if (rvl[-1] > MPW_MAX / 2) { |
| 776 | mpx_uadd(rv + scale, rvl, rv + scale, rvl, dv, dvl); |
| 777 | q--; |
| 778 | } |
| 779 | } |
| 780 | |
| 781 | /* --- Done for another iteration --- */ |
| 782 | |
| 783 | if (qvl - qv > scale) |
| 784 | qv[scale] = q; |
| 785 | r = rr; |
| 786 | rr = rrr; |
| 787 | } |
| 788 | } |
| 789 | |
| 790 | /* --- Now fiddle with unnormalizing and things --- */ |
| 791 | |
| 792 | mpx_lsr(rv, rvl, rv, rvl, norm); |
| 793 | } |
| 794 | |
| 795 | /*----- That's all, folks -------------------------------------------------*/ |