d03ab969 |
1 | /* -*-c-*- |
2 | * |
4f29a732 |
3 | * $Id: mpx.c,v 1.14 2002/10/19 18:55:08 mdw Exp $ |
d03ab969 |
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 $ |
4f29a732 |
33 | * Revision 1.14 2002/10/19 18:55:08 mdw |
34 | * Fix overflows in shift primitives. |
35 | * |
75263f25 |
36 | * Revision 1.13 2002/10/19 17:56:50 mdw |
37 | * Fix bit operations. Test them (a bit) better. |
38 | * |
f09e814a |
39 | * Revision 1.12 2002/10/06 22:52:50 mdw |
40 | * Pile of changes for supporting two's complement properly. |
41 | * |
0f32e0f8 |
42 | * Revision 1.11 2001/04/03 19:36:05 mdw |
43 | * Add some simple bitwise operations so that Perl can use them. |
44 | * |
1a05a8ef |
45 | * Revision 1.10 2000/10/08 12:06:12 mdw |
46 | * Provide @mpx_ueq@ for rapidly testing equality of two integers. |
47 | * |
99b30c23 |
48 | * Revision 1.9 2000/06/26 07:52:50 mdw |
49 | * Portability fix for the bug fix. |
50 | * |
ce76ff16 |
51 | * Revision 1.8 2000/06/25 12:59:02 mdw |
52 | * (mpx_udiv): Fix bug in quotient digit estimation. |
53 | * |
698bd937 |
54 | * Revision 1.7 1999/12/22 15:49:07 mdw |
55 | * New function for division by a small integer. |
56 | * |
42684bdb |
57 | * Revision 1.6 1999/11/20 22:43:44 mdw |
58 | * Integrate testing for MPX routines. |
59 | * |
dd517851 |
60 | * Revision 1.5 1999/11/20 22:23:27 mdw |
61 | * Add function versions of some low-level macros with wider use. |
62 | * |
f45a00c6 |
63 | * Revision 1.4 1999/11/17 18:04:09 mdw |
64 | * Add two's-complement functionality. Improve mpx_udiv a little by |
65 | * performing the multiplication of the divisor by q with the subtraction |
66 | * from r. |
67 | * |
501da53c |
68 | * Revision 1.3 1999/11/13 01:57:31 mdw |
69 | * Remove stray debugging code. |
70 | * |
c8a2f9ef |
71 | * Revision 1.2 1999/11/13 01:50:59 mdw |
72 | * Multiprecision routines finished and tested. |
73 | * |
d03ab969 |
74 | * Revision 1.1 1999/09/03 08:41:12 mdw |
75 | * Initial import. |
76 | * |
77 | */ |
78 | |
79 | /*----- Header files ------------------------------------------------------*/ |
80 | |
c8a2f9ef |
81 | #include <assert.h> |
d03ab969 |
82 | #include <stdio.h> |
83 | #include <stdlib.h> |
84 | #include <string.h> |
85 | |
86 | #include <mLib/bits.h> |
87 | |
88 | #include "mptypes.h" |
89 | #include "mpx.h" |
75263f25 |
90 | #include "bitops.h" |
d03ab969 |
91 | |
92 | /*----- Loading and storing -----------------------------------------------*/ |
93 | |
94 | /* --- @mpx_storel@ --- * |
95 | * |
96 | * Arguments: @const mpw *v, *vl@ = base and limit of source vector |
c8a2f9ef |
97 | * @void *pp@ = pointer to octet array |
d03ab969 |
98 | * @size_t sz@ = size of octet array |
99 | * |
100 | * Returns: --- |
101 | * |
102 | * Use: Stores an MP in an octet array, least significant octet |
103 | * first. High-end octets are silently discarded if there |
104 | * isn't enough space for them. |
105 | */ |
106 | |
c8a2f9ef |
107 | void mpx_storel(const mpw *v, const mpw *vl, void *pp, size_t sz) |
d03ab969 |
108 | { |
109 | mpw n, w = 0; |
c8a2f9ef |
110 | octet *p = pp, *q = p + sz; |
d03ab969 |
111 | unsigned bits = 0; |
112 | |
113 | while (p < q) { |
114 | if (bits < 8) { |
115 | if (v >= vl) { |
116 | *p++ = U8(w); |
117 | break; |
118 | } |
119 | n = *v++; |
120 | *p++ = U8(w | n << bits); |
121 | w = n >> (8 - bits); |
122 | bits += MPW_BITS - 8; |
123 | } else { |
124 | *p++ = U8(w); |
125 | w >>= 8; |
126 | bits -= 8; |
127 | } |
128 | } |
129 | memset(p, 0, q - p); |
130 | } |
131 | |
132 | /* --- @mpx_loadl@ --- * |
133 | * |
134 | * Arguments: @mpw *v, *vl@ = base and limit of destination vector |
c8a2f9ef |
135 | * @const void *pp@ = pointer to octet array |
d03ab969 |
136 | * @size_t sz@ = size of octet array |
137 | * |
138 | * Returns: --- |
139 | * |
140 | * Use: Loads an MP in an octet array, least significant octet |
141 | * first. High-end octets are ignored if there isn't enough |
142 | * space for them. |
143 | */ |
144 | |
c8a2f9ef |
145 | void mpx_loadl(mpw *v, mpw *vl, const void *pp, size_t sz) |
d03ab969 |
146 | { |
147 | unsigned n; |
c8a2f9ef |
148 | mpw w = 0; |
149 | const octet *p = pp, *q = p + sz; |
d03ab969 |
150 | unsigned bits = 0; |
151 | |
152 | if (v >= vl) |
153 | return; |
154 | while (p < q) { |
155 | n = U8(*p++); |
156 | w |= n << bits; |
157 | bits += 8; |
158 | if (bits >= MPW_BITS) { |
159 | *v++ = MPW(w); |
160 | w = n >> (MPW_BITS - bits + 8); |
161 | bits -= MPW_BITS; |
162 | if (v >= vl) |
163 | return; |
164 | } |
165 | } |
166 | *v++ = w; |
167 | MPX_ZERO(v, vl); |
168 | } |
169 | |
170 | /* --- @mpx_storeb@ --- * |
171 | * |
172 | * Arguments: @const mpw *v, *vl@ = base and limit of source vector |
c8a2f9ef |
173 | * @void *pp@ = pointer to octet array |
d03ab969 |
174 | * @size_t sz@ = size of octet array |
175 | * |
176 | * Returns: --- |
177 | * |
178 | * Use: Stores an MP in an octet array, most significant octet |
179 | * first. High-end octets are silently discarded if there |
180 | * isn't enough space for them. |
181 | */ |
182 | |
c8a2f9ef |
183 | void mpx_storeb(const mpw *v, const mpw *vl, void *pp, size_t sz) |
d03ab969 |
184 | { |
185 | mpw n, w = 0; |
c8a2f9ef |
186 | octet *p = pp, *q = p + sz; |
d03ab969 |
187 | unsigned bits = 0; |
188 | |
189 | while (q > p) { |
190 | if (bits < 8) { |
191 | if (v >= vl) { |
192 | *--q = U8(w); |
193 | break; |
194 | } |
195 | n = *v++; |
196 | *--q = U8(w | n << bits); |
197 | w = n >> (8 - bits); |
198 | bits += MPW_BITS - 8; |
199 | } else { |
200 | *--q = U8(w); |
201 | w >>= 8; |
202 | bits -= 8; |
203 | } |
204 | } |
205 | memset(p, 0, q - p); |
206 | } |
207 | |
208 | /* --- @mpx_loadb@ --- * |
209 | * |
210 | * Arguments: @mpw *v, *vl@ = base and limit of destination vector |
