5 * Efficient reduction modulo sparse binary polynomials
7 * (c) 2004 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 /*----- Header files ------------------------------------------------------*/
32 #include <mLib/alloc.h>
33 #include <mLib/darray.h>
34 #include <mLib/macros.h>
38 #include "gfreduce-exp.h"
42 /*----- Data structures ---------------------------------------------------*/
44 DA_DECL(instr_v, gfreduce_instr);
46 /*----- Main code ---------------------------------------------------------*/
48 /* --- What's going on here? --- *
50 * Let's face it, @gfx_div@ sucks. It works (I hope), but it's not in any
51 * sense fast. Here, we do efficient reduction modulo sparse polynomials.
53 * Suppose we have a polynomial @X@ we're trying to reduce mod @P@. If we
54 * take the topmost nonzero word of @X@, call it @w@, then we can eliminate
55 * it by subtracting off @w P x^{k}@ for an appropriate value of @k@. The
56 * trick is in observing that if @P@ is sparse we can do this multiplication
57 * and subtraction efficiently, just by XORing appropriate shifts of @w@ into
60 * The first tricky bit is in working out when to stop. I'll use eight-bit
61 * words to demonstrate what I'm talking about.
63 * xxxxxx xxxxxxxx xxxxxxxx xxxxxxxx xxxxxxxx xxxxxxxx
64 * 001ppppp pppppppp pppppppp pppppppp
66 * |<------------ bp ------------->|
67 * |<------------ nw --------------->|
69 * The trick of taking whole words off of @X@ stops working when there are
70 * only @nw@ words left. Then we have to mask off the bottom bits of @w@
74 /* --- @gfreduce_create@ --- *
76 * Arguments: @gfreduce *r@ = structure to fill in
77 * @mp *x@ = a (hopefully sparse) polynomial
81 * Use: Initializes a context structure for reduction.
84 void gfreduce_create(gfreduce *r, mp *p)
93 size_t w, ww, wi, wl, ll, bb;
95 /* --- Sort out the easy stuff --- */
97 d = mp_bits(p); assert(d); d--;
103 r->mask = MPW(((mpw)-1) << dw);
108 /* --- Stash a new instruction --- */
110 #define INSTR(op_, arg_) do { \
112 DA(&iv)[DA_LEN(&iv)].op = (op_); \
113 DA(&iv)[DA_LEN(&iv)].arg = (arg_); \
119 w = (d + MPW_BITS - 1)/MPW_BITS;
125 for (i = 0, mp_scan(&sc, p); mp_step(&sc) && i < d; i++) {
128 ww = (d - i + MPW_BITS - 1)/MPW_BITS;
131 INSTR(GFRI_STORE, w);
135 INSTR(GFRI_LOAD, ww);
137 INSTR(GFRI_LOAD, w - 1);
138 for (; wi < wl; wi++) {
140 assert(ip->op == GFRI_LSL);
142 INSTR(GFRI_LSR, MPW_BITS - ip->arg);
145 INSTR(GFRI_STORE, w - 1);
146 INSTR(GFRI_LOAD, ww);
153 INSTR(GFRI_LSL, (bb + i)%MPW_BITS);
154 if ((bb + i)%MPW_BITS)
158 INSTR(GFRI_STORE, w);
162 INSTR(GFRI_LOAD, w - 1);
163 for (; wi < wl; wi++) {
165 assert(ip->op == GFRI_LSL);
167 INSTR(GFRI_LSR, MPW_BITS - ip->arg);
169 INSTR(GFRI_STORE, w - 1);
175 r->iv = xmalloc(r->in * sizeof(gfreduce_instr));
177 memcpy(r->iv, DA(&iv), r->in * sizeof(gfreduce_instr));
181 /* --- @gfreduce_destroy@ --- *
183 * Arguments: @gfreduce *r@ = structure to free
187 * Use: Reclaims the resources from a reduction context.
190 void gfreduce_destroy(gfreduce *r)
196 /* --- @gfreduce_dump@ --- *
198 * Arguments: @gfreduce *r@ = structure to dump
199 * @FILE *fp@ = file to dump on
203 * Use: Dumps a reduction context.
