1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444
|
#include <math.h>
#include <gmp.h>
#include "ptypes.h"
/*****************************************************************************
*
* The AKS polynomial-time deterministic primality test.
*
* Multiple variants are implemented. The original algorithm as shown in
* articles such as Rotella (2005) is not here as it is much slower than
* the essentially identical algorithm in the updated (V6) AKS paper. The
* updated paper uses improved theorems based on Lenstra et al. which allows
* loweing some limits used in the algorithm.
*
* All versions have a relatively similar O(log^{6.x}(n)) asymptotic growth,
* which is what we expect from AKS.
*
* A version with improvements from Voloch and Bornemann is included, and
* is similar to Bornemann's 2002 Pari/GP implementation. It is *much* faster
* than the V6 algorithm, and to the best of my knowledge is the fastest
* publicly available AKS implementation in early 2016.
*
* The Bernstein 4.1 algorithm implements theorem 4.1 from Bernstein's 2003
* paper, which has the Voloch improvements as well as many more. It is
* another 10-20x faster than the Bornemann version, hence is substantially
* faster than any other known implementation in mid 2016.
*
* Copyright (2012-2016) Dana Jacobsen.
*
*****************************************************************************/
/* In approximate order of performance. */
#define AKS_VARIANT_V6 1 /* The V6 paper with Lenstra impr */
#define AKS_VARIANT_BERN21 2 /* AKS-Bernstein-Morain theorem 2.1 */
#define AKS_VARIANT_BERN22 3 /* AKS-Bernstein-Morain theorem 2.2 */
#define AKS_VARIANT_BERN23 4 /* AKS-Bernstein-Morain theorem 2.3 */
#define AKS_VARIANT_BORNEMANN 5 /* Based on Folkmar Bornemann's impl */
#define AKS_VARIANT_BERN41 6 /* Bernstein 2003, theorem 4.1 */
#define AKS_VARIANT AKS_VARIANT_BERN41
#include "aks.h"
#include "prime_iterator.h"
#include "factor.h"
#if AKS_VARIANT == AKS_VARIANT_BORNEMANN
#define FUNC_mpz_logn 1
#endif
#define FUNC_mpz_log2 1
#include "utility.h"
static int test_anr(UV a, mpz_t n, UV r, mpz_t* px, mpz_t* py)
{
int retval = 1;
UV i, n_mod_r;
mpz_t t;
for (i = 0; i < r; i++)
mpz_set_ui(px[i], 0);
a %= r;
mpz_set_ui(px[0], a);
mpz_set_ui(px[1], 1);
poly_mod_pow(py, px, n, r, n);
mpz_init(t);
n_mod_r = mpz_mod_ui(t, n, r);
if (n_mod_r >= r) croak("n mod r >= r ?!");
mpz_sub_ui(t, py[n_mod_r], 1);
mpz_mod(py[n_mod_r], t, n);
mpz_sub_ui(t, py[0], a);
mpz_mod(py[0], t, n);
mpz_clear(t);
for (i = 0; i < r; i++)
if (mpz_sgn(py[i]))
retval = 0;
return retval;
}
#if AKS_VARIANT != AKS_VARIANT_V6
static int is_primitive_root_uiprime(mpz_t n, UV r)
{
int res;
mpz_t zr;
mpz_init_set_ui(zr, r);
res = is_primitive_root(n, zr, 1);
mpz_clear(zr);
return res;
}
#endif
#if AKS_VARIANT == AKS_VARIANT_BERN21
static UV largest_factor(UV n) {
UV p = 2;
PRIME_ITERATOR(iter);
while (n >= p*p && !prime_iterator_isprime(&iter, n)) {