c8a2f9ef |
211 | * @const void *pp@ = pointer to octet array |
d03ab969 |
212 | * @size_t sz@ = size of octet array |
213 | * |
214 | * Returns: --- |
215 | * |
216 | * Use: Loads an MP in an octet array, most significant octet |
217 | * first. High-end octets are ignored if there isn't enough |
218 | * space for them. |
219 | */ |
220 | |
c8a2f9ef |
221 | void mpx_loadb(mpw *v, mpw *vl, const void *pp, size_t sz) |
d03ab969 |
222 | { |
223 | unsigned n; |
c8a2f9ef |
224 | mpw w = 0; |
225 | const octet *p = pp, *q = p + sz; |
d03ab969 |
226 | unsigned bits = 0; |
227 | |
228 | if (v >= vl) |
229 | return; |
230 | while (q > p) { |
231 | n = U8(*--q); |
232 | w |= n << bits; |
233 | bits += 8; |
234 | if (bits >= MPW_BITS) { |
235 | *v++ = MPW(w); |
236 | w = n >> (MPW_BITS - bits + 8); |
237 | bits -= MPW_BITS; |
238 | if (v >= vl) |
239 | return; |
240 | } |
241 | } |
242 | *v++ = w; |
243 | MPX_ZERO(v, vl); |
244 | } |
245 | |
f09e814a |
246 | /* --- @mpx_storel2cn@ --- * |
247 | * |
248 | * Arguments: @const mpw *v, *vl@ = base and limit of source vector |
249 | * @void *pp@ = pointer to octet array |
250 | * @size_t sz@ = size of octet array |
251 | * |
252 | * Returns: --- |
253 | * |
254 | * Use: Stores a negative MP in an octet array, least significant |
255 | * octet first, as two's complement. High-end octets are |
256 | * silently discarded if there isn't enough space for them. |
257 | * This obviously makes the output bad. |
258 | */ |
259 | |
260 | void mpx_storel2cn(const mpw *v, const mpw *vl, void *pp, size_t sz) |
261 | { |
262 | unsigned c = 1; |
263 | unsigned b = 0; |
264 | mpw n, w = 0; |
265 | octet *p = pp, *q = p + sz; |
266 | unsigned bits = 0; |
267 | |
268 | while (p < q) { |
269 | if (bits < 8) { |
270 | if (v >= vl) { |
271 | b = w; |
272 | break; |
273 | } |
274 | n = *v++; |
275 | b = w | n << bits; |
276 | w = n >> (8 - bits); |
277 | bits += MPW_BITS - 8; |
278 | } else { |
279 | b = w; |
280 | w >>= 8; |
281 | bits -= 8; |
282 | } |
283 | b = U8(~b + c); |
284 | c = !b; |
285 | *p++ = b; |
286 | } |
287 | while (p < q) { |
288 | b = U8(~b + c); |
289 | c = !b; |
290 | *p++ = b; |
291 | b = 0; |
292 | } |
293 | } |
294 | |
295 | /* --- @mpx_loadl2cn@ --- * |
296 | * |
297 | * Arguments: @mpw *v, *vl@ = base and limit of destination vector |
298 | * @const void *pp@ = pointer to octet array |
299 | * @size_t sz@ = size of octet array |
300 | * |
301 | * Returns: --- |
302 | * |
303 | * Use: Loads a negative MP in an octet array, least significant |
304 | * octet first, as two's complement. High-end octets are |
305 | * ignored if there isn't enough space for them. This probably |
306 | * means you made the wrong choice coming here. |
307 | */ |
308 | |
309 | void mpx_loadl2cn(mpw *v, mpw *vl, const void *pp, size_t sz) |
310 | { |
311 | unsigned n; |
312 | unsigned c = 1; |
313 | mpw w = 0; |
314 | const octet *p = pp, *q = p + sz; |
315 | unsigned bits = 0; |
316 | |
317 | if (v >= vl) |
318 | return; |
319 | while (p < q) { |
320 | n = U8(~(*p++) + c); |
321 | c = !n; |
322 | w |= n << bits; |
323 | bits += 8; |
324 | if (bits >= MPW_BITS) { |
325 | *v++ = MPW(w); |
326 | w = n >> (MPW_BITS - bits + 8); |
327 | bits -= MPW_BITS; |
328 | if (v >= vl) |
329 | return; |
330 | } |
331 | } |
332 | *v++ = w; |
333 | MPX_ZERO(v, vl); |
334 | } |
335 | |
336 | /* --- @mpx_storeb2cn@ --- * |
337 | * |
338 | * Arguments: @const mpw *v, *vl@ = base and limit of source vector |
339 | * @void *pp@ = pointer to octet array |
340 | * @size_t sz@ = size of octet array |
341 | * |
342 | * Returns: --- |
343 | * |
344 | * Use: Stores a negative MP in an octet array, most significant |
345 | * octet first, as two's complement. High-end octets are |
346 | * silently discarded if there isn't enough space for them, |
347 | * which probably isn't what you meant. |
348 | */ |
349 | |
350 | void mpx_storeb2cn(const mpw *v, const mpw *vl, void *pp, size_t sz) |
351 | { |
352 | mpw n, w = 0; |
353 | unsigned b = 0; |
354 | unsigned c = 1; |
355 | octet *p = pp, *q = p + sz; |
356 | unsigned bits = 0; |
357 | |
358 | while (q > p) { |
359 | if (bits < 8) { |
360 | if (v >= vl) { |
361 | b = w; |
362 | break; |
363 | } |
364 | n = *v++; |
365 | b = w | n << bits; |
366 | w = n >> (8 - bits); |
367 | bits += MPW_BITS - 8; |
368 | } else { |
369 | b = w; |
370 | w >>= 8; |
371 | bits -= 8; |
372 | } |
373 | b = U8(~b + c); |
374 | c = !b; |
375 | *--q = b; |
376 | } |
377 | while (q > p) { |
378 | b = ~b + c; |
379 | c = !(b & 0xff); |
380 | *--q = b; |
381 | b = 0; |
382 | } |
383 | } |
384 | |
385 | /* --- @mpx_loadb2cn@ --- * |
386 | * |
387 | * Arguments: @mpw *v, *vl@ = base and limit of destination vector |
388 | * @const void *pp@ = pointer to octet array |
389 | * @size_t sz@ = size of octet array |
390 | * |
391 | * Returns: --- |
392 | * |
393 | * Use: Loads a negative MP in an octet array, most significant octet |
394 | * first as two's complement. High-end octets are ignored if |
395 | * there isn't enough space for them. This probably means you |
396 | * chose this function wrongly. |
397 | */ |
398 | |
399 | void mpx_loadb2cn(mpw *v, mpw *vl, const void *pp, size_t sz) |
400 | { |
401 | unsigned n; |
402 | unsigned c = 1; |
403 | mpw w = 0; |
404 | const octet *p = pp, *q = p + sz; |
405 | unsigned bits = 0; |
406 | |
407 | if (v >= vl) |
408 | return; |
409 | while (q > p) { |
410 | n = U8(~(*--q) + c); |
411 | c = !n; |
412 | w |= n << bits; |
413 | bits += 8; |
414 | if (bits >= MPW_BITS) { |
415 | *v++ = MPW(w); |
416 | w = n >> (MPW_BITS - bits + 8); |
417 | bits -= MPW_BITS; |
418 | if (v >= vl) |
419 | return; |
420 | } |
421 | } |
422 | *v++ = w; |
423 | MPX_ZERO(v, vl); |
424 | } |
425 | |
d03ab969 |
