projects / u / mdw / catacomb / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
mpreduce: Don't stop bit scanner too early.
[u/mdw/catacomb] / mpreduce.c
CommitLineData
f46efa79 1/* -*-c-*-
2 *
a69a3efd 3 * $Id$
f46efa79 4 *
5 * Efficient reduction modulo nice primes
6 *
7 * (c) 2004 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
f46efa79 30/*----- Header files ------------------------------------------------------*/
31
32#include <mLib/darray.h>
33#include <mLib/macros.h>
34
35#include "mp.h"
36#include "mpreduce.h"
37#include "mpreduce-exp.h"
38
39/*----- Data structures ---------------------------------------------------*/
40
41DA_DECL(instr_v, mpreduce_instr);
42
43/*----- Main code ---------------------------------------------------------*/
44
45/* --- @mpreduce_create@ --- *
46 *
47 * Arguments: @gfreduce *r@ = structure to fill in
48 * @mp *x@ = an integer
49 *
f4535c64 50 * Returns: Zero if successful; nonzero on failure.
f46efa79 51 *
52 * Use: Initializes a context structure for reduction.
53 */
54
f4535c64 55int mpreduce_create(mpreduce *r, mp *p)
f46efa79 56{
57 mpscan sc;
58 enum { Z = 0, Z1 = 2, X = 4, X0 = 6 };
59 unsigned st = Z;
60 instr_v iv = DA_INIT;
61 unsigned long d, i;
62 unsigned op;
63 size_t w, b;
64
65 /* --- Fill in the easy stuff --- */
66
f4535c64 67 if (!MP_POSP(p))
68 return (-1);
f46efa79 69 d = mp_bits(p);
70 r->lim = d/MPW_BITS;
71 r->s = d%MPW_BITS;
72 if (r->s)
73 r->lim++;
74 r->p = mp_copy(p);
75
76 /* --- Stash a new instruction --- */
77
78#define INSTR(op_, argx_, argy_) do { \
79 DA_ENSURE(&iv, 1); \
80 DA(&iv)[DA_LEN(&iv)].op = (op_); \
81 DA(&iv)[DA_LEN(&iv)].argx = (argx_); \
82 DA(&iv)[DA_LEN(&iv)].argy = (argy_); \
83 DA_EXTEND(&iv, 1); \
84} while (0)
85
86 /* --- Main loop --- *
87 *
88 * A simple state machine decomposes @p@ conveniently into positive and
89 * negative powers of 2. The pure form of the state machine is left below
90 * for reference (and in case I need inspiration for a NAF exponentiator).
91 */
92
93#ifdef DEBUG
94 for (i = 0, mp_scan(&sc, p); mp_step(&sc); i++) {
95 switch (st | mp_bit(&sc)) {
96 case Z | 1: st = Z1; break;
97 case Z1 | 0: st = Z; printf("+ %lu\n", i - 1); break;
98 case Z1 | 1: st = X; printf("- %lu\n", i - 1); break;
99 case X | 0: st = X0; break;
100 case X0 | 1: st = X; printf("- %lu\n", i - 1); break;
101 case X0 | 0: st = Z; printf("+ %lu\n", i - 1); break;
102 }
103 }
f9c435c0
MW		 104 if (st >= X) printf("+ %lu\n", i - 1);
105 st = Z;
f46efa79 106#endif
107
bccb92dd 108 for (i = 0, mp_scan(&sc, p); i < d && mp_step(&sc); i++) {
f46efa79 109 switch (st | mp_bit(&sc)) {
110 case Z | 1: st = Z1; break;