while ( (n % p) == 0 && n >= p*p ) { n /= p; }
p = prime_iterator_next(&iter);
}
prime_iterator_destroy(&iter);
return n;
}
#endif
#if AKS_VARIANT == AKS_VARIANT_BERN41
int bern41_acceptable(mpz_t n, UV r, UV s, mpz_t t1, mpz_t t2)
{
double scmp = ceil(sqrt( (r-1)/3.0 )) * mpz_log2(n);
UV d = (UV) (0.5 * (r-1));
UV i = (UV) (0.475 * (r-1));
UV j = i;
/* Ensure conditions are correct */
if (d > r-2) d = r-2;
if (i > d) i = d;
if (j > (r-2-d)) j = r-2-d;
mpz_bin_uiui(t2, 2*s, i);
mpz_bin_uiui(t1, d, i); mpz_mul(t2, t2, t1);
mpz_bin_uiui(t1, 2*s-i, j); mpz_mul(t2, t2, t1);
mpz_bin_uiui(t1, r-2-d, j); mpz_mul(t2, t2, t1);
return (mpz_log2(t2) >= scmp);
}
#endif
int is_aks_prime(mpz_t n)
{
mpz_t *px, *py;
int retval;
UV i, s, r, a;
UV starta = 1;
int _verbose = get_verbose_level();
if (mpz_cmp_ui(n, 4) < 0)
return (mpz_cmp_ui(n, 1) <= 0) ? 0 : 1;
/* Just for performance: check small divisors: 2*3*5*7*11*13*17*19*23 */
if (mpz_gcd_ui(0, n, 223092870UL) != 1 && mpz_cmp_ui(n, 23) > 0)
return 0;
if (mpz_perfect_power_p(n))
return 0;
#if AKS_VARIANT == AKS_VARIANT_V6 /* From the V6 AKS paper */
{
mpz_t sqrtn, t;
double log2n;
UV limit, startr;
PRIME_ITERATOR(iter);
mpz_init(sqrtn);
mpz_sqrt(sqrtn, n);
log2n = mpz_log2(n);
limit = (UV) floor( log2n * log2n );
if (_verbose>1) gmp_printf("# AKS checking order_r(%Zd) to %"UVuf"\n", n, (unsigned long) limit);
/* Using a native r limits us to ~2000 digits in the worst case (r ~ log^5n)
* but would typically work for 100,000+ digits (r ~ log^3n). This code is
* far too slow to matter either way. Composite r is ok here, but it will
* always end up prime, so save time and just check primes. */
retval = 0;
/* Start order search at a good spot. Idea from Nemana and Venkaiah. */
startr = (mpz_sizeinbase(n,2)-1) * (mpz_sizeinbase(n,2)-1);
startr = (startr < 1002) ? 2 : startr - 100;
for (r = 2; /* */; r = prime_iterator_next(&iter)) {
if (mpz_divisible_ui_p(n, r) ) /* r divides n. composite. */
{ retval = 0; break; }
if (mpz_cmp_ui(sqrtn, r) <= 0) /* no r <= sqrtn divides n. prime. */
{ retval = 1; break; }
if (r < startr) continue;
if (mpz_order_ui(r, n, limit) > limit)
{ retval = 2; break; }
}
prime_iterator_destroy(&iter);
mpz_clear(sqrtn);
if (retval != 2) return retval;
/* Since r is prime, phi(r) = r-1. */
s = (UV) floor( sqrt(r-1) * log2n );
}
#elif AKS_VARIANT == AKS_VARIANT_BORNEMANN /* Bernstein + Voloch */
{
UV slim;
double c2, x;
/* small t = few iters of big poly. big t = many iters of small poly */
double const t = (mpz_sizeinbase(n, 2) <= 64) ? 32 : 40;
double const t1 = (1.0/((t+1)*log(t+1)-t*log(t)));
double const dlogn = mpz_logn(n);
mpz_t tmp;
PRIME_ITERATOR(iter);
mpz_init(tmp);
prime_iterator_setprime(&iter, (UV) (t1*t1 * dlogn*dlogn) );
r = prime_iterator_next(&iter);
while (!is_primitive_root_uiprime(n,r))
r = prime_iterator_next(&iter);
prime_iterator_destroy(&iter);
slim = (UV) (2*t*(r-1));
c2 = dlogn * floor(sqrt(r));
{ /* Binary search for first s in [1,slim] where x >= 0 */