426 | /*----- Logical shifting --------------------------------------------------*/ |
427 | |
428 | /* --- @mpx_lsl@ --- * |
429 | * |
430 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
431 | * @const mpw *av, *avl@ = source vector base and limit |
432 | * @size_t n@ = number of bit positions to shift by |
433 | * |
434 | * Returns: --- |
435 | * |
436 | * Use: Performs a logical shift left operation on an integer. |
437 | */ |
438 | |
439 | void mpx_lsl(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n) |
440 | { |
441 | size_t nw; |
442 | unsigned nb; |
443 | |
444 | /* --- Trivial special case --- */ |
445 | |
446 | if (n == 0) |
447 | MPX_COPY(dv, dvl, av, avl); |
448 | |
449 | /* --- Single bit shifting --- */ |
450 | |
451 | else if (n == 1) { |
452 | mpw w = 0; |
453 | while (av < avl) { |
454 | mpw t; |
455 | if (dv >= dvl) |
456 | goto done; |
457 | t = *av++; |
458 | *dv++ = MPW((t << 1) | w); |
459 | w = t >> (MPW_BITS - 1); |
460 | } |
461 | if (dv >= dvl) |
462 | goto done; |
463 | *dv++ = MPW(w); |
464 | MPX_ZERO(dv, dvl); |
c8a2f9ef |
465 | goto done; |
d03ab969 |
466 | } |
467 | |
468 | /* --- Break out word and bit shifts for more sophisticated work --- */ |
469 | |
470 | nw = n / MPW_BITS; |
471 | nb = n % MPW_BITS; |
472 | |
473 | /* --- Handle a shift by a multiple of the word size --- */ |
474 | |
475 | if (nb == 0) { |
4f29a732 |
476 | if (nw >= dvl - dv) |
477 | MPX_ZERO(dv, dvl); |
478 | else { |
479 | MPX_COPY(dv + nw, dvl, av, avl); |
480 | memset(dv, 0, MPWS(nw)); |
481 | } |
d03ab969 |
482 | } |
483 | |
c8a2f9ef |
484 | /* --- And finally the difficult case --- * |
485 | * |
486 | * This is a little convoluted, because I have to start from the end and |
487 | * work backwards to avoid overwriting the source, if they're both the same |
488 | * block of memory. |
489 | */ |
d03ab969 |
490 | |
491 | else { |
492 | mpw w; |
493 | size_t nr = MPW_BITS - nb; |
c8a2f9ef |
494 | size_t dvn = dvl - dv; |
495 | size_t avn = avl - av; |
d03ab969 |
496 | |
c8a2f9ef |
497 | if (dvn <= nw) { |
d03ab969 |
498 | MPX_ZERO(dv, dvl); |
499 | goto done; |
500 | } |
d03ab969 |
501 | |
c8a2f9ef |
502 | if (dvn > avn + nw) { |
503 | size_t off = avn + nw + 1; |
504 | MPX_ZERO(dv + off, dvl); |
505 | dvl = dv + off; |
506 | w = 0; |
507 | } else { |
508 | avl = av + dvn - nw; |
509 | w = *--avl << nb; |
d03ab969 |
510 | } |
511 | |
c8a2f9ef |
512 | while (avl > av) { |
513 | mpw t = *--avl; |
514 | *--dvl = (t >> nr) | w; |
515 | w = t << nb; |
d03ab969 |
516 | } |
c8a2f9ef |
517 | |
518 | *--dvl = w; |
519 | MPX_ZERO(dv, dvl); |
d03ab969 |
520 | } |
521 | |
522 | done:; |
523 | } |
524 | |
525 | /* --- @mpx_lsr@ --- * |
526 | * |
527 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
528 | * @const mpw *av, *avl@ = source vector base and limit |
529 | * @size_t n@ = number of bit positions to shift by |
530 | * |
531 | * Returns: --- |
532 | * |
533 | * Use: Performs a logical shift right operation on an integer. |
534 | */ |
535 | |
536 | void mpx_lsr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, size_t n) |
537 | { |
538 | size_t nw; |
539 | unsigned nb; |
540 | |
541 | /* --- Trivial special case --- */ |
542 | |
543 | if (n == 0) |
544 | MPX_COPY(dv, dvl, av, avl); |
545 | |
546 | /* --- Single bit shifting --- */ |
547 | |
548 | else if (n == 1) { |
549 | mpw w = *av++ >> 1; |
550 | while (av < avl) { |
551 | mpw t; |
552 | if (dv >= dvl) |
553 | goto done; |
554 | t = *av++; |
555 | *dv++ = MPW((t << (MPW_BITS - 1)) | w); |
556 | w = t >> 1; |
557 | } |
558 | if (dv >= dvl) |
559 | goto done; |
560 | *dv++ = MPW(w); |
561 | MPX_ZERO(dv, dvl); |
c8a2f9ef |
562 | goto done; |
d03ab969 |
563 | } |
564 | |
565 | /* --- Break out word and bit shifts for more sophisticated work --- */ |
566 | |
567 | nw = n / MPW_BITS; |
568 | nb = n % MPW_BITS; |
569 | |
570 | /* --- Handle a shift by a multiple of the word size --- */ |
571 | |
4f29a732 |
572 | if (nb == 0) { |
573 | if (nw >= avl - av) |
574 | MPX_ZERO(dv, dvl); |
575 | else |
576 | MPX_COPY(dv, dvl, av + nw, avl); |
577 | } |
d03ab969 |
578 | |
579 | /* --- And finally the difficult case --- */ |
580 | |
581 | else { |
582 | mpw w; |
583 | size_t nr = MPW_BITS - nb; |
584 | |
585 | av += nw; |
4f29a732 |
586 | w = av < avl ? *av++ : 0; |
d03ab969 |
587 | while (av < avl) { |
588 | mpw t; |
589 | if (dv >= dvl) |
590 | goto done; |
591 | t = *av++; |
592 | *dv++ = MPW((w >> nb) | (t << nr)); |
593 | w = t; |
594 | } |
595 | if (dv < dvl) { |
596 | *dv++ = MPW(w >> nb); |
597 | MPX_ZERO(dv, dvl); |
598 | } |
599 | } |
600 | |
601 | done:; |
602 | } |
603 | |
0f32e0f8 |
604 | /*----- Bitwise operations ------------------------------------------------*/ |
605 | |
f09e814a |
606 | /* --- @mpx_bitop@ --- * |
0f32e0f8 |
607 | * |
608 | * Arguments: @mpw *dv, *dvl@ = destination vector |
609 | * @const mpw *av, *avl@ = first source vector |
610 | * @const mpw *bv, *bvl@ = second source vector |
611 | * |
612 | * Returns: --- |
613 | * |
f09e814a |
614 | * Use; Provides the dyadic boolean functions. |
0f32e0f8 |
615 | */ |
616 | |
f09e814a |
617 | #define MPX_BITBINOP(string) \ |
0f32e0f8 |
618 | \ |
f09e814a |
619 | void mpx_bit##string(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, \ |
620 | const mpw *bv, const mpw *bvl) \ |
0f32e0f8 |
621 | { \ |
622 | MPX_SHRINK(av, avl); \ |
623 | MPX_SHRINK(bv, bvl); \ |
624 | \ |
625 | while (dv < dvl) { \ |
626 | mpw a, b; \ |
627 | a = (av < avl) ? *av++ : 0; \ |
628 | b = (bv < bvl) ? *bv++ : 0; \ |
75263f25 |
629 | *dv++ = B##string(a, b); \ |
0f32e0f8 |
630 | } \ |
631 | } |
632 | |
f09e814a |
633 | MPX_DOBIN(MPX_BITBINOP) |
0f32e0f8 |
634 | |
635 | void mpx_not(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl) |