111 case Z1 | 0: st = Z; op = MPRI_SUB; goto instr;
112 case Z1 | 1: st = X; op = MPRI_ADD; goto instr;
113 case X | 0: st = X0; break;
114 case X0 | 1: st = X; op = MPRI_ADD; goto instr;
115 case X0 | 0: st = Z; op = MPRI_SUB; goto instr;
116 instr:
117 w = (d - i)/MPW_BITS + 1;
118 b = (MPW_BITS + i - d - 1)%MPW_BITS;
119 INSTR(op | !!b, w, b);
120 }
121 }
122 if ((DA(&iv)[DA_LEN(&iv) - 1].op & ~1u) == MPRI_SUB) {
f4535c64 123 mp_drop(r->p);
124 DA_DESTROY(&iv);
125 return (-1);
f46efa79 126 }
127
128#undef INSTR
129
130 /* --- Wrap up --- */
131
132 r->in = DA_LEN(&iv);
133 if (!r->s) {
134 r->iv = xmalloc(r->in * sizeof(mpreduce_instr));
135 memcpy(r->iv, DA(&iv), r->in * sizeof(mpreduce_instr));
136 } else {
137 r->iv = xmalloc(r->in * 2 * sizeof(mpreduce_instr));
138 for (i = 0; i < r->in; i++) {
139 r->iv[i] = DA(&iv)[i];
140 op = r->iv[i].op & ~1u;
141 w = r->iv[i].argx;
142 b = r->iv[i].argy;
143 b += r->s;
144 if (b >= MPW_BITS) {
145 b -= MPW_BITS;
146 w--;
147 }
148 if (b) op |= 1;
149 r->iv[i + r->in].op = op;
150 r->iv[i + r->in].argx = w;
151 r->iv[i + r->in].argy = b;
152 }
153 }
f4535c64 154 DA_DESTROY(&iv);
155 return (0);
f46efa79 156}
157
158/* --- @mpreduce_destroy@ --- *
159 *
160 * Arguments: @mpreduce *r@ = structure to free
161 *
162 * Returns: ---
163 *
164 * Use: Reclaims the resources from a reduction context.
165 */
166
167void mpreduce_destroy(mpreduce *r)
168{
169 mp_drop(r->p);
170 xfree(r->iv);
171}
172
173/* --- @mpreduce_dump@ --- *
174 *
175 * Arguments: @mpreduce *r@ = structure to dump
176 * @FILE *fp@ = file to dump on
177 *
178 * Returns: ---
179 *
180 * Use: Dumps a reduction context.
181 */
182
183void mpreduce_dump(mpreduce *r, FILE *fp)
184{
185 size_t i;
186 static const char *opname[] = { "add", "addshift", "sub", "subshift" };
187
188 fprintf(fp, "mod = "); mp_writefile(r->p, fp, 16);
189 fprintf(fp, "\n lim = %lu; s = %d\n", (unsigned long)r->lim, r->s);
190 for (i = 0; i < r->in; i++) {
191 assert(r->iv[i].op < N(opname));
192 fprintf(fp, " %s %lu %lu\n",
193 opname[r->iv[i].op],
194 (unsigned long)r->iv[i].argx,
195 (unsigned long)r->iv[i].argy);
196 }
197 if (r->s) {
198 fprintf(fp, "tail end charlie\n");
199 for (i = r->in; i < 2 * r->in; i++) {
200 assert(r->iv[i].op < N(opname));
201 fprintf(fp, " %s %lu %lu\n",
202 opname[r->iv[i].op],
203 (unsigned long)r->iv[i].argx,
204 (unsigned long)r->iv[i].argy);
205 }
206 }
207}
208
209/* --- @mpreduce_do@ --- *
210 *
211 * Arguments: @mpreduce *r@ = reduction context
212 * @mp *d@ = destination
213 * @mp *x@ = source
214 *
215 * Returns: Destination, @x@ reduced modulo the reduction poly.