206 void gfreduce_dump(gfreduce *r, FILE *fp)
210 fprintf(fp, "poly = "); mp_writefile(r->p, fp, 16);
211 fprintf(fp, "\n lim = %lu; mask = %lx\n",
212 (unsigned long)r->lim, (unsigned long)r->mask);
213 for (i = 0; i < r->in; i++) {
214 static const char *opname[] = { "load", "lsl", "lsr", "store" };
215 assert(r->iv[i].op < N(opname));
216 fprintf(fp, " %s %lu\n",
218 (unsigned long)r->iv[i].arg);
222 /* --- @gfreduce_do@ --- *
224 * Arguments: @gfreduce *r@ = reduction context
225 * @mp *d@ = destination
228 * Returns: Destination, @x@ reduced modulo the reduction poly.
231 static void run(const gfreduce_instr *i, const gfreduce_instr *il,
236 for (; i < il; i++) {
238 case GFRI_LOAD: w = *(v - i->arg); break;
239 case GFRI_LSL: w ^= z << i->arg; break;
240 case GFRI_LSR: w ^= z >> i->arg; break;
241 case GFRI_STORE: *(v - i->arg) = MPW(w); break;
247 mp *gfreduce_do(gfreduce *r, mp *d, mp *x)
250 const gfreduce_instr *il;
253 /* --- Try to reuse the source's space --- */
257 MP_DEST(x, MP_LEN(x), x->f);
259 /* --- Do the reduction --- */
262 if (MP_LEN(x) >= r->lim) {
269 run(r->iv, il, vl, z);
273 while (*vl & r->mask) {
276 run(r->iv, il, vl, z);
287 /* --- @gfreduce_sqrt@ --- *
289 * Arguments: @gfreduce *r@ = pointer to reduction context
290 * @mp *d@ = destination
291 * @mp *x@ = some polynomial
293 * Returns: The square root of @x@ modulo @r->p@, or null.
296 mp *gfreduce_sqrt(gfreduce *r, mp *d, mp *x)
299 mp *z, *spare = MP_NEW;
300 unsigned long m = mp_bits(r->p) - 1;
303 for (i = 0; i < m - 1; i++) {
304 mp *t = gf_sqr(spare, y);
306 y = gfreduce_do(r, t, t);
308 z = gf_sqr(spare, y);
309 z = gfreduce_do(r, z, z);
319 /* --- @gfreduce_trace@ --- *
321 * Arguments: @gfreduce *r@ = pointer to reduction context
322 * @mp *x@ = some polynomial
324 * Returns: The trace of @x@. (%$\Tr(x)=x + x^2 + \cdots + x^{2^{m-1}}$%
325 * if %$x \in \gf{2^m}$%).
328 int gfreduce_trace(gfreduce *r, mp *x)
332 unsigned long m = mp_bits(r->p) - 1;
336 for (i = 0; i < m - 1; i++) {
337 mp *t = gf_sqr(spare, y);
339 y = gfreduce_do(r, t, t);
348 /* --- @gfreduce_halftrace@ --- *
350 * Arguments: @gfreduce *r@ = pointer to reduction context
351 * @mp *d@ = destination
352 * @mp *x@ = some polynomial
354 * Returns: The half-trace of @x@.
355 * (%$\HfTr(x)= x + x^{2^2} + \cdots + x^{2^{m-1}}$%
356 * if %$x \in \gf{2^m}$% with %$m$% odd).
359 mp *gfreduce_halftrace(gfreduce *r, mp *d, mp *x)
363 unsigned long m = mp_bits(r->p) - 1;
367 for (i = 0; i < m - 1; i += 2) {
368 mp *t = gf_sqr(spare, y);
370 y = gfreduce_do(r, t, t);
371 t = gf_sqr(spare, y);
373 y = gfreduce_do(r, t, t);
380 /* --- @gfreduce_quadsolve@ --- *
382 * Arguments: @gfreduce *r@ = pointer to reduction context
383 * @mp *d@ = destination
384 * @mp *x@ = some polynomial
386 * Returns: A polynomial @y@ such that %$y^2 + y = x$%, or null.