636 | { |
637 | MPX_SHRINK(av, avl); |
638 | |
639 | while (dv < dvl) { |
640 | mpw a; |
641 | a = (av < avl) ? *av++ : 0; |
642 | *dv++ = ~a; |
643 | } |
644 | } |
645 | |
d03ab969 |
646 | /*----- Unsigned arithmetic -----------------------------------------------*/ |
647 | |
f45a00c6 |
648 | /* --- @mpx_2c@ --- * |
649 | * |
650 | * Arguments: @mpw *dv, *dvl@ = destination vector |
651 | * @const mpw *v, *vl@ = source vector |
652 | * |
653 | * Returns: --- |
654 | * |
655 | * Use: Calculates the two's complement of @v@. |
656 | */ |
657 | |
658 | void mpx_2c(mpw *dv, mpw *dvl, const mpw *v, const mpw *vl) |
659 | { |
660 | mpw c = 0; |
661 | while (dv < dvl && v < vl) |
662 | *dv++ = c = MPW(~*v++); |
663 | if (dv < dvl) { |
664 | if (c > MPW_MAX / 2) |
665 | c = MPW(~0); |
666 | while (dv < dvl) |
667 | *dv++ = c; |
668 | } |
669 | MPX_UADDN(dv, dvl, 1); |
670 | } |
671 | |
1a05a8ef |
672 | /* --- @mpx_ueq@ --- * |
673 | * |
674 | * Arguments: @const mpw *av, *avl@ = first argument vector base and limit |
675 | * @const mpw *bv, *bvl@ = second argument vector base and limit |
676 | * |
677 | * Returns: Nonzero if the two vectors are equal. |
678 | * |
679 | * Use: Performs an unsigned integer test for equality. |
680 | */ |
681 | |
682 | int mpx_ueq(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl) |
683 | { |
684 | MPX_SHRINK(av, avl); |
685 | MPX_SHRINK(bv, bvl); |
686 | if (avl - av != bvl - bv) |
687 | return (0); |
688 | while (av < avl) { |
689 | if (*av++ != *bv++) |
690 | return (0); |
691 | } |
692 | return (1); |
693 | } |
694 | |
d03ab969 |
695 | /* --- @mpx_ucmp@ --- * |
696 | * |
697 | * Arguments: @const mpw *av, *avl@ = first argument vector base and limit |
698 | * @const mpw *bv, *bvl@ = second argument vector base and limit |
699 | * |
700 | * Returns: Less than, equal to, or greater than zero depending on |
701 | * whether @a@ is less than, equal to or greater than @b@, |
702 | * respectively. |
703 | * |
704 | * Use: Performs an unsigned integer comparison. |
705 | */ |
706 | |
707 | int mpx_ucmp(const mpw *av, const mpw *avl, const mpw *bv, const mpw *bvl) |
708 | { |
709 | MPX_SHRINK(av, avl); |
710 | MPX_SHRINK(bv, bvl); |
711 | |
712 | if (avl - av > bvl - bv) |
713 | return (+1); |
714 | else if (avl - av < bvl - bv) |
715 | return (-1); |
716 | else while (avl > av) { |
717 | mpw a = *--avl, b = *--bvl; |
718 | if (a > b) |
719 | return (+1); |
720 | else if (a < b) |
721 | return (-1); |
722 | } |
723 | return (0); |
724 | } |
1a05a8ef |
725 | |
d03ab969 |
726 | /* --- @mpx_uadd@ --- * |
727 | * |
728 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
729 | * @const mpw *av, *avl@ = first addend vector base and limit |
730 | * @const mpw *bv, *bvl@ = second addend vector base and limit |
731 | * |
732 | * Returns: --- |
733 | * |
734 | * Use: Performs unsigned integer addition. If the result overflows |
735 | * the destination vector, high-order bits are discarded. This |
736 | * means that two's complement addition happens more or less for |
737 | * free, although that's more a side-effect than anything else. |
738 | * The result vector may be equal to either or both source |
739 | * vectors, but may not otherwise overlap them. |
740 | */ |
741 | |
742 | void mpx_uadd(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, |
743 | const mpw *bv, const mpw *bvl) |
744 | { |
745 | mpw c = 0; |
746 | |
747 | while (av < avl || bv < bvl) { |
748 | mpw a, b; |
749 | mpd x; |
750 | if (dv >= dvl) |
751 | return; |
752 | a = (av < avl) ? *av++ : 0; |
753 | b = (bv < bvl) ? *bv++ : 0; |
754 | x = (mpd)a + (mpd)b + c; |
755 | *dv++ = MPW(x); |
756 | c = x >> MPW_BITS; |
757 | } |
758 | if (dv < dvl) { |
759 | *dv++ = c; |
760 | MPX_ZERO(dv, dvl); |
761 | } |
762 | } |
763 | |
dd517851 |
764 | /* --- @mpx_uaddn@ --- * |
765 | * |
766 | * Arguments: @mpw *dv, *dvl@ = source and destination base and limit |
767 | * @mpw n@ = other addend |
768 | * |
769 | * Returns: --- |
770 | * |
771 | * Use: Adds a small integer to a multiprecision number. |
772 | */ |
773 | |
774 | void mpx_uaddn(mpw *dv, mpw *dvl, mpw n) { MPX_UADDN(dv, dvl, n); } |
775 | |
d03ab969 |
776 | /* --- @mpx_usub@ --- * |
777 | * |
778 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
779 | * @const mpw *av, *avl@ = first argument vector base and limit |
780 | * @const mpw *bv, *bvl@ = second argument vector base and limit |
781 | * |
782 | * Returns: --- |
783 | * |
784 | * Use: Performs unsigned integer subtraction. If the result |
785 | * overflows the destination vector, high-order bits are |
786 | * discarded. This means that two's complement subtraction |
787 | * happens more or less for free, althuogh that's more a side- |
788 | * effect than anything else. The result vector may be equal to |
789 | * either or both source vectors, but may not otherwise overlap |
790 | * them. |
791 | */ |
792 | |
793 | void mpx_usub(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, |
794 | const mpw *bv, const mpw *bvl) |
795 | { |
796 | mpw c = 0; |
797 | |
798 | while (av < avl || bv < bvl) { |
799 | mpw a, b; |
800 | mpd x; |
801 | if (dv >= dvl) |
802 | return; |
803 | a = (av < avl) ? *av++ : 0; |
804 | b = (bv < bvl) ? *bv++ : 0; |
c8a2f9ef |
805 | x = (mpd)a - (mpd)b - c; |
d03ab969 |
806 | *dv++ = MPW(x); |
c8a2f9ef |
807 | if (x >> MPW_BITS) |
808 | c = 1; |
809 | else |
810 | c = 0; |
d03ab969 |
811 | } |
c8a2f9ef |
812 | if (c) |
813 | c = MPW_MAX; |
d03ab969 |
814 | while (dv < dvl) |
c8a2f9ef |
815 | *dv++ = c; |
d03ab969 |
816 | } |
817 | |
dd517851 |
818 | /* --- @mpx_usubn@ --- * |
819 | * |
820 | * Arguments: @mpw *dv, *dvl@ = source and destination base and limit |
821 | * @n@ = subtrahend |
822 | * |
823 | * Returns: --- |