216 */
217
218static void run(const mpreduce_instr *i, const mpreduce_instr *il,
219 mpw *v, mpw z)
220{
221 for (; i < il; i++) {
222#ifdef DEBUG
223 mp vv;
224 mp_build(&vv, v - i->argx, v + 1);
225 printf(" 0x"); mp_writefile(&vv, stdout, 16);
226 printf(" %c (0x%lx << %u) == 0x",
227 (i->op & ~1u) == MPRI_ADD ? '+' : '-',
228 (unsigned long)z,
229 i->argy);
230#endif
231 switch (i->op) {
232 case MPRI_ADD: MPX_UADDN(v - i->argx, v + 1, z); break;
233 case MPRI_ADDLSL: mpx_uaddnlsl(v - i->argx, v + 1, z, i->argy); break;
234 case MPRI_SUB: MPX_USUBN(v - i->argx, v + 1, z); break;
235 case MPRI_SUBLSL: mpx_usubnlsl(v - i->argx, v + 1, z, i->argy); break;
236 default:
237 abort();
238 }
239#ifdef DEBUG
240 mp_build(&vv, v - i->argx, v + 1);
241 mp_writefile(&vv, stdout, 16);
242 printf("\n");
243#endif
244 }
245}
246
247mp *mpreduce_do(mpreduce *r, mp *d, mp *x)
248{
249 mpw *v, *vl;
250 const mpreduce_instr *il;
251 mpw z;
252
253#ifdef DEBUG
254 mp *_r = 0, *_rr = 0;
255#endif
256
257 /* --- If source is negative, divide --- */
258
a69a3efd 259 if (MP_NEGP(x)) {
f46efa79 260 mp_div(0, &d, x, r->p);
261 return (d);
262 }
263
264 /* --- Try to reuse the source's space --- */
265
266 MP_COPY(x);
267 if (d) MP_DROP(d);
268 MP_DEST(x, MP_LEN(x), x->f);
269
270 /* --- Do the reduction --- */
271
272#ifdef DEBUG
273 _r = MP_NEW;
274 mp_div(0, &_r, x, r->p);
275 MP_PRINTX("x", x);
276 _rr = 0;
277#endif
278
279 il = r->iv + r->in;
280 if (MP_LEN(x) >= r->lim) {
281 v = x->v + r->lim;
282 vl = x->vl;
283 while (vl-- > v) {
284 while (*vl) {
285 z = *vl;
286 *vl = 0;
287 run(r->iv, il, vl, z);
288#ifdef DEBUG
106b481c
MW		 289 MP_PRINTX("x", x);
290 mp_div(0, &_rr, x, r->p);
291 assert(MP_EQ(_r, _rr));
f46efa79 292#endif
293 }
294 }
295 if (r->s) {
296 while (*vl >> r->s) {
297 z = *vl >> r->s;
298 *vl &= ((1 << r->s) - 1);
299 run(r->iv + r->in, il + r->in, vl, z);
300#ifdef DEBUG
106b481c
MW		 301 MP_PRINTX("x", x);
302 mp_div(0, &_rr, x, r->p);
303 assert(MP_EQ(_r, _rr));
f46efa79 304#endif
305 }
306 }
307 }
308
309 /* --- Finishing touches --- */
310
311 MP_SHRINK(x);
312 if (MP_CMP(x, >=, r->p))
313 x = mp_sub(x, x, r->p);
314
315 /* --- Done --- */
316
317#ifdef DEBUG
318 assert(MP_EQ(_r, x));
319 mp_drop(_r);
320 mp_drop(_rr);
321#endif
322 return (x);
323}
324
325/* --- @mpreduce_exp@ --- *
326 *
327 * Arguments: @mpreduce *mr@ = pointer to reduction context
328 * @mp *d@ = fake destination
329 * @mp *a@ = base
330 * @mp *e@ = exponent
331 *
332 * Returns: Result, %$a^e \bmod m$%.
