3 * $Id: mpmont.c,v 1.10 2000/07/29 17:05:43 mdw Exp $
7 * (c) 1999 Straylight/Edgeware
10 /*----- Licensing notice --------------------------------------------------*
12 * This file is part of Catacomb.
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.
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.
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,
30 /*----- Revision history --------------------------------------------------*
33 * Revision 1.10 2000/07/29 17:05:43 mdw
34 * (mpmont_expr): Use sliding window exponentiation, with a drop-through
35 * for small exponents to use a simple left-to-right bitwise routine. This
36 * can reduce modexp times by up to a quarter.
38 * Revision 1.9 2000/06/17 11:45:09 mdw
39 * Major memory management overhaul. Added arena support. Use the secure
40 * arena for secret integers. Replace and improve the MP management macros
41 * (e.g., replace MP_MODIFY by MP_DEST).
43 * Revision 1.8 1999/12/22 15:55:00 mdw
44 * Adjust Karatsuba parameters.
46 * Revision 1.7 1999/12/11 01:51:14 mdw
47 * Use a Karatsuba-based reduction for large moduli.
49 * Revision 1.6 1999/12/10 23:18:39 mdw
50 * Change interface for suggested destinations.
52 * Revision 1.5 1999/11/22 13:58:40 mdw
53 * Add an option to disable Montgomery reduction, so that performance
54 * comparisons can be done.
56 * Revision 1.4 1999/11/21 12:27:06 mdw
57 * Remove a division from the Montgomery setup by calculating
58 * %$R^2 \bmod m$% first and then %$R \bmod m$% by Montgomery reduction of
61 * Revision 1.3 1999/11/21 11:35:10 mdw
62 * Performance improvement: use @mp_sqr@ and @mpmont_reduce@ instead of
63 * @mpmont_mul@ for squaring in exponentiation.
65 * Revision 1.2 1999/11/19 13:17:26 mdw
66 * Add extra interface to exponentiation which returns a Montgomerized
69 * Revision 1.1 1999/11/17 18:02:16 mdw
70 * New multiprecision integer arithmetic suite.
74 /*----- Header files ------------------------------------------------------*/
79 /*----- Tweakables --------------------------------------------------------*/
81 /* --- @MPMONT_DISABLE@ --- *
83 * Replace all the clever Montgomery reduction with good old-fashioned long
87 /* #define MPMONT_DISABLE */
89 /*----- Main code ---------------------------------------------------------*/
91 /* --- @mpmont_create@ --- *
93 * Arguments: @mpmont *mm@ = pointer to Montgomery reduction context
94 * @mp *m@ = modulus to use
98 * Use: Initializes a Montgomery reduction context ready for use.
99 * The argument @m@ must be a positive odd integer.
102 #ifdef MPMONT_DISABLE
104 void mpmont_create(mpmont *mm, mp *m)
115 void mpmont_create(mpmont *mm, mp *m)
117 size_t n = MP_LEN(m);
118 mp *r2 = mp_new(2 * n + 1, 0);
121 /* --- Validate the arguments --- */
123 assert(((void)"Montgomery modulus must be positive",
124 (m->f & MP_NEG) == 0));
125 assert(((void)"Montgomery modulus must be odd", m->v[0] & 1));
127 /* --- Take a copy of the modulus --- */
132 /* --- Determine %$R^2$% --- */
135 MPX_ZERO(r2->v, r2->vl - 1);
138 /* --- Find the magic value @mi@ --- */
140 mp_build(&r, r2->v + n, r2->vl);
142 mp_gcd(0, 0, &mm->mi, &r, m);
143 mm->mi = mp_sub(mm->mi, &r, mm->mi);
145 /* --- Discover the values %$R \bmod m$% and %$R^2 \bmod m$% --- */
148 mp_div(0, &mm->r2, r2, m);
149 mm->r = mpmont_reduce(mm, MP_NEW, mm->r2);
155 /* --- @mpmont_destroy@ --- *
157 * Arguments: @mpmont *mm@ = pointer to a Montgomery reduction context
161 * Use: Disposes of a context when it's no longer of any use to
165 void mpmont_destroy(mpmont *mm)
173 /* --- @mpmont_reduce@ --- *
175 * Arguments: @mpmont *mm@ = pointer to Montgomery reduction context
176 * @mp *d@ = destination
177 * @mp *a@ = source, assumed positive
179 * Returns: Result, %$a R^{-1} \bmod m$%.