824 | * |
825 | * Use: Subtracts a small integer from a multiprecision number. |
826 | */ |
827 | |
828 | void mpx_usubn(mpw *dv, mpw *dvl, mpw n) { MPX_USUBN(dv, dvl, n); } |
829 | |
d03ab969 |
830 | /* --- @mpx_umul@ --- * |
831 | * |
832 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
833 | * @const mpw *av, *avl@ = multiplicand vector base and limit |
834 | * @const mpw *bv, *bvl@ = multiplier vector base and limit |
835 | * |
836 | * Returns: --- |
837 | * |
838 | * Use: Performs unsigned integer multiplication. If the result |
839 | * overflows the desination vector, high-order bits are |
840 | * discarded. The result vector may not overlap the argument |
841 | * vectors in any way. |
842 | */ |
843 | |
844 | void mpx_umul(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, |
845 | const mpw *bv, const mpw *bvl) |
846 | { |
847 | /* --- This is probably worthwhile on a multiply --- */ |
848 | |
849 | MPX_SHRINK(av, avl); |
850 | MPX_SHRINK(bv, bvl); |
851 | |
852 | /* --- Deal with a multiply by zero --- */ |
853 | |
854 | if (bv == bvl) { |
c8a2f9ef |
855 | MPX_ZERO(dv, dvl); |
d03ab969 |
856 | return; |
857 | } |
858 | |
859 | /* --- Do the initial multiply and initialize the accumulator --- */ |
860 | |
861 | MPX_UMULN(dv, dvl, av, avl, *bv++); |
862 | |
863 | /* --- Do the remaining multiply/accumulates --- */ |
864 | |
c8a2f9ef |
865 | while (dv < dvl && bv < bvl) { |
d03ab969 |
866 | mpw m = *bv++; |
c8a2f9ef |
867 | mpw c = 0; |
d03ab969 |
868 | const mpw *avv = av; |
869 | mpw *dvv = ++dv; |
870 | |
871 | while (avv < avl) { |
872 | mpd x; |
873 | if (dvv >= dvl) |
874 | goto next; |
c8a2f9ef |
875 | x = (mpd)*dvv + (mpd)m * (mpd)*avv++ + c; |
876 | *dvv++ = MPW(x); |
d03ab969 |
877 | c = x >> MPW_BITS; |
878 | } |
c8a2f9ef |
879 | MPX_UADDN(dvv, dvl, c); |
d03ab969 |
880 | next:; |
881 | } |
882 | } |
883 | |
dd517851 |
884 | /* --- @mpx_umuln@ --- * |
885 | * |
886 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
887 | * @const mpw *av, *avl@ = multiplicand vector base and limit |
888 | * @mpw m@ = multiplier |
889 | * |
890 | * Returns: --- |
891 | * |
892 | * Use: Multiplies a multiprecision integer by a single-word value. |
893 | * The destination and source may be equal. The destination |
894 | * is completely cleared after use. |
895 | */ |
896 | |
897 | void mpx_umuln(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m) |
898 | { |
899 | MPX_UMULN(dv, dvl, av, avl, m); |
900 | } |
901 | |
902 | /* --- @mpx_umlan@ --- * |
903 | * |
904 | * Arguments: @mpw *dv, *dvl@ = destination/accumulator base and limit |
905 | * @const mpw *av, *avl@ = multiplicand vector base and limit |
906 | * @mpw m@ = multiplier |
907 | * |
908 | * Returns: --- |
909 | * |
910 | * Use: Multiplies a multiprecision integer by a single-word value |
911 | * and adds the result to an accumulator. |
912 | */ |
913 | |
914 | void mpx_umlan(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl, mpw m) |
915 | { |
916 | MPX_UMLAN(dv, dvl, av, avl, m); |
917 | } |
918 | |
c8a2f9ef |
919 | /* --- @mpx_usqr@ --- * |
920 | * |
921 | * Arguments: @mpw *dv, *dvl@ = destination vector base and limit |
922 | * @const mpw *av, *av@ = source vector base and limit |
923 | * |
924 | * Returns: --- |
925 | * |
926 | * Use: Performs unsigned integer squaring. The result vector must |
927 | * not overlap the source vector in any way. |
928 | */ |
929 | |
930 | void mpx_usqr(mpw *dv, mpw *dvl, const mpw *av, const mpw *avl) |
931 | { |
932 | MPX_ZERO(dv, dvl); |
933 | |
934 | /* --- Main loop --- */ |
935 | |
936 | while (av < avl) { |
937 | const mpw *avv = av; |
938 | mpw *dvv = dv; |
939 | mpw a = *av; |
940 | mpd c; |
941 | |
942 | /* --- Stop if I've run out of destination --- */ |
943 | |
944 | if (dvv >= dvl) |
945 | break; |
946 | |
947 | /* --- Work out the square at this point in the proceedings --- */ |
948 | |
949 | { |
c8a2f9ef |
950 | mpd x = (mpd)a * (mpd)a + *dvv; |
951 | *dvv++ = MPW(x); |
952 | c = MPW(x >> MPW_BITS); |
953 | } |
954 | |
955 | /* --- Now fix up the rest of the vector upwards --- */ |
956 | |
957 | avv++; |
958 | while (dvv < dvl && avv < avl) { |
c8a2f9ef |
959 | mpd x = (mpd)a * (mpd)*avv++; |
960 | mpd y = ((x << 1) & MPW_MAX) + c + *dvv; |
961 | c = (x >> (MPW_BITS - 1)) + (y >> MPW_BITS); |
962 | *dvv++ = MPW(y); |
963 | } |
964 | while (dvv < dvl && c) { |
965 | mpd x = c + *dvv; |
966 | *dvv++ = MPW(x); |
967 | c = x >> MPW_BITS; |
968 | } |
969 | |
970 | /* --- Get ready for the next round --- */ |
971 | |
972 | av++; |
973 | dv += 2; |
974 | } |
975 | } |
976 | |
d03ab969 |
977 | /* --- @mpx_udiv@ --- * |
978 | * |
979 | * Arguments: @mpw *qv, *qvl@ = quotient vector base and limit |
980 | * @mpw *rv, *rvl@ = dividend/remainder vector base and limit |
981 | * @const mpw *dv, *dvl@ = divisor vector base and limit |
c8a2f9ef |
982 | * @mpw *sv, *svl@ = scratch workspace |
d03ab969 |
983 | * |
984 | * Returns: --- |
985 | * |
986 | * Use: Performs unsigned integer division. If the result overflows |
987 | * the quotient vector, high-order bits are discarded. (Clearly |
988 | * the remainder vector can't overflow.) The various vectors |
989 | * may not overlap in any way. Yes, I know it's a bit odd |
990 | * requiring the dividend to be in the result position but it |
991 | * does make some sense really. The remainder must have |
c8a2f9ef |
992 | * headroom for at least two extra words. The scratch space |
f45a00c6 |
993 | * must be at least one word larger than the divisor. |
d03ab969 |
994 | */ |
995 | |
996 | void mpx_udiv(mpw *qv, mpw *qvl, mpw *rv, mpw *rvl, |
c8a2f9ef |
997 | const mpw *dv, const mpw *dvl, |
998 | mpw *sv, mpw *svl) |
d03ab969 |
999 | { |
d03ab969 |
1000 | unsigned norm = 0; |