UV bi = 1;
UV bj = slim;
while (bi < bj) {
s = bi + (bj-bi)/2;
mpz_bin_uiui(tmp, r+s-1, s);
x = mpz_logn(tmp) / c2 - 1.0;
if (x < 0) bi = s+1;
else bj = s;
}
s = bi-1;
}
s = (s+3) >> 1;
/* Bornemann checks factors up to (s-1)^2, we check to max(r,s) */
/* slim = (s-1)*(s-1); */
slim = (r > s) ? r : s;
if (_verbose > 1) printf("# aks trial to %"UVuf"\n", slim);
if (_GMP_trial_factor(n, 2, slim) > 1)
{ mpz_clear(tmp); return 0; }
mpz_sqrt(tmp, n);
if (mpz_cmp_ui(tmp, slim) <= 0)
{ mpz_clear(tmp); return 1; }
mpz_clear(tmp);
}
#elif AKS_VARIANT == AKS_VARIANT_BERN21
{ /* Bernstein 2003, theorem 2.1 (simplified) */
UV q;
double slim, scmp, x;
mpz_t t, t2;
PRIME_ITERATOR(iter);
mpz_init(t); mpz_init(t2);
r = s = 0;
while (1) {
/* todo: Check r|n and r >= sqrt(n) here instead of waiting */
if (mpz_cmp_ui(n, r) <= 0) break;
r = prime_iterator_next(&iter);
q = largest_factor(r-1);
mpz_set_ui(t, r);
mpz_powm_ui(t, n, (r-1)/q, t);
if (mpz_cmp_ui(t, 1) <= 0) continue;
scmp = 2 * floor(sqrt(r)) * mpz_log2(n);
slim = 20 * (r-1);
/* Check viability */
mpz_bin_uiui(t, q+slim-1, slim); if (mpz_log2(t) < scmp) continue;
for (s = 2; s < slim; s++) {
mpz_bin_uiui(t, q+s-1, s);
if (mpz_log2(t) > scmp) break;
}
if (s < slim) break;
}
mpz_clear(t); mpz_clear(t2);
prime_iterator_destroy(&iter);
if (_GMP_trial_factor(n, 2, s) > 1)
return 0;
}
#elif AKS_VARIANT == AKS_VARIANT_BERN22
{ /* Bernstein 2003, theorem 2.2 (simplified) */
UV q;
double slim, scmp, x;
mpz_t t, t2;
PRIME_ITERATOR(iter);
mpz_init(t); mpz_init(t2);
r = s = 0;
while (1) {
/* todo: Check r|n and r >= sqrt(n) here instead of waiting */
if (mpz_cmp_ui(n, r) <= 0) break;
r = prime_iterator_next(&iter);
if (!is_primitive_root_uiprime(n,r)) continue;
q = r-1; /* Since r is prime, phi(r) = r-1 */
scmp = 2 * floor(sqrt(r-1)) * mpz_log2(n);
slim = 20 * (r-1);
/* Check viability */
mpz_bin_uiui(t, q+slim-1, slim); if (mpz_log2(t) < scmp) continue;
for (s = 2; s < slim; s++) {
mpz_bin_uiui(t, q+s-1, s);
if (mpz_log2(t) > scmp) break;
}
if (s < slim) break;
}
mpz_clear(t); mpz_clear(t2);
prime_iterator_destroy(&iter);
if (_GMP_trial_factor(n, 2, s) > 1)
return 0;
}
#elif AKS_VARIANT == AKS_VARIANT_BERN23
{ /* Bernstein 2003, theorem 2.3 (simplified) */
UV q, d, limit;
double slim, scmp, sbin, x, log2n;
mpz_t t, t2;
PRIME_ITERATOR(iter);
mpz_init(t); mpz_init(t2);
log2n = mpz_log2(n);
limit = (UV) floor( log2n * log2n );
r = 2;
s = 0;
while (1) {
/* todo: Check r|n and r >= sqrt(n) here instead of waiting */
if (mpz_cmp_ui(n, r) <= 0) break;
r++;
UV gcd = mpz_gcd_ui(NULL, n, r);
if (gcd != 1) { mpz_clear(t); mpz_clear(t2); return 0; }
UV v = mpz_order_ui(r, n, limit);
if (v >= limit) continue;
mpz_set_ui(t2, r);
totient(t, t2);
q = mpz_get_ui(t);
UV phiv = q/v;
/* printf("phi(%lu)/v = %lu/%lu = %lu\n", r, q, v, phiv); */
/* This is extremely inefficient. */
/* Choose an s value we'd be happy with */
slim = 20 * (r-1);
/* Quick check to see if it could work with s=slim, d=1 */