182 #ifdef MPMONT_DISABLE
184 mp *mpmont_reduce(mpmont *mm, mp *d, mp *a)
186 mp_div(0, &d, a, mm->m);
192 mp *mpmont_reduce(mpmont *mm, mp *d, mp *a)
196 /* --- Check for serious Karatsuba reduction --- */
198 if (n > KARATSUBA_CUTOFF * 3) {
207 mp_build(&al, a->v, vl);
208 u = mp_mul(MP_NEW, &al, mm->mi);
211 u = mp_mul(u, u, mm->m);
216 /* --- Otherwise do it the hard way --- */
224 /* --- Initial conditioning of the arguments --- */
230 MP_DEST(d, 2 * n + 1, a->f);
232 dv = d->v; dvl = d->vl;
233 mv = mm->m->v; mvl = mm->m->vl;
235 /* --- Let's go to work --- */
239 mpw u = MPW(*dv * mi);
240 MPX_UMLAN(dv, dvl, mv, mvl, u);
245 /* --- Wrap everything up --- */
247 memmove(d->v, d->v + n, MPWS(MP_LEN(d) - n));
249 if (MP_CMP(d, >=, mm->m))
250 d = mp_sub(d, d, mm->m);
257 /* --- @mpmont_mul@ --- *
259 * Arguments: @mpmont *mm@ = pointer to Montgomery reduction context
260 * @mp *d@ = destination
261 * @mp *a, *b@ = sources, assumed positive
263 * Returns: Result, %$a b R^{-1} \bmod m$%.
266 #ifdef MPMONT_DISABLE
268 mp *mpmont_mul(mpmont *mm, mp *d, mp *a, mp *b)
271 mp_div(0, &d, d, mm->m);
277 mp *mpmont_mul(mpmont *mm, mp *d, mp *a, mp *b)
279 if (mm->n > KARATSUBA_CUTOFF * 3) {
281 d = mpmont_reduce(mm, d, d);
291 /* --- Initial conditioning of the arguments --- */
293 if (MP_LEN(a) > MP_LEN(b)) {
294 mp *t = a; a = b; b = t;
300 MP_DEST(d, 2 * n + 1, a->f | b->f | MP_UNDEF);
301 dv = d->v; dvl = d->vl;
303 av = a->v; avl = a->vl;
304 bv = b->v; bvl = b->vl;
305 mv = mm->m->v; mvl = mm->m->vl;
308 /* --- Montgomery multiplication phase --- */
312 while (i < n && av < avl) {
314 mpw u = MPW((*dv + x * y) * mi);
315 MPX_UMLAN(dv, dvl, bv, bvl, x);
316 MPX_UMLAN(dv, dvl, mv, mvl, u);
321 /* --- Simpler Montgomery reduction phase --- */
324 mpw u = MPW(*dv * mi);
325 MPX_UMLAN(dv, dvl, mv, mvl, u);
332 memmove(d->v, dv, MPWS(dvl - dv));
335 d->f = (a->f | b->f) & MP_BURN;
336 if (MP_CMP(d, >=, mm->m))
337 d = mp_sub(d, d, mm->m);
347 /* --- @mpmont_expr@ --- *
349 * Arguments: @mpmont *mm@ = pointer to Montgomery reduction context
350 * @mp *d@ = fake destination
354 * Returns: Result, %$a^e R \bmod m$%.