389 mp *gfreduce_quadsolve(gfreduce *r, mp *d, mp *x)
391 unsigned long m = mp_bits(r->p) - 1;
396 d = gfreduce_halftrace(r, d, x);
398 mp *z, *w, *rho = MP_NEW;
400 grand *fr = fibrand_create(0);
404 rho = mprand(rho, m, fr, 0);
407 for (i = 0; i < m - 1; i++) {
408 t = gf_sqr(spare, z); spare = z; z = gfreduce_do(r, t, t);
409 t = gf_sqr(spare, w); spare = w; w = gfreduce_do(r, t, t);
410 t = gf_mul(spare, w, x); t = gfreduce_do(r, t, t); spare = t;
412 w = gf_add(w, w, rho);
423 fr->ops->destroy(fr);
427 t = gf_sqr(MP_NEW, d); t = gfreduce_do(r, t, t); t = gf_add(t, t, d);
434 if (d) d->v[0] &= ~(mpw)1;
438 /* --- @gfreduce_exp@ --- *
440 * Arguments: @gfreduce *gr@ = pointer to reduction context
441 * @mp *d@ = fake destination
445 * Returns: Result, %$a^e \bmod m$%.
448 mp *gfreduce_exp(gfreduce *gr, mp *d, mp *a, mp *e)
451 mp *spare = (e->f & MP_BURN) ? MP_NEWSEC : MP_NEW;
459 a = gf_modinv(a, a, gr->p);
460 if (MP_LEN(e) < EXP_THRESH)
471 /*----- Test rig ----------------------------------------------------------*/
475 #define MP(x) mp_readstring(MP_NEW, #x, 0, 0)
477 static int vreduce(dstr *v)
479 mp *d = *(mp **)v[0].buf;
480 mp *n = *(mp **)v[1].buf;
481 mp *r = *(mp **)v[2].buf;
486 gfreduce_create(&rr, d);
487 c = gfreduce_do(&rr, MP_NEW, n);
489 fprintf(stderr, "\n*** reduction failed\n*** ");
490 gfreduce_dump(&rr, stderr);
491 fprintf(stderr, "\n*** n = "); mp_writefile(n, stderr, 16);
492 fprintf(stderr, "\n*** r = "); mp_writefile(r, stderr, 16);
493 fprintf(stderr, "\n*** c = "); mp_writefile(c, stderr, 16);
494 fprintf(stderr, "\n");
497 gfreduce_destroy(&rr);
498 mp_drop(n); mp_drop(d); mp_drop(r); mp_drop(c);
499 assert(mparena_count(MPARENA_GLOBAL) == 0);
503 static int vmodexp(dstr *v)
505 mp *p = *(mp **)v[0].buf;
506 mp *g = *(mp **)v[1].buf;
507 mp *x = *(mp **)v[2].buf;
508 mp *r = *(mp **)v[3].buf;
513 gfreduce_create(&rr, p);
514 c = gfreduce_exp(&rr, MP_NEW, g, x);
516 fprintf(stderr, "\n*** modexp failed\n*** ");
517 fprintf(stderr, "\n*** p = "); mp_writefile(p, stderr, 16);
518 fprintf(stderr, "\n*** g = "); mp_writefile(g, stderr, 16);
519 fprintf(stderr, "\n*** x = "); mp_writefile(x, stderr, 16);
520 fprintf(stderr, "\n*** c = "); mp_writefile(c, stderr, 16);
521 fprintf(stderr, "\n*** r = "); mp_writefile(r, stderr, 16);
522 fprintf(stderr, "\n");
525 gfreduce_destroy(&rr);
526 mp_drop(p); mp_drop(g); mp_drop(r); mp_drop(x); mp_drop(c);
527 assert(mparena_count(MPARENA_GLOBAL) == 0);
531 static int vsqrt(dstr *v)
533 mp *p = *(mp **)v[0].buf;
534 mp *x = *(mp **)v[1].buf;
535 mp *r = *(mp **)v[2].buf;
540 gfreduce_create(&rr, p);
541 c = gfreduce_sqrt(&rr, MP_NEW, x);
543 fprintf(stderr, "\n*** sqrt failed\n*** ");
544 fprintf(stderr, "\n*** p = "); mp_writefile(p, stderr, 16);