333 */
334
335mp *mpreduce_exp(mpreduce *mr, mp *d, mp *a, mp *e)
336{
337 mp *x = MP_ONE;
338 mp *spare = (e->f & MP_BURN) ? MP_NEWSEC : MP_NEW;
339
340 MP_SHRINK(e);
a69a3efd 341 MP_COPY(a);
342 if (MP_ZEROP(e))
f46efa79 343 ;
a69a3efd 344 else {
345 if (MP_NEGP(e))
346 a = mp_modinv(a, a, mr->p);
347 if (MP_LEN(e) < EXP_THRESH)
348 EXP_SIMPLE(x, a, e);
349 else
350 EXP_WINDOW(x, a, e);
351 }
352 mp_drop(a);
f46efa79 353 mp_drop(d);
354 mp_drop(spare);
355 return (x);
356}
357
358/*----- Test rig ----------------------------------------------------------*/
359
360
361#ifdef TEST_RIG
362
363#define MP(x) mp_readstring(MP_NEW, #x, 0, 0)
364
365static int vreduce(dstr *v)
366{
367 mp *d = *(mp **)v[0].buf;
368 mp *n = *(mp **)v[1].buf;
369 mp *r = *(mp **)v[2].buf;
370 mp *c;
371 int ok = 1;
372 mpreduce rr;
373
374 mpreduce_create(&rr, d);
375 c = mpreduce_do(&rr, MP_NEW, n);
376 if (!MP_EQ(c, r)) {
377 fprintf(stderr, "\n*** reduction failed\n*** ");
378 mpreduce_dump(&rr, stderr);
379 fprintf(stderr, "\n*** n = "); mp_writefile(n, stderr, 10);
380 fprintf(stderr, "\n*** r = "); mp_writefile(r, stderr, 10);
381 fprintf(stderr, "\n*** c = "); mp_writefile(c, stderr, 10);
382 fprintf(stderr, "\n");
383 ok = 0;
384 }
385 mpreduce_destroy(&rr);
386 mp_drop(n); mp_drop(d); mp_drop(r); mp_drop(c);
387 assert(mparena_count(MPARENA_GLOBAL) == 0);
388 return (ok);
389}
390
391static int vmodexp(dstr *v)
392{
393 mp *p = *(mp **)v[0].buf;
394 mp *g = *(mp **)v[1].buf;
395 mp *x = *(mp **)v[2].buf;
396 mp *r = *(mp **)v[3].buf;
397 mp *c;
398 int ok = 1;
399 mpreduce rr;
400
401 mpreduce_create(&rr, p);
402 c = mpreduce_exp(&rr, MP_NEW, g, x);
403 if (!MP_EQ(c, r)) {
404 fprintf(stderr, "\n*** modexp failed\n*** ");
405 fprintf(stderr, "\n*** p = "); mp_writefile(p, stderr, 10);
406 fprintf(stderr, "\n*** g = "); mp_writefile(g, stderr, 10);
407 fprintf(stderr, "\n*** x = "); mp_writefile(x, stderr, 10);
408 fprintf(stderr, "\n*** c = "); mp_writefile(c, stderr, 10);
409 fprintf(stderr, "\n*** r = "); mp_writefile(r, stderr, 10);
410 fprintf(stderr, "\n");
411 ok = 0;
412 }
413 mpreduce_destroy(&rr);
414 mp_drop(p); mp_drop(g); mp_drop(r); mp_drop(x); mp_drop(c);
415 assert(mparena_count(MPARENA_GLOBAL) == 0);
416 return (ok);
417}
418
419static test_chunk defs[] = {
420 { "reduce", vreduce, { &type_mp, &type_mp, &type_mp, 0 } },
421 { "modexp", vmodexp, { &type_mp, &type_mp, &type_mp, &type_mp, 0 } },
422 { 0, 0, { 0 } }
423};
424
425int main(int argc, char *argv[])
426{
427 test_run(argc, argv, defs, SRCDIR"/tests/mpreduce");
428 return (0);
429}
430
431#endif
432
433/*----- That's all, folks -------------------------------------------------*/
Catacomb cryptographic library (downstream of http://git.distorted.org.uk/~mdw/catacomb/)