358 #define TABSZ (1 << (WINSZ - 1))
360 #define THRESH (((MPW_BITS / WINSZ) << 2) + 1)
362 static mp *exp_simple(mpmont *mm, mp *d, mp *a, mp *e)
366 mp *x = MP_COPY(mm->r);
367 mp *spare = (e->f & MP_BURN) ? MP_NEWSEC : MP_NEW;
373 while (!MP_RBIT(&sc))
376 /* --- Do the main body of the work --- */
378 ar = mpmont_mul(mm, MP_NEW, a, mm->r2);
383 y = mp_sqr(spare, x);
384 y = mpmont_reduce(mm, y, y);
388 { mp *y = mpmont_mul(mm, spare, x, ar); spare = x; x = y; }
399 /* --- Do a final round of squaring --- */
404 y = mp_sqr(spare, x);
405 y = mpmont_reduce(mm, y, y);
421 mp *mpmont_expr(mpmont *mm, mp *d, mp *a, mp *e)
425 mp *spare = (e->f & MP_BURN) ? MP_NEWSEC : MP_NEW;
426 mp *x = MP_COPY(mm->r);
430 /* --- Do we bother? --- */
435 if (MP_LEN(e) < THRESH) {
437 return (exp_simple(mm, d, a, e));
440 /* --- Do the precomputation --- */
442 ar = mpmont_mul(mm, MP_NEW, a, mm->r2);
443 a2 = mp_sqr(MP_NEW, ar);
444 a2 = mpmont_reduce(mm, a2, a2);
445 tab = xmalloc(TABSZ * sizeof(mp *));
447 for (i = 1; i < TABSZ; i++)
448 tab[i] = mpmont_mul(mm, MP_NEW, tab[i - 1], a2);
452 /* --- Skip top-end zero bits --- *
454 * If the initial step worked, there must be a set bit somewhere, so keep
455 * stepping until I find it.
459 while (!MP_RBIT(&sc)) {
463 /* --- Now for the main work --- */
469 /* --- The next bit is set, so read a window index --- *
471 * Reset @i@ to zero and increment @sq@. Then, until either I read
472 * @WINSZ@ bits or I run out of bits, scan in a bit: if it's clear, bump
473 * the @z@ counter; if it's set, push a set bit into @i@, shift it over
474 * by @z@ bits, bump @sq@ by @z + 1@ and clear @z@. By the end of this
475 * palaver, @i@ is an index to the precomputed value in @tab@.
482 if (l >= WINSZ || !MP_RSTEP(&sc))
487 i = ((i << 1) | 1) << z;
493 /* --- Do the squaring --- *
495 * Remember that @sq@ carries over from the zero-skipping stuff below.
500 y = mp_sqr(spare, x);
501 y = mpmont_reduce(mm, y, y);
506 /* --- Do the multiply --- */
508 { mp *y = mpmont_mul(mm, spare, x, tab[i]); spare = x; x = y; }
510 /* --- Now grind along through the rest of the bits --- */
522 /* --- Do a final round of squaring --- */
527 y = mp_sqr(spare, x);
528 y = mpmont_reduce(mm, y, y);
535 for (i = 0; i < TABSZ; i++)
546 /* --- @mpmont_exp@ --- *
548 * Arguments: @mpmont *mm@ = pointer to Montgomery reduction context
549 * @mp *d@ = fake destination
553 * Returns: Result, %$a^e \bmod m$%.