545 fprintf(stderr, "\n*** x = "); mp_writefile(x, stderr, 16);
546 fprintf(stderr, "\n*** c = "); mp_writefile(c, stderr, 16);
547 fprintf(stderr, "\n*** r = "); mp_writefile(r, stderr, 16);
548 fprintf(stderr, "\n");
551 gfreduce_destroy(&rr);
552 mp_drop(p); mp_drop(r); mp_drop(x); mp_drop(c);
553 assert(mparena_count(MPARENA_GLOBAL) == 0);
557 static int vtr(dstr *v)
559 mp *p = *(mp **)v[0].buf;
560 mp *x = *(mp **)v[1].buf;
561 int r = *(int *)v[2].buf, c;
565 gfreduce_create(&rr, p);
566 c = gfreduce_trace(&rr, x);
568 fprintf(stderr, "\n*** trace failed\n*** ");
569 fprintf(stderr, "\n*** p = "); mp_writefile(p, stderr, 16);
570 fprintf(stderr, "\n*** x = "); mp_writefile(x, stderr, 16);
571 fprintf(stderr, "\n*** c = %d", c);
572 fprintf(stderr, "\n*** r = %d", r);
573 fprintf(stderr, "\n");
576 gfreduce_destroy(&rr);
577 mp_drop(p); mp_drop(x);
578 assert(mparena_count(MPARENA_GLOBAL) == 0);
582 static int vhftr(dstr *v)
584 mp *p = *(mp **)v[0].buf;
585 mp *x = *(mp **)v[1].buf;
586 mp *r = *(mp **)v[2].buf;
591 gfreduce_create(&rr, p);
592 c = gfreduce_halftrace(&rr, MP_NEW, x);
594 fprintf(stderr, "\n*** halftrace failed\n*** ");
595 fprintf(stderr, "\n*** p = "); mp_writefile(p, stderr, 16);
596 fprintf(stderr, "\n*** x = "); mp_writefile(x, stderr, 16);
597 fprintf(stderr, "\n*** c = "); mp_writefile(c, stderr, 16);
598 fprintf(stderr, "\n*** r = "); mp_writefile(r, stderr, 16);
599 fprintf(stderr, "\n");
602 gfreduce_destroy(&rr);
603 mp_drop(p); mp_drop(r); mp_drop(x); mp_drop(c);
604 assert(mparena_count(MPARENA_GLOBAL) == 0);
608 static int vquad(dstr *v)
610 mp *p = *(mp **)v[0].buf;
611 mp *x = *(mp **)v[1].buf;
612 mp *r = *(mp **)v[2].buf;
617 gfreduce_create(&rr, p);
618 c = gfreduce_quadsolve(&rr, MP_NEW, x);
620 fprintf(stderr, "\n*** quadsolve failed\n*** ");
621 fprintf(stderr, "\n*** p = "); mp_writefile(p, stderr, 16);
622 fprintf(stderr, "\n*** x = "); mp_writefile(x, stderr, 16);
623 fprintf(stderr, "\n*** c = "); mp_writefile(c, stderr, 16);
624 fprintf(stderr, "\n*** r = "); mp_writefile(r, stderr, 16);
625 fprintf(stderr, "\n");
628 gfreduce_destroy(&rr);
629 mp_drop(p); mp_drop(r); mp_drop(x); mp_drop(c);
630 assert(mparena_count(MPARENA_GLOBAL) == 0);
634 static test_chunk defs[] = {
635 { "reduce", vreduce, { &type_mp, &type_mp, &type_mp, 0 } },
636 { "modexp", vmodexp, { &type_mp, &type_mp, &type_mp, &type_mp, 0 } },
637 { "sqrt", vsqrt, { &type_mp, &type_mp, &type_mp, 0 } },
638 { "trace", vtr, { &type_mp, &type_mp, &type_int, 0 } },
639 { "halftrace", vhftr, { &type_mp, &type_mp, &type_mp, 0 } },
640 { "quadsolve", vquad, { &type_mp, &type_mp, &type_mp, 0 } },
644 int main(int argc, char *argv[])
646 test_run(argc, argv, defs, SRCDIR"/tests/gfreduce");
652 /*----- That's all, folks -------------------------------------------------*/