1001 | size_t scale; |
1002 | mpw d, dd; |
1003 | |
1004 | /* --- Initialize the quotient --- */ |
1005 | |
1006 | MPX_ZERO(qv, qvl); |
1007 | |
c8a2f9ef |
1008 | /* --- Perform some sanity checks --- */ |
1009 | |
1010 | MPX_SHRINK(dv, dvl); |
1011 | assert(((void)"division by zero in mpx_udiv", dv < dvl)); |
1012 | |
d03ab969 |
1013 | /* --- Normalize the divisor --- * |
1014 | * |
1015 | * The algorithm requires that the divisor be at least two digits long. |
1016 | * This is easy to fix. |
1017 | */ |
1018 | |
c8a2f9ef |
1019 | { |
1020 | unsigned b; |
d03ab969 |
1021 | |
c8a2f9ef |
1022 | d = dvl[-1]; |
1023 | for (b = MPW_BITS / 2; b; b >>= 1) { |
1024 | if (d < (MPW_MAX >> b)) { |
1025 | d <<= b; |
1026 | norm += b; |
1027 | } |
1028 | } |
1029 | if (dv + 1 == dvl) |
1030 | norm += MPW_BITS; |
d03ab969 |
1031 | } |
d03ab969 |
1032 | |
1033 | /* --- Normalize the dividend/remainder to match --- */ |
1034 | |
c8a2f9ef |
1035 | if (norm) { |
c8a2f9ef |
1036 | mpx_lsl(rv, rvl, rv, rvl, norm); |
f45a00c6 |
1037 | mpx_lsl(sv, svl, dv, dvl, norm); |
c8a2f9ef |
1038 | dv = sv; |
f45a00c6 |
1039 | dvl = svl; |
c8a2f9ef |
1040 | MPX_SHRINK(dv, dvl); |
1041 | } |
1042 | |
d03ab969 |
1043 | MPX_SHRINK(rv, rvl); |
c8a2f9ef |
1044 | d = dvl[-1]; |
1045 | dd = dvl[-2]; |
d03ab969 |
1046 | |
1047 | /* --- Work out the relative scales --- */ |
1048 | |
1049 | { |
1050 | size_t rvn = rvl - rv; |
c8a2f9ef |
1051 | size_t dvn = dvl - dv; |
d03ab969 |
1052 | |
1053 | /* --- If the divisor is clearly larger, notice this --- */ |
1054 | |
1055 | if (dvn > rvn) { |
1056 | mpx_lsr(rv, rvl, rv, rvl, norm); |
1057 | return; |
1058 | } |
1059 | |
1060 | scale = rvn - dvn; |
1061 | } |
1062 | |
1063 | /* --- Calculate the most significant quotient digit --- * |
1064 | * |
1065 | * Because the divisor has its top bit set, this can only happen once. The |
1066 | * pointer arithmetic is a little contorted, to make sure that the |
1067 | * behaviour is defined. |
1068 | */ |
1069 | |
1070 | if (MPX_UCMP(rv + scale, rvl, >=, dv, dvl)) { |
1071 | mpx_usub(rv + scale, rvl, rv + scale, rvl, dv, dvl); |
1072 | if (qvl - qv > scale) |
1073 | qv[scale] = 1; |
1074 | } |
1075 | |
1076 | /* --- Now for the main loop --- */ |
1077 | |
1078 | { |
c8a2f9ef |
1079 | mpw *rvv = rvl - 2; |
d03ab969 |
1080 | |
1081 | while (scale) { |
c8a2f9ef |
1082 | mpw q; |
1083 | mpd rh; |
d03ab969 |
1084 | |
1085 | /* --- Get an estimate for the next quotient digit --- */ |
1086 | |
c8a2f9ef |
1087 | mpw r = rvv[1]; |
1088 | mpw rr = rvv[0]; |
1089 | mpw rrr = *--rvv; |
1090 | |
1091 | scale--; |
1092 | rh = ((mpd)r << MPW_BITS) | rr; |
d03ab969 |
1093 | if (r == d) |
1094 | q = MPW_MAX; |
c8a2f9ef |
1095 | else |
1096 | q = MPW(rh / d); |
d03ab969 |
1097 | |
1098 | /* --- Refine the estimate --- */ |
1099 | |
1100 | { |
1101 | mpd yh = (mpd)d * q; |
ce76ff16 |
1102 | mpd yy = (mpd)dd * q; |
1103 | mpw yl; |
c8a2f9ef |
1104 | |
ce76ff16 |
1105 | if (yy > MPW_MAX) |
1106 | yh += yy >> MPW_BITS; |
1107 | yl = MPW(yy); |
c8a2f9ef |
1108 | |
1109 | while (yh > rh || (yh == rh && yl > rrr)) { |
1110 | q--; |
1111 | yh -= d; |
ce76ff16 |
1112 | if (yl < dd) |
1113 | yh--; |
99b30c23 |
1114 | yl = MPW(yl - dd); |
c8a2f9ef |
1115 | } |
1116 | } |
1117 | |
1118 | /* --- Remove a chunk from the dividend --- */ |
1119 | |
1120 | { |
1121 | mpw *svv; |
1122 | const mpw *dvv; |
f45a00c6 |
1123 | mpw mc = 0, sc = 0; |
c8a2f9ef |
1124 | |
f45a00c6 |
1125 | /* --- Calculate the size of the chunk --- * |
1126 | * |
1127 | * This does the whole job of calculating @r >> scale - qd@. |
1128 | */ |
c8a2f9ef |
1129 | |
f45a00c6 |
1130 | for (svv = rv + scale, dvv = dv; |
1131 | dvv < dvl && svv < rvl; |
1132 | svv++, dvv++) { |
1133 | mpd x = (mpd)*dvv * (mpd)q + mc; |
1134 | mc = x >> MPW_BITS; |
1135 | x = (mpd)*svv - MPW(x) - sc; |
c8a2f9ef |
1136 | *svv = MPW(x); |
f45a00c6 |
1137 | if (x >> MPW_BITS) |
1138 | sc = 1; |
1139 | else |
1140 | sc = 0; |
1141 | } |
1142 | |
1143 | if (svv < rvl) { |
1144 | mpd x = (mpd)*svv - mc - sc; |
1145 | *svv++ = MPW(x); |
1146 | if (x >> MPW_BITS) |
1147 | sc = MPW_MAX; |
1148 | else |
1149 | sc = 0; |
1150 | while (svv < rvl) |
1151 | *svv++ = sc; |
c8a2f9ef |
1152 | } |
c8a2f9ef |
1153 | |
f45a00c6 |
1154 | /* --- Fix if the quotient was too large --- * |
c8a2f9ef |
1155 | * |
f45a00c6 |
1156 | * This doesn't seem to happen very often. |
c8a2f9ef |
1157 | */ |
1158 | |
c8a2f9ef |
1159 | if (rvl[-1] > MPW_MAX / 2) { |
1160 | mpx_uadd(rv + scale, rvl, rv + scale, rvl, dv, dvl); |
1161 | q--; |
1162 | } |
1163 | } |
1164 | |
1165 | /* --- Done for another iteration --- */ |
1166 | |
1167 | if (qvl - qv > scale) |
1168 | qv[scale] = q; |
1169 | r = rr; |
1170 | rr = rrr; |
1171 | } |
1172 | } |
1173 | |
1174 | /* --- Now fiddle with unnormalizing and things --- */ |
1175 | |
1176 | mpx_lsr(rv, rvl, rv, rvl, norm); |
d03ab969 |
1177 | } |
1178 | |
698bd937 |
1179 | /* --- @mpx_udivn@ --- * |
1180 | * |
1181 | * Arguments: @mpw *qv, *qvl@ = storage for the quotient (may overlap |
1182 | * dividend) |
1183 | * @const mpw *rv, *rvl@ = dividend |
1184 | * @mpw d@ = single-precision divisor |
1185 | * |
1186 | * Returns: Remainder after divison. |
1187 | * |
1188 | * Use: Performs a single-precision division operation. |
1189 | */ |
1190 | |
1191 | mpw mpx_udivn(mpw *qv, mpw *qvl, const mpw *rv, const mpw *rvl, mpw d) |
1192 | { |
1193 | size_t i; |
1194 | size_t ql = qvl - qv; |
1195 | mpd r = 0; |
1196 | |
1197 | i = rvl - rv; |
1198 | while (i > 0) { |
1199 | i--; |
1200 | r = (r << MPW_BITS) | rv[i]; |
1201 | if (i < ql) |
1202 | qv[i] = r / d; |
1203 | r %= d; |
1204 | } |
1205 | return (MPW(r)); |
1206 | } |
1207 | |
42684bdb |
1208 | /*----- Test rig ----------------------------------------------------------*/ |