556 mp *mpmont_exp(mpmont *mm, mp *d, mp *a, mp *e)
558 d = mpmont_expr(mm, d, a, e);
559 d = mpmont_reduce(mm, d, d);
563 /*----- Test rig ----------------------------------------------------------*/
567 static int tcreate(dstr *v)
569 mp *m = *(mp **)v[0].buf;
570 mp *mi = *(mp **)v[1].buf;
571 mp *r = *(mp **)v[2].buf;
572 mp *r2 = *(mp **)v[3].buf;
577 mpmont_create(&mm, m);
579 if (mm.mi->v[0] != mi->v[0]) {
580 fprintf(stderr, "\n*** bad mi: found %lu, expected %lu",
581 (unsigned long)mm.mi->v[0], (unsigned long)mi->v[0]);
582 fputs("\nm = ", stderr); mp_writefile(m, stderr, 10);
587 if (MP_CMP(mm.r, !=, r)) {
588 fputs("\n*** bad r", stderr);
589 fputs("\nm = ", stderr); mp_writefile(m, stderr, 10);
590 fputs("\nexpected ", stderr); mp_writefile(r, stderr, 10);
591 fputs("\n found ", stderr); mp_writefile(mm.r, stderr, 10);
596 if (MP_CMP(mm.r2, !=, r2)) {
597 fputs("\n*** bad r2", stderr);
598 fputs("\nm = ", stderr); mp_writefile(m, stderr, 10);
599 fputs("\nexpected ", stderr); mp_writefile(r2, stderr, 10);
600 fputs("\n found ", stderr); mp_writefile(mm.r2, stderr, 10);
610 assert(mparena_count(MPARENA_GLOBAL) == 0);
614 static int tmul(dstr *v)
616 mp *m = *(mp **)v[0].buf;
617 mp *a = *(mp **)v[1].buf;
618 mp *b = *(mp **)v[2].buf;
619 mp *r = *(mp **)v[3].buf;
623 mpmont_create(&mm, m);
626 mp *qr = mp_mul(MP_NEW, a, b);
627 mp_div(0, &qr, qr, m);
629 if (MP_CMP(qr, !=, r)) {
630 fputs("\n*** classical modmul failed", stderr);
631 fputs("\n m = ", stderr); mp_writefile(m, stderr, 10);
632 fputs("\n a = ", stderr); mp_writefile(a, stderr, 10);
633 fputs("\n b = ", stderr); mp_writefile(b, stderr, 10);
634 fputs("\n r = ", stderr); mp_writefile(r, stderr, 10);
635 fputs("\nqr = ", stderr); mp_writefile(qr, stderr, 10);
644 mp *ar = mpmont_mul(&mm, MP_NEW, a, mm.r2);
645 mp *br = mpmont_mul(&mm, MP_NEW, b, mm.r2);
646 mp *mr = mpmont_mul(&mm, MP_NEW, ar, br);
647 mr = mpmont_reduce(&mm, mr, mr);
648 if (MP_CMP(mr, !=, r)) {
649 fputs("\n*** montgomery modmul failed", stderr);
650 fputs("\n m = ", stderr); mp_writefile(m, stderr, 10);
651 fputs("\n a = ", stderr); mp_writefile(a, stderr, 10);
652 fputs("\n b = ", stderr); mp_writefile(b, stderr, 10);
653 fputs("\n r = ", stderr); mp_writefile(r, stderr, 10);
654 fputs("\nmr = ", stderr); mp_writefile(mr, stderr, 10);
658 MP_DROP(ar); MP_DROP(br);
668 assert(mparena_count(MPARENA_GLOBAL) == 0);
672 static int texp(dstr *v)
674 mp *m = *(mp **)v[0].buf;
675 mp *a = *(mp **)v[1].buf;
676 mp *b = *(mp **)v[2].buf;
677 mp *r = *(mp **)v[3].buf;
682 mpmont_create(&mm, m);
684 mr = mpmont_exp(&mm, MP_NEW, a, b);
686 if (MP_CMP(mr, !=, r)) {
687 fputs("\n*** montgomery modexp failed", stderr);
688 fputs("\n m = ", stderr); mp_writefile(m, stderr, 10);
689 fputs("\n a = ", stderr); mp_writefile(a, stderr, 10);
690 fputs("\n e = ", stderr); mp_writefile(b, stderr, 10);
691 fputs("\n r = ", stderr); mp_writefile(r, stderr, 10);
692 fputs("\nmr = ", stderr); mp_writefile(mr, stderr, 10);
703 assert(mparena_count(MPARENA_GLOBAL) == 0);
708 static test_chunk tests[] = {
709 { "create", tcreate, { &type_mp, &type_mp, &type_mp, &type_mp, 0 } },
710 { "mul", tmul, { &type_mp, &type_mp, &type_mp, &type_mp, 0 } },
711 { "exp", texp, { &type_mp, &type_mp, &type_mp, &type_mp, 0 } },
715 int main(int argc, char *argv[])
718 test_run(argc, argv, tests, SRCDIR "/tests/mpmont");
724 /*----- That's all, folks -------------------------------------------------*/