mpz_bin_uiui(t, q+slim-1, slim);
sbin = mpz_log2(t);
if (sbin < 2*floor(sqrt(q))*log2n)
continue;
for (s = 2; s < slim; s++) {
mpz_bin_uiui(t, q+s-1, s);
sbin = mpz_log2(t);
if (sbin < 2*floor(sqrt(q))*log2n) continue; /* d=1 */
/* Check each number dividing phi(r)/v */
for (d = 2; d < phiv; d++) {
if ((phiv % d) != 0) continue;
scmp = 2 * d * floor(sqrt(q/d)) * log2n;
if (sbin < scmp) break;
}
/* if we did not exit early, this s worked for each d. This s wins. */
if (d >= phiv) break;
}
if (s < slim) break;
}
mpz_clear(t); mpz_clear(t2);
prime_iterator_destroy(&iter);
if (_GMP_trial_factor(n, 2, s) > 1)
return 0;
}
#elif AKS_VARIANT == AKS_VARIANT_BERN41
{
double const log2n = mpz_log2(n);
/* Tuning: Initial 'r' selection */
double const r0 = 0.008 * log2n * log2n;
/* Tuning: Try a larger 'r' if 's' looks very large */
UV const rmult = 8;
UV slim;
mpz_t tmp, tmp2;
PRIME_ITERATOR(iter);
mpz_init(tmp); mpz_init(tmp2);
/* r has to be at least 3. */
prime_iterator_setprime(&iter, (r0 < 2) ? 2 : (UV) r0);
r = prime_iterator_next(&iter);
/* r must be a primitive root. For performance, skip if s looks too big. */
while ( !is_primitive_root_uiprime(n, r) ||
!bern41_acceptable(n, r, rmult*(r-1), tmp, tmp2) )
r = prime_iterator_next(&iter);
prime_iterator_destroy(&iter);
{ /* Binary search for first s in [1,lim] where conditions met */
UV bi = 1;
UV bj = rmult * (r-1);
while (bi < bj) {
s = bi + (bj-bi)/2;
if (!bern41_acceptable(n,r,s,tmp,tmp2)) bi = s+1;
else bj = s;
}
s = bj;
/* Our S goes from 2 to s+1. */
starta = 2;
s = s+1;
}
/* printf("chose r=%lu s=%lu d = %lu i = %lu j = %lu\n", r, s, d, i, j); */
/* Check divisibility to s(s-1) to cover both gcd conditions */
slim = s * (s-1);
if (_verbose > 1) printf("# aks trial to %"UVuf"\n", slim);
if (_GMP_trial_factor(n, 2, slim) > 1)
{ mpz_clear(tmp); mpz_clear(tmp2); return 0; }
/* If we checked divisibility to sqrt(n), then it is prime. */
mpz_sqrt(tmp, n);
if (mpz_cmp_ui(tmp, slim) <= 0)
{ mpz_clear(tmp); mpz_clear(tmp2); return 1; }
/* Check b^(n-1) = 1 mod n for b in [2..s] */
mpz_sub_ui(tmp2, n, 1);
for (i = 2; i <= s; i++) {
mpz_set_ui(tmp, i);
mpz_powm(tmp, tmp, tmp2, n);
if (mpz_cmp_ui(tmp, 1) != 0)
{ mpz_clear(tmp); mpz_clear(tmp2); return 0; }
}
mpz_clear(tmp); mpz_clear(tmp2);
}
#endif
if (_verbose) gmp_printf("# AKS %Zd. r = %"UVuf" s = %"UVuf"\n", n, (unsigned long) r, (unsigned long) s);
/* Create the three polynomials we will use */
New(0, px, r, mpz_t);
New(0, py, r, mpz_t);
if ( !px || !py )
croak("allocation failure\n");
for (i = 0; i < r; i++) {
mpz_init(px[i]);
mpz_init(py[i]);
}
retval = 1;
for (a = starta; a <= s; a++) {
retval = test_anr(a, n, r, px, py);
if (!retval) break;
if (_verbose>1) { printf("."); fflush(stdout); }
}
if (_verbose>1) { printf("\n"); fflush(stdout); };
/* Free the polynomials */
for (i = 0; i < r; i++) {
mpz_clear(px[i]);
mpz_clear(py[i]);
}
Safefree(px);
Safefree(py);
return retval;
}
|