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