1209 | |
1210 | #ifdef TEST_RIG |
1211 | |
1212 | #include <mLib/alloc.h> |
1213 | #include <mLib/dstr.h> |
1214 | #include <mLib/quis.h> |
1215 | #include <mLib/testrig.h> |
1216 | |
1217 | #include "mpscan.h" |
1218 | |
1219 | #define ALLOC(v, vl, sz) do { \ |
1220 | size_t _sz = (sz); \ |
1221 | mpw *_vv = xmalloc(MPWS(_sz)); \ |
1222 | mpw *_vvl = _vv + _sz; \ |
1223 | (v) = _vv; \ |
1224 | (vl) = _vvl; \ |
1225 | } while (0) |
1226 | |
1227 | #define LOAD(v, vl, d) do { \ |
1228 | const dstr *_d = (d); \ |
1229 | mpw *_v, *_vl; \ |
1230 | ALLOC(_v, _vl, MPW_RQ(_d->len)); \ |
1231 | mpx_loadb(_v, _vl, _d->buf, _d->len); \ |
1232 | (v) = _v; \ |
1233 | (vl) = _vl; \ |
1234 | } while (0) |
1235 | |
1236 | #define MAX(x, y) ((x) > (y) ? (x) : (y)) |
1237 | |
1238 | static void dumpbits(const char *msg, const void *pp, size_t sz) |
1239 | { |
1240 | const octet *p = pp; |
1241 | fputs(msg, stderr); |
1242 | for (; sz; sz--) |
1243 | fprintf(stderr, " %02x", *p++); |
1244 | fputc('\n', stderr); |
1245 | } |
1246 | |
1247 | static void dumpmp(const char *msg, const mpw *v, const mpw *vl) |
1248 | { |
1249 | fputs(msg, stderr); |
1250 | MPX_SHRINK(v, vl); |
1251 | while (v < vl) |
1252 | fprintf(stderr, " %08lx", (unsigned long)*--vl); |
1253 | fputc('\n', stderr); |
1254 | } |
1255 | |
1256 | static int chkscan(const mpw *v, const mpw *vl, |
1257 | const void *pp, size_t sz, int step) |
1258 | { |
1259 | mpscan mps; |
1260 | const octet *p = pp; |
1261 | unsigned bit = 0; |
1262 | int ok = 1; |
1263 | |
1264 | mpscan_initx(&mps, v, vl); |
1265 | while (sz) { |
1266 | unsigned x = *p; |
1267 | int i; |
1268 | p += step; |
1269 | for (i = 0; i < 8 && MPSCAN_STEP(&mps); i++) { |
1270 | if (MPSCAN_BIT(&mps) != (x & 1)) { |
1271 | fprintf(stderr, |
1272 | "\n*** error, step %i, bit %u, expected %u, found %u\n", |
1273 | step, bit, x & 1, MPSCAN_BIT(&mps)); |
1274 | ok = 0; |
1275 | } |
1276 | x >>= 1; |
1277 | bit++; |
1278 | } |
1279 | sz--; |
1280 | } |
1281 | |
1282 | return (ok); |
1283 | } |
1284 | |
1285 | static int loadstore(dstr *v) |
1286 | { |
1287 | dstr d = DSTR_INIT; |
1288 | size_t sz = MPW_RQ(v->len) * 2, diff; |
1289 | mpw *m, *ml; |
1290 | int ok = 1; |
1291 | |
1292 | dstr_ensure(&d, v->len); |
1293 | m = xmalloc(MPWS(sz)); |
1294 | |
1295 | for (diff = 0; diff < sz; diff += 5) { |
1296 | size_t oct; |
1297 | |
1298 | ml = m + sz - diff; |
1299 | |
1300 | mpx_loadl(m, ml, v->buf, v->len); |
1301 | if (!chkscan(m, ml, v->buf, v->len, +1)) |
1302 | ok = 0; |
1303 | MPX_OCTETS(oct, m, ml); |
1304 | mpx_storel(m, ml, d.buf, d.sz); |
1305 | if (memcmp(d.buf, v->buf, oct) != 0) { |
1306 | dumpbits("\n*** storel failed", d.buf, d.sz); |
1307 | ok = 0; |
1308 | } |
1309 | |
1310 | mpx_loadb(m, ml, v->buf, v->len); |
1311 | if (!chkscan(m, ml, v->buf + v->len - 1, v->len, -1)) |
1312 | ok = 0; |
1313 | MPX_OCTETS(oct, m, ml); |
1314 | mpx_storeb(m, ml, d.buf, d.sz); |
1315 | if (memcmp(d.buf + d.sz - oct, v->buf + v->len - oct, oct) != 0) { |
1316 | dumpbits("\n*** storeb failed", d.buf, d.sz); |
1317 | ok = 0; |
1318 | } |
1319 | } |
1320 | |
1321 | if (!ok) |
1322 | dumpbits("input data", v->buf, v->len); |
1323 | |
1324 | free(m); |
1325 | dstr_destroy(&d); |
1326 | return (ok); |
1327 | } |
1328 | |
f09e814a |
1329 | static int twocl(dstr *v) |
1330 | { |
1331 | dstr d = DSTR_INIT; |
1332 | mpw *m, *ml; |
1333 | size_t sz; |
1334 | int ok = 1; |
1335 | |
1336 | sz = v[0].len; if (v[1].len > sz) sz = v[1].len; |
1337 | dstr_ensure(&d, sz); |
1338 | |
1339 | sz = MPW_RQ(sz); |
1340 | m = xmalloc(MPWS(sz)); |
1341 | ml = m + sz; |
1342 | |
1343 | mpx_loadl(m, ml, v[0].buf, v[0].len); |
1344 | mpx_storel2cn(m, ml, d.buf, v[1].len); |
1345 | if (memcmp(d.buf, v[1].buf, v[1].len)) { |
1346 | dumpbits("\n*** storel2cn failed", d.buf, v[1].len); |
1347 | ok = 0; |
1348 | } |
1349 | |
1350 | mpx_loadl2cn(m, ml, v[1].buf, v[1].len); |
1351 | mpx_storel(m, ml, d.buf, v[0].len); |
1352 | if (memcmp(d.buf, v[0].buf, v[0].len)) { |
1353 | dumpbits("\n*** loadl2cn failed", d.buf, v[0].len); |
1354 | ok = 0; |
1355 | } |
1356 | |
1357 | if (!ok) { |
1358 | dumpbits("pos", v[0].buf, v[0].len); |
1359 | dumpbits("neg", v[1].buf, v[1].len); |
1360 | } |
1361 | |
1362 | free(m); |
1363 | dstr_destroy(&d); |
1364 | |
1365 | return (ok); |
1366 | } |
1367 | |
1368 | static int twocb(dstr *v) |
1369 | { |
1370 | dstr d = DSTR_INIT; |
1371 | mpw *m, *ml; |
1372 | size_t sz; |
1373 | int ok = 1; |
1374 | |
1375 | sz = v[0].len; if (v[1].len > sz) sz = v[1].len; |
1376 | dstr_ensure(&d, sz); |
1377 | |
1378 | sz = MPW_RQ(sz); |
1379 | m = xmalloc(MPWS(sz)); |
1380 | ml = m + sz; |
1381 | |
1382 | mpx_loadb(m, ml, v[0].buf, v[0].len); |
1383 | mpx_storeb2cn(m, ml, d.buf, v[1].len); |
1384 | if (memcmp(d.buf, v[1].buf, v[1].len)) { |
1385 | dumpbits("\n*** storeb2cn failed", d.buf, v[1].len); |
1386 | ok = 0; |
1387 | } |
1388 | |
1389 | mpx_loadb2cn(m, ml, v[1].buf, v[1].len); |
1390 | mpx_storeb(m, ml, d.buf, v[0].len); |
1391 | if (memcmp(d.buf, v[0].buf, v[0].len)) { |
1392 | dumpbits("\n*** loadb2cn failed", d.buf, v[0].len); |
1393 | ok = 0; |
1394 | } |
1395 | |
1396 | if (!ok) { |
1397 | dumpbits("pos", v[0].buf, v[0].len); |
1398 | dumpbits("neg", v[1].buf, v[1].len); |
1399 | } |
1400 | |
1401 | free(m); |
1402 | dstr_destroy(&d); |
1403 | |
1404 | return (ok); |
1405 | } |
1406 | |
42684bdb |
1407 | static int lsl(dstr *v) |
1408 | { |
1409 | mpw *a, *al; |
1410 | int n = *(int *)v[1].buf; |
1411 | mpw *c, *cl; |
1412 | mpw *d, *dl; |
1413 | int ok = 1; |
1414 | |
1415 | LOAD(a, al, &v[0]); |
1416 | LOAD(c, cl, &v[2]); |
1417 | ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS); |
1418 | |
1419 | mpx_lsl(d, dl, a, al, n); |
1a05a8ef |
1420 | if (!mpx_ueq(d, dl, c, cl)) { |
42684bdb |
1421 | fprintf(stderr, "\n*** lsl(%i) failed\n", n); |
1422 | dumpmp(" a", a, al); |
1423 | dumpmp("expected", c, cl); |
1424 | dumpmp(" result", d, dl); |
1425 | ok = 0; |
1426 | } |
1427 | |
1428 | free(a); free(c); free(d); |
1429 | return (ok); |
1430 | } |
1431 | |
1432 | static int lsr(dstr *v) |
1433 | { |
1434 | mpw *a, *al; |
1435 | int n = *(int *)v[1].buf; |
1436 | mpw *c, *cl; |
1437 | mpw *d, *dl; |
1438 | int ok = 1; |
1439 | |
1440 | LOAD(a, al, &v[0]); |
1441 | LOAD(c, cl, &v[2]); |
1442 | ALLOC(d, dl, al - a + (n + MPW_BITS - 1) / MPW_BITS + 1); |
1443 | |
1444 | mpx_lsr(d, dl, a, al, n); |
1a05a8ef |
1445 | if (!mpx_ueq(d, dl, c, cl)) { |
42684bdb |
1446 | fprintf(stderr, "\n*** lsr(%i) failed\n", n); |
1447 | dumpmp(" a", a, al); |
1448 | dumpmp("expected", c, cl); |
1449 | dumpmp(" result", d, dl); |
1450 | ok = 0; |
1451 | } |
1452 | |
1453 | free(a); free(c); free(d); |
1454 | return (ok); |
1455 | } |
1456 | |
1457 | static int uadd(dstr *v) |
1458 | { |
1459 | mpw *a, *al; |
1460 | mpw *b, *bl; |
1461 | mpw *c, *cl; |
1462 | mpw *d, *dl; |
1463 | int ok = 1; |
1464 | |
1465 | LOAD(a, al, &v[0]); |
1466 | LOAD(b, bl, &v[1]); |
1467 | LOAD(c, cl, &v[2]); |
1468 | ALLOC(d, dl, MAX(al - a, bl - b) + 1); |
1469 | |
1470 | mpx_uadd(d, dl, a, al, b, bl); |
1a05a8ef |
1471 | if (!mpx_ueq(d, dl, c, cl)) { |
42684bdb |
1472 | fprintf(stderr, "\n*** uadd failed\n"); |
1473 | dumpmp(" a", a, al); |
1474 | dumpmp(" b", b, bl); |
1475 | dumpmp("expected", c, cl); |
1476 | dumpmp(" result", d, dl); |
1477 | ok = 0; |
1478 | } |
1479 | |
1480 | free(a); free(b); free(c); free(d); |
1481 | return (ok); |
1482 | } |
1483 | |
1484 | static int usub(dstr *v) |
1485 | { |
1486 | mpw *a, *al; |
1487 | mpw *b, *bl; |
1488 | mpw *c, *cl; |
1489 | mpw *d, *dl; |
1490 | int ok = 1; |
1491 | |
1492 | LOAD(a, al, &v[0]); |
1493 | LOAD(b, bl, &v[1]); |
1494 | LOAD(c, cl, &v[2]); |
1495 | ALLOC(d, dl, al - a); |
1496 | |
1497 | mpx_usub(d, dl, a, al, b, bl); |
1a05a8ef |
1498 | if (!mpx_ueq(d, dl, c, cl)) { |
42684bdb |
1499 | fprintf(stderr, "\n*** usub failed\n"); |
1500 | dumpmp(" a", a, al); |
1501 | dumpmp(" b", b, bl); |
1502 | dumpmp("expected", c, cl); |
1503 | dumpmp(" result", d, dl); |
1504 | ok = 0; |
1505 | } |
1506 | |
1507 | free(a); free(b); free(c); free(d); |
1508 | return (ok); |
1509 | } |
1510 | |
1511 | static int umul(dstr *v) |
1512 | { |
1513 | mpw *a, *al; |
1514 | mpw *b, *bl; |
1515 | mpw *c, *cl; |
1516 | mpw *d, *dl; |
1517 | int ok = 1; |
1518 | |
1519 | LOAD(a, al, &v[0]); |
1520 | LOAD(b, bl, &v[1]); |
1521 | LOAD(c, cl, &v[2]); |
1522 | ALLOC(d, dl, (al - a) + (bl - b)); |
1523 | |
1524 | mpx_umul(d, dl, a, al, b, bl); |
1a05a8ef |
1525 | if (!mpx_ueq(d, dl, c, cl)) { |
42684bdb |
1526 | fprintf(stderr, "\n*** umul failed\n"); |
1527 | dumpmp(" a", a, al); |
1528 | dumpmp(" b", b, bl); |
1529 | dumpmp("expected", c, cl); |
1530 | dumpmp(" result", d, dl); |
1531 | ok = 0; |
1532 | } |
1533 | |
1534 | free(a); free(b); free(c); free(d); |
1535 | return (ok); |
1536 | } |
1537 | |
1538 | static int usqr(dstr *v) |
1539 | { |
1540 | mpw *a, *al; |
1541 | mpw *c, *cl; |
1542 | mpw *d, *dl; |
1543 | int ok = 1; |
1544 | |
1545 | LOAD(a, al, &v[0]); |
1546 | LOAD(c, cl, &v[1]); |
1547 | ALLOC(d, dl, 2 * (al - a)); |
1548 | |
1549 | mpx_usqr(d, dl, a, al); |
1a05a8ef |
1550 | if (!mpx_ueq(d, dl, c, cl)) { |
42684bdb |
1551 | fprintf(stderr, "\n*** usqr failed\n"); |
1552 | dumpmp(" a", a, al); |
1553 | dumpmp("expected", c, cl); |
1554 | dumpmp(" result", d, dl); |
1555 | ok = 0; |
1556 | } |
1557 | |
1558 | free(a); free(c); free(d); |
1559 | return (ok); |
1560 | } |
1561 | |
1562 | static int udiv(dstr *v) |
1563 | { |
1564 | mpw *a, *al; |
1565 | mpw *b, *bl; |
1566 | mpw *q, *ql; |
1567 | mpw *r, *rl; |
1568 | mpw *qq, *qql; |
1569 | mpw *s, *sl; |
1570 | int ok = 1; |
1571 | |
1572 | ALLOC(a, al, MPW_RQ(v[0].len) + 2); mpx_loadb(a, al, v[0].buf, v[0].len); |
1573 | LOAD(b, bl, &v[1]); |
1574 | LOAD(q, ql, &v[2]); |
1575 | LOAD(r, rl, &v[3]); |
1576 | ALLOC(qq, qql, al - a); |
1577 | ALLOC(s, sl, (bl - b) + 1); |
1578 | |
1579 | mpx_udiv(qq, qql, a, al, b, bl, s, sl); |
1a05a8ef |
1580 | if (!mpx_ueq(qq, qql, q, ql) || |
1581 | !mpx_ueq(a, al, r, rl)) { |
42684bdb |
1582 | fprintf(stderr, "\n*** udiv failed\n"); |
1583 | dumpmp(" divisor", b, bl); |
1584 | dumpmp("expect r", r, rl); |
1585 | dumpmp("result r", a, al); |
1586 | dumpmp("expect q", q, ql); |
1587 | dumpmp("result q", qq, qql); |
1588 | ok = 0; |
1589 | } |
1590 | |
1591 | free(a); free(b); free(r); free(q); free(s); free(qq); |
1592 | return (ok); |
1593 | } |
1594 | |
1595 | static test_chunk defs[] = { |
1596 | { "load-store", loadstore, { &type_hex, 0 } }, |
f09e814a |
1597 | { "2cl", twocl, { &type_hex, &type_hex, } }, |
1598 | { "2cb", twocb, { &type_hex, &type_hex, } }, |
42684bdb |
1599 | { "lsl", lsl, { &type_hex, &type_int, &type_hex, 0 } }, |
1600 | { "lsr", lsr, { &type_hex, &type_int, &type_hex, 0 } }, |
1601 | { "uadd", uadd, { &type_hex, &type_hex, &type_hex, 0 } }, |
1602 | { "usub", usub, { &type_hex, &type_hex, &type_hex, 0 } }, |
1603 | { "umul", umul, { &type_hex, &type_hex, &type_hex, 0 } }, |
1604 | { "usqr", usqr, { &type_hex, &type_hex, 0 } }, |
1605 | { "udiv", udiv, { &type_hex, &type_hex, &type_hex, &type_hex, 0 } }, |
1606 | { 0, 0, { 0 } } |
1607 | }; |
1608 | |
1609 | int main(int argc, char *argv[]) |
1610 | { |
1611 | test_run(argc, argv, defs, SRCDIR"/tests/mpx"); |
1612 | return (0); |
1613 | } |
1614 | |
42684bdb |
1615 | #endif |
1616 | |
d03ab969 |
1617 | /*----- That's all, folks -------------------------------------------------*/ |